Best cloud GPU for Qwen2.5 72B Instruct LLM inference (FP16)
Compare whole-GPU quotes for Qwen2.5 72B Instruct. “Best” means the lowest instance bill among reviewed GPU models where one GPU passes the cited memory estimate and software floor in this snapshot; a tensor-parallel multi-GPU instance is chosen only when no single GPU fits, and cheaper tensor-parallel options are listed separately. Throughput and successful execution have not been benchmarked.
Only current whole-GPU on-demand and Spot USD quotes in standard or unspecified regions qualify. Fractional GPUs, GPU-component prices, promotional “from” prices, tax-included quotes and explicit out-of-stock records are excluded. Quotes with unknown tax basis remain labelled; tier B/C quotes are display-only.
Minimum VRAM screening rule
Labelled estimate, batch 1: 72.706203648B parameters × 2 bytes/parameter = 145.412 GB weights; required = 145.412 × 1.2 + 2.684 GB FP16 KV cache at 8192 tokens = 177.179 GB.
Exact input: 72706203648 parameters, derived from the official configuration (hidden size 8192, intermediate size 29568, vocabulary 152064, including embeddings, output projection, norms and any Q/K/V biases), cross-checked against the BF16 weight index total_size / 2. We use 8192 tokens, below the configured maximum of 32768; no context extension is assumed.
Hugging Face Accelerate says inference may need up to about 20% extra memory beyond loading weights. We use that guidance as the 1.2 allowance, then add the stated KV cache. This is our estimate, not a runtime requirement published by the model developer; kernels, context, batches and backend can change usage.
KV input: 2 × 80 layers × 8 KV heads × head dimension 128 × 2 bytes × 8192 tokens / 10⁹ = 2.684 GB, batch 1.
Tensor-parallel inputs: 64 attention heads and 8 KV heads; GPU count must divide both.
Model-developer cross-check, Qwen: Qwen: speed and GPU-memory benchmark. Transformers 4.46.0, PyTorch 2.3.1, CUDA 12.1, A100 80GB; batch 1, input 6144 + output 2048 tokens. The benchmark labels memory GB, but Qwen: benchmark memory calculation (bytes / 1024³) sums peak allocated bytes across GPUs and divides by 1024³: these are GiB. We convert to decimal GB for screening (1 GiB = 1.073741824 GB). The benchmark does not list vLLM memory because vLLM preallocates GPU memory.
BF16 measurement published by Qwen: 140.00 GiB total across 2 GPU(s). BF16 and FP16 both store weights in two bytes, but use different formats and can have different runtime usage.
Governing capacity screen: max(our estimate 177.179 GB, model-developer peak 150.324 GB after conversion) = 177.179 GB. A single GPU, or the total capacity of an eligible tensor-parallel instance, must meet this figure. For 4-bit we use the larger GPTQ/AWQ peak.
Qwen ran this benchmark on 2 × A100 80GB, a setup below our conservative capacity screen. Our estimate remains the conservative screen; the measurement is specific to that backend and format.
Reviewed software floor: vLLM on Linux, NVIDIA CUDA compute capability ≥ 7.5; or AMD MI200s (gfx90a), MI300 (gfx942), MI350 (gfx950), Radeon RX 7900 series (gfx1100/1101), Radeon RX 9000 series (gfx1200/1201), Ryzen AI MAX / AI 300 Series (gfx1151/1150). ROCm ≥ 6.3; MI350 ≥ 7.0; Ryzen ≥ 7.0.2. Architecture eligibility does not certify a particular quantization kernel or host wheel. On aarch64, stable PyTorch wheels target only 8.0/9.0/10.0/11.0/12.0; T4G (CC 7.5) is memory-only. AMD Arm hosts are outside this screen.
Capacity comes only from reviewed per-model manufacturer specifications linked in each row. Decimal GB estimates are compared conservatively with those published capacity labels, without credit for a possible GiB interpretation. Unreviewed models (listed below), over-reported capacities, mining SKUs and laptop GPUs are omitted. One-GPU fits and tensor-parallel instance fits are screened separately.
Price observations dated: 2026-11-18
Qualifying current listings: 429. Distinct model/type minima shown: 63.
“Best” on this page
On-demand: B200 SXM6 180GB · Verda / DataCrunch · 1 GPU per instance · $9.80 per instance-hour
Spot: B200 SXM6 180GB · Verda / DataCrunch · 1 GPU per instance · $4.90 per instance-hour
One GPU’s memory is enough: cheapest instance
On-demand: B200 SXM6 180GB · Verda / DataCrunch · 1 GPU per instance · $9.80 per instance-hour
Spot: B200 SXM6 180GB · Verda / DataCrunch · 1 GPU per instance · $4.90 per instance-hour
Cheapest multi-GPU instance
On-demand: RTX A5000 24GB · Clore.ai · 8 GPUs per instance · $1.13 per instance-hour
Spot: H100 NVL 94GB · Clore.ai · 2 GPUs per instance · $0.917 per instance-hour
One-GPU fit tables show each model/type’s cheapest per-GPU quote; equal rates prefer a smaller instance bill, then fewer GPUs. Tensor-parallel tables rank by instance-hour bill. Headlines use the cheapest qualifying instance bill. Regions, providers, allocations and tax bases may differ; these are independent minima, not matched Spot discounts. Quote cards retain billing and availability conditions.
Not yet reviewed — memory unverified (model names only)
A100PCIE40GB · B300 HGX · B300 SXM · CMP 170HX · CMP 30HX · CMP 40HX · CMP 50HX · CMP 70HX · CMP 90HX · GTX 1050 Ti · GTX 1650 · GTX 970 · GeForce GTX 1060 · GeForce GTX 1070 · GeForce GTX 1080 Ti · K40 · P100 · P104-100 · P106-100 · P4 · Quadro P2200 · Quadro-RTX5000 · RTX 2060 · RTX 3050 Laptop GPU · RTX 3050 Ti Laptop GPU · RTX 3060 Laptop GPU · RTX 3070 Laptop GPU · RTX 4090 Laptop GPU · RTX 5090 Laptop GPU · RTX A5500 Laptop GPU · RTX PRO 4000 Blackwell SFF Edition · RTX PRO 5000 72GB Blackwell · RTX PRO 6000 Blackwell Max-Q · Radeon Pro V710 · Tesla P4 · Tesla V100S
Fits on one GPU
| GPU model | Reviewed GB per GPU | Price type | Per GPU $/hr | GPUs per instance | Instance $/hr | Estimated headroom (single: per GPU; multi: total) | Reviewed architecture and checks |
|---|---|---|---|---|---|---|---|
| MI300X 192GB | 192 | Spot | $1.11 | 8 | $8.87 | tight < 1.25× (estimate) | AMD · CDNA 3 · gfx942 · non-CUDA · one GPU’s memory passes, but this offer is sold only as an instance with 8 GPUs |
| B200 SXM6 180GB | 180 | Spot | $4.90 | 4 | $19.59 | tight < 1.25× (estimate) | NVIDIA · Blackwell · CUDA CC 10.0 · one GPU’s memory passes, but this offer is sold only as an instance with 4 GPUs |
| B200 CC SXM6 180GB | 180 | Spot | $5.00 | 4 | $19.98 | tight < 1.25× (estimate) | NVIDIA · Blackwell · CUDA CC 10.0 · one GPU’s memory passes, but this offer is sold only as an instance with 4 GPUs |
| MI300X 192GB | 192 | On-demand | $6.00 | 8 | $48.00 | tight < 1.25× (estimate) | AMD · CDNA 3 · gfx942 · non-CUDA · one GPU’s memory passes, but this offer is sold only as an instance with 8 GPUs |
| B300 SXM6 268GB | 268 | Spot | $6.39 | 1 | $6.39 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 10.3 |
| B300 CC SXM6 268GB | 268 | Spot | $6.52 | 4 | $26.09 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 10.3 · one GPU’s memory passes, but this offer is sold only as an instance with 4 GPUs |
| B200 SXM6 180GB | 180 | On-demand | $9.79 | 2 | $19.59 | tight < 1.25× (estimate) | NVIDIA · Blackwell · CUDA CC 10.0 · one GPU’s memory passes, but this offer is sold only as an instance with 2 GPUs |
| B200 CC SXM6 180GB | 180 | On-demand | $9.99 | 2 | $19.98 | tight < 1.25× (estimate) | NVIDIA · Blackwell · CUDA CC 10.0 · one GPU’s memory passes, but this offer is sold only as an instance with 2 GPUs |
| B300 SXM6 268GB | 268 | On-demand | $12.79 | 8 | $102.30 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 10.3 · one GPU’s memory passes, but this offer is sold only as an instance with 8 GPUs |
| B300 CC SXM6 268GB | 268 | On-demand | $13.04 | 4 | $52.18 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 10.3 · one GPU’s memory passes, but this offer is sold only as an instance with 4 GPUs |
| B200 180GB | 180 | On-demand | $14.00 | 8 | $112.00 | tight < 1.25× (estimate) | NVIDIA · Blackwell · CUDA CC 10.0 · one GPU’s memory passes, but this offer is sold only as an instance with 8 GPUs |
| B200 179GB | 179 | On-demand | $14.24 | 8 | $113.93 | tight < 1.25× (estimate) | NVIDIA · Blackwell · CUDA CC 10.0 · one GPU’s memory passes, but this offer is sold only as an instance with 8 GPUs |
| B300 268GB | 268 | On-demand | $17.80 | 8 | $142.42 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 10.3 · one GPU’s memory passes, but this offer is sold only as an instance with 8 GPUs |
Multi-GPU instances (tensor parallel)
Use 2, 4 or 8 GPUs in one eligible instance. The count must divide both attention and KV heads, and count × reviewed per-GPU GB must meet the estimate. This is a capacity screen; tensor-parallel execution and interconnect must be checked.
| GPU model | Reviewed GB per GPU | Price type | Per GPU $/hr | GPUs per instance | Instance $/hr | Estimated headroom (single: per GPU; multi: total) | Reviewed architecture and checks |
|---|---|---|---|---|---|---|---|
| H100 NVL 94GB | 94 | Spot | $0.458 | 2 | $0.917 | tight < 1.25× (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 2 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX A5000 24GB | 24 | On-demand | $0.142 | 8 | $1.13 | tight < 1.25× (estimate) | NVIDIA · Ampere · CUDA CC 8.6 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX A5000 24GB | 24 | Spot | $0.142 | 8 | $1.13 | tight < 1.25× (estimate) | NVIDIA · Ampere · CUDA CC 8.6 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H100 NVL 94GB | 94 | On-demand | $0.833 | 2 | $1.67 | tight < 1.25× (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 2 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX A6000 48GB | 48 | Spot | $0.458 | 4 | $1.83 | tight < 1.25× (estimate) | NVIDIA · Ampere · CUDA CC 8.6 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX 3090 24GB | 24 | On-demand | $0.260 | 8 | $2.08 | tight < 1.25× (estimate) | NVIDIA · Ampere · CUDA CC 8.6 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX 3090 24GB | 24 | Spot | $0.260 | 8 | $2.08 | tight < 1.25× (estimate) | NVIDIA · Ampere · CUDA CC 8.6 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX A6000 48GB | 48 | On-demand | $0.550 | 4 | $2.20 | tight < 1.25× (estimate) | NVIDIA · Ampere · CUDA CC 8.6 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| A30 24GB | 24 | On-demand | $0.350 | 8 | $2.80 | tight < 1.25× (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX PRO 6000 96GB | 96 | Spot | $1.46 | 2 | $2.92 | tight < 1.25× (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 2 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX 6000 Ada 48GB | 48 | On-demand | $0.790 | 4 | $3.16 | tight < 1.25× (estimate) | NVIDIA · Ada Lovelace · CUDA CC 8.9 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX 6000 Ada 48GB | 48 | Spot | $0.827 | 4 | $3.31 | tight < 1.25× (estimate) | NVIDIA · Ada Lovelace · CUDA CC 8.9 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX PRO 6000 Blackwell Workstation 96GB | 96 | On-demand | $1.67 | 2 | $3.33 | tight < 1.25× (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 2 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX PRO 6000 Blackwell Workstation 96GB | 96 | Spot | $1.67 | 2 | $3.33 | tight < 1.25× (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 2 GPUs; needs a serving engine with TP support; check the interconnect |
| L40 48GB | 48 | On-demand | $0.860 | 4 | $3.44 | tight < 1.25× (estimate) | NVIDIA · Ada Lovelace · CUDA CC 8.9 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect · PCIe only, no NVLink; vLLM suggests pipeline parallelism |
| RTX A6000 48GB | 48 | Spot | $0.458 | 8 | $3.66 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.6 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| L40S 48GB | 48 | On-demand | $0.970 | 4 | $3.88 | tight < 1.25× (estimate) | NVIDIA · Ada Lovelace · CUDA CC 8.9 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect · PCIe only, no NVLink; vLLM suggests pipeline parallelism |
| L40S 48GB | 48 | Spot | $1.04 | 4 | $4.16 | tight < 1.25× (estimate) | NVIDIA · Ada Lovelace · CUDA CC 8.9 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect · PCIe only, no NVLink; vLLM suggests pipeline parallelism |
| RTX PRO 6000 Blackwell 96GB | 96 | On-demand | $2.19 | 2 | $4.38 | tight < 1.25× (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 2 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX A6000 48GB | 48 | On-demand | $0.550 | 8 | $4.40 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.6 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| A100 SXM4 80GB | 80 | Spot | $1.21 | 4 | $4.84 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX 5090 32GB | 32 | On-demand | $0.625 | 8 | $5.00 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX 5090 32GB | 32 | Spot | $0.625 | 8 | $5.00 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| A100 80GB | 80 | On-demand | $1.35 | 4 | $5.40 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX PRO 6000 96GB | 96 | On-demand | $2.92 | 2 | $5.84 | tight < 1.25× (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 2 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX PRO 6000 96GB | 96 | Spot | $1.46 | 4 | $5.84 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| A100 40GB | 40 | Spot | $0.738 | 8 | $5.90 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H100 NVL | 94 | On-demand | $3.11 | 2 | $6.22 | tight < 1.25× (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 2 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX 4090 24GB | 24 | On-demand | $0.781 | 8 | $6.25 | tight < 1.25× (estimate) | NVIDIA · Ada Lovelace · CUDA CC 8.9 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX 4090 24GB | 24 | Spot | $0.781 | 8 | $6.25 | tight < 1.25× (estimate) | NVIDIA · Ada Lovelace · CUDA CC 8.9 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX 6000 Ada 48GB | 48 | On-demand | $0.790 | 8 | $6.32 | comfortable (estimate) | NVIDIA · Ada Lovelace · CUDA CC 8.9 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX 6000 Ada 48GB | 48 | Spot | $0.827 | 8 | $6.62 | comfortable (estimate) | NVIDIA · Ada Lovelace · CUDA CC 8.9 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| A100 80GB | 80 | Spot | $0.859 | 8 | $6.87 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H200 SXM5 141GB | 141 | Spot | $3.45 | 2 | $6.90 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 2 GPUs; needs a serving engine with TP support; check the interconnect |
| L4 24GB | 24 | On-demand | $0.865 | 8 | $6.92 | tight < 1.25× (estimate) | NVIDIA · Ada Lovelace · CUDA CC 8.9 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect · PCIe only, no NVLink; vLLM suggests pipeline parallelism |
| A100 SXM4 40GB | 40 | Spot | $0.867 | 8 | $6.93 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H200 NVL 141GB | 141 | On-demand | $3.62 | 2 | $7.24 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 2 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX PRO 4500 Blackwell 32GB | 32 | On-demand | $0.920 | 8 | $7.36 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX PRO 6000 Blackwell MaxQ Works 96GB | 96 | On-demand | $1.88 | 4 | $7.50 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX PRO 6000 Blackwell MaxQ Works 96GB | 96 | Spot | $1.88 | 4 | $7.50 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| L40S 48GB | 48 | On-demand | $0.970 | 8 | $7.76 | comfortable (estimate) | NVIDIA · Ada Lovelace · CUDA CC 8.9 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect · PCIe only, no NVLink; vLLM suggests pipeline parallelism |
| RTX PRO 6000 Blackwell Server Edition 96GB | 96 | On-demand | $4.14 | 2 | $8.29 | tight < 1.25× (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 2 GPUs; needs a serving engine with TP support; check the interconnect |
| L40S 48GB | 48 | Spot | $1.04 | 8 | $8.32 | comfortable (estimate) | NVIDIA · Ada Lovelace · CUDA CC 8.9 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect · PCIe only, no NVLink; vLLM suggests pipeline parallelism |
| RTX PRO 6000 Blackwell 96GB | 96 | On-demand | $2.19 | 4 | $8.76 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| A100 SXM4 80GB | 80 | On-demand | $2.42 | 4 | $9.69 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| A100 SXM4 80GB | 80 | Spot | $1.21 | 8 | $9.69 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H100 SXM 80GB | 80 | Spot | $2.70 | 4 | $10.80 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| A100 80GB | 80 | On-demand | $1.35 | 8 | $10.80 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H100 80GB | 80 | On-demand | $2.73 | 4 | $10.92 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| A100 SXM4 80GB | 80 | On-demand | $1.38 | 8 | $11.04 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| DGX A100 80GB | 80 | On-demand | $1.38 | 8 | $11.04 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H100 NVL 94GB | 94 | On-demand | $2.92 | 4 | $11.68 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX PRO 6000 96GB | 96 | On-demand | $2.92 | 4 | $11.69 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX PRO 6000 96GB | 96 | Spot | $1.46 | 8 | $11.69 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H100 NVL | 94 | On-demand | $3.11 | 4 | $12.44 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| H200 SXM5 141GB | 141 | On-demand | $6.91 | 2 | $13.81 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 2 GPUs; needs a serving engine with TP support; check the interconnect |
| H200 SXM5 141GB | 141 | Spot | $3.45 | 4 | $13.81 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| A100 SXM4 40GB | 40 | On-demand | $1.73 | 8 | $13.86 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H100 SXM 80GB | 80 | On-demand | $3.51 | 4 | $14.03 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| H200 NVL 141GB | 141 | On-demand | $3.62 | 4 | $14.48 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| A10G 24GB | 24 | On-demand | $2.04 | 8 | $16.29 | tight < 1.25× (estimate) | NVIDIA · Ampere · CUDA CC 8.6 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect · PCIe only, no NVLink; vLLM suggests pipeline parallelism |
| RTX PRO 6000 Blackwell Server Edition 96GB | 96 | On-demand | $4.14 | 4 | $16.57 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX PRO 6000 Blackwell 96GB | 96 | On-demand | $2.19 | 8 | $17.52 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| A100 NVSwitch 80GB | 80 | On-demand | $4.53 | 4 | $18.11 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| H100 SXM 80GB | 80 | Spot | $2.27 | 8 | $18.17 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| A100 40GB | 40 | On-demand | $2.74 | 8 | $21.96 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H100 NVL 94GB | 94 | On-demand | $2.92 | 8 | $23.36 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX PRO 6000 96GB | 96 | On-demand | $2.92 | 8 | $23.38 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H100 NVL | 94 | On-demand | $3.11 | 8 | $24.88 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H100 SXM 80GB | 80 | On-demand | $3.14 | 8 | $25.12 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H200 SXM5 141GB | 141 | On-demand | $6.91 | 4 | $27.62 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 4 GPUs; needs a serving engine with TP support; check the interconnect |
| H200 SXM5 141GB | 141 | Spot | $3.45 | 8 | $27.62 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX PRO 4500 32GB | 32 | On-demand | $3.56 | 8 | $28.51 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H200 NVL 141GB | 141 | On-demand | $3.62 | 8 | $28.96 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| A100 NVSwitch 40GB | 40 | On-demand | $3.91 | 8 | $31.30 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| RTX PRO 6000 Blackwell Server Edition 96GB | 96 | On-demand | $4.14 | 8 | $33.14 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 12.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| A100 NVSwitch 80GB | 80 | On-demand | $4.53 | 8 | $36.22 | comfortable (estimate) | NVIDIA · Ampere · CUDA CC 8.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H200 141GB | 141 | On-demand | $5.77 | 8 | $46.15 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H100 80GB | 80 | On-demand | $6.88 | 8 | $55.04 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H200 SXM5 141GB | 141 | On-demand | $6.90 | 8 | $55.23 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
| H200 141GB | 141 | Spot | $10.60 | 8 | $84.80 | comfortable (estimate) | NVIDIA · Hopper · CUDA CC 9.0 · tensor parallel across 8 GPUs; needs a serving engine with TP support; check the interconnect |
Meets memory only, check software support
| GPU model | Reviewed GB per GPU | Price type | Per GPU $/hr | GPUs per instance | Instance $/hr | Estimated headroom (single: per GPU; multi: total) | Reviewed architecture and checks |
|---|---|---|---|---|---|---|---|
| GB300 SXM6 288GB | 288 | Spot | $7.35 | 1 | $7.35 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 10.3 · Arm (aarch64) host · below floor |
| GB300 SXM6 288GB | 288 | On-demand | $14.71 | 4 | $58.83 | comfortable (estimate) | NVIDIA · Blackwell · CUDA CC 10.3 · Arm (aarch64) host · below floor |
H100 NVL 94GB 2×
US
USD (stablecoin) • platform fees may apply
$0.458 per GPU-hour
Minimum billable instance cost: USD $0.917/hr · Minimum billing duration unknown
Details & source
SKU: 122309 · Availability: Available · A · Official API
Reported GPU name: NVIDIA H100 NVL
Reported VRAM: 94 GB
Hourly amount in source scope: USD $0.917/hr · Instance
Source billing unit: Instance-hour
72/72
512 GB
94 GB
Listed whole-server daily price divided by 24 and GPU count. Platform service fees may apply; tax treatment unknown. Stablecoin-denominated USD price (USD-Blockchain).
API record pageRTX A5000 24GB 8×
ES
USD (stablecoin) • platform fees may apply
$0.142 per GPU-hour
Minimum billable instance cost: USD $1.13/hr · Minimum billing duration unknown
Details & source
SKU: 124560 · Availability: Available · A · Official API
Reported GPU name: NVIDIA RTX A5000
Reported VRAM: 24 GB
Hourly amount in source scope: USD $1.13/hr · Instance
Source billing unit: Instance-hour
32/64
256 GB
24 GB
Listed whole-server daily price divided by 24 and GPU count. Platform service fees may apply; tax treatment unknown. Stablecoin-denominated USD price (USD-Blockchain).
API record pageRTX A5000 24GB 8×
ES
USD (stablecoin) • platform fees may apply
$0.142 per GPU-hour
Minimum billable instance cost: USD $1.13/hr · Minimum billing duration unknown
Details & source
SKU: 124560 · Availability: Available · A · Official API
Reported GPU name: NVIDIA RTX A5000
Reported VRAM: 24 GB
Hourly amount in source scope: USD $1.13/hr · Instance
Source billing unit: Instance-hour
32/64
256 GB
24 GB
Listed whole-server daily price divided by 24 and GPU count. Platform service fees may apply; tax treatment unknown. Stablecoin-denominated USD price (USD-Blockchain).
API record pageH100 NVL 94GB 2×
US
USD (stablecoin) • platform fees may apply
$0.833 per GPU-hour
Minimum billable instance cost: USD $1.67/hr · Minimum billing duration unknown
Details & source
SKU: 122309 · Availability: Available · A · Official API
Reported GPU name: NVIDIA H100 NVL
Reported VRAM: 94 GB
Hourly amount in source scope: USD $1.67/hr · Instance
Source billing unit: Instance-hour
72/72
512 GB
94 GB
Listed whole-server daily price divided by 24 and GPU count. Platform service fees may apply; tax treatment unknown. Stablecoin-denominated USD price (USD-Blockchain).
API record page$0.458 per GPU-hour
Minimum billable instance cost: USD $1.83/hr · Minimum billing duration unknown
Details & source
SKU: 4A6000.40V · Availability: Unknown · A · Official API
Reported GPU name: 4x RTX A6000 48GB
Reported VRAM: 192GB GPU RAM
Hourly amount in source scope: USD $1.83/hr · Instance
Source billing unit: Instance-hour
40 CPU
240GB RAM
192GB GPU RAM
RTX 3090 24GB 8×
HR
USD (stablecoin) • platform fees may apply
$0.260 per GPU-hour
Minimum billable instance cost: USD $2.08/hr · Minimum billing duration unknown
Details & source
SKU: 103366 · Availability: Available · A · Official API
Reported GPU name: NVIDIA GeForce RTX 3090
Reported VRAM: 24 GB
Hourly amount in source scope: USD $2.08/hr · Instance
Source billing unit: Instance-hour
4/4
16 GB
24 GB
Listed whole-server daily price divided by 24 and GPU count. Platform service fees may apply; tax treatment unknown. Stablecoin-denominated USD price (USD-Blockchain).
API record pageRTX 3090 24GB 8×
HR
USD (stablecoin) • platform fees may apply
$0.260 per GPU-hour
Minimum billable instance cost: USD $2.08/hr · Minimum billing duration unknown
Details & source
SKU: 103366 · Availability: Available · A · Official API
Reported GPU name: NVIDIA GeForce RTX 3090
Reported VRAM: 24 GB
Hourly amount in source scope: USD $2.08/hr · Instance
Source billing unit: Instance-hour
4/4
16 GB
24 GB
Listed whole-server daily price divided by 24 and GPU count. Platform service fees may apply; tax treatment unknown. Stablecoin-denominated USD price (USD-Blockchain).
API record pageRTX A6000 48GB 4×
vCPU: 30
Region not specified by provider
$0.550 per GPU-hour
Minimum billable instance cost: USD $2.20/hr · Minimum billing duration unknown
Details & source
SKU: RTX A6000 (48GB) [ALT Config] | 4 GPU | 1024 storage · Availability: Unknown · C · Display only
Reported GPU name: RTX A6000 (48GB) [ALT Config]
Reported VRAM: 48GB
Hourly amount in source scope: USD $2.20/hr · Instance
Source billing unit: Instance-hour
96 GB
48GB
A30 24GB 8×
vCPU: 94
Region not specified by provider
$0.350 per GPU-hour
Minimum billable instance cost: USD $2.80/hr · Minimum billing duration unknown
Details & source
SKU: A30 (24GB) | 8 GPU | 2048 storage · Availability: Unknown · C · Display only
Reported GPU name: A30 (24GB)
Reported VRAM: 24GB
Hourly amount in source scope: USD $2.80/hr · Instance
Source billing unit: Instance-hour
384 GB
24GB
RTX PRO 6000 96GB 2×
Region not specified by provider
$1.46 per GPU-hour
Minimum billable instance cost: USD $2.92/hr · Minimum billing duration unknown
Details & source
SKU: 2RTXPRO6000.60V · Availability: Unknown · A · Official API
Reported GPU name: 2x RTX PRO 6000 96GB
Reported VRAM: 192GB GPU RAM
Hourly amount in source scope: USD $2.92/hr · Instance
Source billing unit: Instance-hour
60 CPU
180GB RAM
192GB GPU RAM
RTX 6000 Ada 48GB 4×
vCPU: 52
Region not specified by provider
$0.790 per GPU-hour
Minimum billable instance cost: USD $3.16/hr · Minimum billing duration unknown
Details & source
SKU: RTX 6000 ADA (48GB) | 4 GPU | 1400 storage · Availability: Unknown · C · Display only
Reported GPU name: RTX 6000 ADA (48GB)
Reported VRAM: 48GB
Hourly amount in source scope: USD $3.16/hr · Instance
Source billing unit: Instance-hour
288 GB
48GB
RTX 6000 Ada 48GB 4×
Region not specified by provider
$0.827 per GPU-hour
Minimum billable instance cost: USD $3.31/hr · Minimum billing duration unknown
Details & source
SKU: 4RTX6000ADA.40V · Availability: Unknown · A · Official API
Reported GPU name: 4x RTX 6000 Ada 48GB
Reported VRAM: 192GB GPU RAM
Hourly amount in source scope: USD $3.31/hr · Instance
Source billing unit: Instance-hour
40 CPU
240GB RAM
192GB GPU RAM
RTX PRO 6000 Blackwell Workstation 96GB 2×
US
USD (stablecoin) • platform fees may apply
$1.67 per GPU-hour
Minimum billable instance cost: USD $3.33/hr · Minimum billing duration unknown
Details & source
SKU: 121985 · Availability: Available · A · Official API
Reported GPU name: NVIDIA RTX PRO 6000 Blackwell Workstation
Reported VRAM: 96 GB
Hourly amount in source scope: USD $3.33/hr · Instance
Source billing unit: Instance-hour
32/64
128 GB
96 GB
Listed whole-server daily price divided by 24 and GPU count. Platform service fees may apply; tax treatment unknown. Stablecoin-denominated USD price (USD-Blockchain).
API record pageRTX PRO 6000 Blackwell Workstation 96GB 2×
US
USD (stablecoin) • platform fees may apply
$1.67 per GPU-hour
Minimum billable instance cost: USD $3.33/hr · Minimum billing duration unknown
Details & source
SKU: 121985 · Availability: Available · A · Official API
Reported GPU name: NVIDIA RTX PRO 6000 Blackwell Workstation
Reported VRAM: 96 GB
Hourly amount in source scope: USD $3.33/hr · Instance
Source billing unit: Instance-hour
32/64
128 GB
96 GB
Listed whole-server daily price divided by 24 and GPU count. Platform service fees may apply; tax treatment unknown. Stablecoin-denominated USD price (USD-Blockchain).
API record pageL40 48GB 4×
vCPU: 50
Region not specified by provider
$0.860 per GPU-hour
Minimum billable instance cost: USD $3.44/hr · Minimum billing duration unknown
Details & source
SKU: L40 (48GB) | 4 GPU | 2500 storage · Availability: Unknown · C · Display only
Reported GPU name: L40 (48GB)
Reported VRAM: 48GB
Hourly amount in source scope: USD $3.44/hr · Instance
Source billing unit: Instance-hour
288 GB
48GB
$0.458 per GPU-hour
Minimum billable instance cost: USD $3.66/hr · Minimum billing duration unknown
Details & source
SKU: 8A6000.80V · Availability: Unknown · A · Official API
Reported GPU name: 8x RTX A6000 48GB
Reported VRAM: 384GB GPU RAM
Hourly amount in source scope: USD $3.66/hr · Instance
Source billing unit: Instance-hour
80 CPU
480GB RAM
384GB GPU RAM
L40S 48GB 4×
vCPU: 46
Region not specified by provider
$0.970 per GPU-hour
Minimum billable instance cost: USD $3.88/hr · Minimum billing duration unknown
Details & source
SKU: L40S (48GB) | 4 GPU | 2500 storage · Availability: Unknown · C · Display only
Reported GPU name: L40S (48GB)
Reported VRAM: 48GB
Hourly amount in source scope: USD $3.88/hr · Instance
Source billing unit: Instance-hour
288 GB
48GB
$1.04 per GPU-hour
Minimum billable instance cost: USD $4.16/hr · Minimum billing duration unknown
Details & source
SKU: 4L40S.80V · Availability: Unknown · A · Official API
Reported GPU name: 4x L40S 48GB
Reported VRAM: 192GB GPU RAM
Hourly amount in source scope: USD $4.16/hr · Instance
Source billing unit: Instance-hour
80 CPU
240GB RAM
192GB GPU RAM
RTX PRO 6000 Blackwell 96GB 2×
vCPU: 30
Region not specified by provider
$2.19 per GPU-hour
Minimum billable instance cost: USD $4.38/hr · Minimum billing duration unknown
Details & source
SKU: RTX PRO 6000 Blackwell (96GB) [Reduced Storage] | 2 GPU | 700 storage · Availability: Unknown · C · Display only
Reported GPU name: RTX PRO 6000 Blackwell (96GB) [Reduced Storage]
Reported VRAM: 96GB
Hourly amount in source scope: USD $4.38/hr · Instance
Source billing unit: Instance-hour
288 GB
96GB
RTX A6000 48GB 8×
vCPU: 62
Region not specified by provider
$0.550 per GPU-hour
Minimum billable instance cost: USD $4.40/hr · Minimum billing duration unknown
Details & source
SKU: RTX A6000 (48GB) [ALT Config] | 8 GPU | 2048 storage · Availability: Unknown · C · Display only
Reported GPU name: RTX A6000 (48GB) [ALT Config]
Reported VRAM: 48GB
Hourly amount in source scope: USD $4.40/hr · Instance
Source billing unit: Instance-hour
192 GB
48GB
$1.21 per GPU-hour
Minimum billable instance cost: USD $4.84/hr · Minimum billing duration unknown
Details & source
SKU: 4A100.88V · Availability: Unknown · A · Official API
Reported GPU name: 4x A100 SXM4 80GB
Reported VRAM: 320GB GPU RAM
Hourly amount in source scope: USD $4.84/hr · Instance
Source billing unit: Instance-hour
88 CPU
480GB RAM
320GB GPU RAM
$4.90 per GPU-hour
Minimum billable instance cost: USD $4.90/hr · Minimum billing duration unknown
Details & source
SKU: 1B200.30V · Availability: Unknown · A · Official API
Reported GPU name: 1x B200 SXM6 180GB
Reported VRAM: 180GB GPU RAM
Hourly amount in source scope: USD $4.90/hr · Instance
Source billing unit: Instance-hour
30 CPU
170GB RAM
180GB GPU RAM
RTX 5090 32GB 8×
BY
USD (stablecoin) • platform fees may apply
$0.625 per GPU-hour
Minimum billable instance cost: USD $5.00/hr · Minimum billing duration unknown
Details & source
SKU: 85081 · Availability: Available · A · Official API
Reported GPU name: NVIDIA GeForce RTX 5090
Reported VRAM: 32 GB
Hourly amount in source scope: USD $5.00/hr · Instance
Source billing unit: Instance-hour
48/96
256 GB
32 GB
Listed whole-server daily price divided by 24 and GPU count. Platform service fees may apply; tax treatment unknown. Stablecoin-denominated USD price (USD-Blockchain).
API record pageRTX 5090 32GB 8×
BY
USD (stablecoin) • platform fees may apply
$0.625 per GPU-hour
Minimum billable instance cost: USD $5.00/hr · Minimum billing duration unknown
Details & source
SKU: 85081 · Availability: Available · A · Official API
Reported GPU name: NVIDIA GeForce RTX 5090
Reported VRAM: 32 GB
Hourly amount in source scope: USD $5.00/hr · Instance
Source billing unit: Instance-hour
48/96
256 GB
32 GB
Listed whole-server daily price divided by 24 and GPU count. Platform service fees may apply; tax treatment unknown. Stablecoin-denominated USD price (USD-Blockchain).
API record pageA100 80GB 4×
vCPU: 60
Region not specified by provider
$1.35 per GPU-hour
Minimum billable instance cost: USD $5.40/hr · Minimum billing duration unknown
Details & source
SKU: A100 (80GB) | 4 GPU | 2048 storage · Availability: Unknown · C · Display only
Reported GPU name: A100 (80GB)
Reported VRAM: 80GB
Hourly amount in source scope: USD $5.40/hr · Instance
Source billing unit: Instance-hour
300 GB
80GB
RTX PRO 6000 96GB 2×
Region not specified by provider
$2.92 per GPU-hour
Minimum billable instance cost: USD $5.84/hr · Minimum billing duration unknown
Details & source
SKU: 2RTXPRO6000.60V · Availability: Unknown · A · Official API
Reported GPU name: 2x RTX PRO 6000 96GB
Reported VRAM: 192GB GPU RAM
Hourly amount in source scope: USD $5.84/hr · Instance
Source billing unit: Instance-hour
60 CPU
180GB RAM
192GB GPU RAM
RTX PRO 6000 96GB 4×
Region not specified by provider
$1.46 per GPU-hour
Minimum billable instance cost: USD $5.84/hr · Minimum billing duration unknown
Details & source
SKU: 4RTXPRO6000.120V · Availability: Unknown · A · Official API
Reported GPU name: 4x RTX PRO 6000 96GB
Reported VRAM: 384GB GPU RAM
Hourly amount in source scope: USD $5.84/hr · Instance
Source billing unit: Instance-hour
120 CPU
360GB RAM
384GB GPU RAM
$0.738 per GPU-hour
Minimum billable instance cost: USD $5.90/hr · Minimum billing duration unknown
Details & source
SKU: Standard_ND96asr_v4 · Availability: Unknown · A · Official API
Reported GPU name: A100 40GB
Hourly amount in source scope: USD $5.90/hr · Instance
Source billing unit: Instance-hour
H100 NVL 2×
vCPU: 36
Region not specified by provider
$3.11 per GPU-hour
Minimum billable instance cost: USD $6.22/hr · Minimum billing duration unknown
Details & source
SKU: H100 NVL | 2 GPU | 1500 storage · Availability: Unknown · C · Display only
Reported GPU name: H100 NVL
Hourly amount in source scope: USD $6.22/hr · Instance
Source billing unit: Instance-hour
256 GB
RTX 4090 24GB 8×
HR
USD (stablecoin) • platform fees may apply
$0.781 per GPU-hour
Minimum billable instance cost: USD $6.25/hr · Minimum billing duration unknown
Details & source
SKU: 87636 · Availability: Available · A · Official API
Reported GPU name: NVIDIA GeForce RTX 4090
Reported VRAM: 24 GB
Hourly amount in source scope: USD $6.25/hr · Instance
Source billing unit: Instance-hour
4/4
16 GB
24 GB
Listed whole-server daily price divided by 24 and GPU count. Platform service fees may apply; tax treatment unknown. Stablecoin-denominated USD price (USD-Blockchain).
API record pageRTX 4090 24GB 8×
HR
USD (stablecoin) • platform fees may apply
$0.781 per GPU-hour
Minimum billable instance cost: USD $6.25/hr · Minimum billing duration unknown
Details & source
SKU: 87636 · Availability: Available · A · Official API
Reported GPU name: NVIDIA GeForce RTX 4090
Reported VRAM: 24 GB
Hourly amount in source scope: USD $6.25/hr · Instance
Source billing unit: Instance-hour
4/4
16 GB
24 GB
Listed whole-server daily price divided by 24 and GPU count. Platform service fees may apply; tax treatment unknown. Stablecoin-denominated USD price (USD-Blockchain).
API record pageRTX 6000 Ada 48GB 8×
vCPU: 104
Region not specified by provider
$0.790 per GPU-hour
Minimum billable instance cost: USD $6.32/hr · Minimum billing duration unknown
Details & source
SKU: RTX 6000 ADA (48GB) | 8 GPU | 2800 storage · Availability: Unknown · C · Display only
Reported GPU name: RTX 6000 ADA (48GB)
Reported VRAM: 48GB
Hourly amount in source scope: USD $6.32/hr · Instance
Source billing unit: Instance-hour
640 GB
48GB
$6.39 per GPU-hour
Minimum billable instance cost: USD $6.39/hr · Minimum billing duration unknown
Details & source
SKU: 1B300.30V · Availability: Unknown · A · Official API
Reported GPU name: 1x B300 SXM6 268GB
Reported VRAM: 268GB GPU RAM
Hourly amount in source scope: USD $6.39/hr · Instance
Source billing unit: Instance-hour
30 CPU
255GB RAM
268GB GPU RAM
RTX 6000 Ada 48GB 8×
Region not specified by provider
$0.827 per GPU-hour
Minimum billable instance cost: USD $6.62/hr · Minimum billing duration unknown
Details & source
SKU: 8RTX6000ADA.80V · Availability: Unknown · A · Official API
Reported GPU name: 8x RTX 6000 Ada 48GB
Reported VRAM: 384GB GPU RAM
Hourly amount in source scope: USD $6.62/hr · Instance
Source billing unit: Instance-hour
80 CPU
480GB RAM
384GB GPU RAM
$0.859 per GPU-hour
Minimum billable instance cost: USD $6.87/hr · Minimum billing duration unknown
Details & source
SKU: Standard_ND96amsr_A100_v4 · Availability: Unknown · A · Official API
Reported GPU name: A100 80GB
Hourly amount in source scope: USD $6.87/hr · Instance
Source billing unit: Instance-hour
$3.45 per GPU-hour
Minimum billable instance cost: USD $6.90/hr · Minimum billing duration unknown
Details & source
SKU: 2H200.141S.88V · Availability: Unknown · A · Official API
Reported GPU name: 2x H200 SXM5 141GB
Reported VRAM: 282GB GPU RAM
Hourly amount in source scope: USD $6.90/hr · Instance
Source billing unit: Instance-hour
88 CPU
340GB RAM
282GB GPU RAM
$0.865 per GPU-hour
Original hourly price: EUR 6.3/hr · Instance
Minimum billable instance cost: USD $6.92/hr · Minimum billing duration unknown
Details & source
SKU: /instance/server/l4_8_24g/fr-par-1 · Availability: Unknown · A · Official API
Reported GPU name: L4
Hourly amount in source scope: USD $6.92/hr · Instance
Source billing unit: Instance-minute
Original hourly price: EUR 6.3 · USD conversion rate: 1.0989 · ECB 2026-11-17
AMD EPYC 7413 (2.65 GHz), x64, vCPUs: 64
384 GiB
$0.867 per GPU-hour
Minimum billable instance cost: USD $6.93/hr · Minimum billing duration unknown
Details & source
SKU: 8A100.40S.176V · Availability: Unknown · A · Official API
Reported GPU name: 8x A100 SXM4 40GB
Reported VRAM: 320GB GPU RAM
Hourly amount in source scope: USD $6.93/hr · Instance
Source billing unit: Instance-hour
176 CPU
960GB RAM
320GB GPU RAM
H200 NVL 141GB 2×
vCPU: 30
Region not specified by provider
$3.62 per GPU-hour
Minimum billable instance cost: USD $7.24/hr · Minimum billing duration unknown
Details & source
SKU: H200 NVL (141GB) | 2 GPU | 1500 storage · Availability: Unknown · C · Display only
Reported GPU name: H200 NVL (141GB)
Reported VRAM: 141GB
Hourly amount in source scope: USD $7.24/hr · Instance
Source billing unit: Instance-hour
360 GB
141GB
GB300 SXM6 288GB 1×
Region not specified by provider
$7.35 per GPU-hour
Minimum billable instance cost: USD $7.35/hr · Minimum billing duration unknown
Details & source
SKU: 1GB300.32V · Availability: Unknown · A · Official API
Reported GPU name: 1x GB300 SXM6 288GB
Reported VRAM: 288GB GPU RAM
Hourly amount in source scope: USD $7.35/hr · Instance
Source billing unit: Instance-hour
32 CPU
225GB RAM
288GB GPU RAM
RTX PRO 4500 Blackwell 32GB 8×
vCPU: 126
Region not specified by provider
$0.920 per GPU-hour
Minimum billable instance cost: USD $7.36/hr · Minimum billing duration unknown
Details & source
SKU: RTX PRO 4500 Blackwell (32GB) | 8 GPU | 6000 storage · Availability: Unknown · C · Display only
Reported GPU name: RTX PRO 4500 Blackwell (32GB)
Reported VRAM: 32GB
Hourly amount in source scope: USD $7.36/hr · Instance
Source billing unit: Instance-hour
592 GB
32GB
RTX PRO 6000 Blackwell MaxQ Works 96GB 4×
TW
USD (stablecoin) • platform fees may apply
$1.88 per GPU-hour
Minimum billable instance cost: USD $7.50/hr · Minimum billing duration unknown
Details & source
SKU: 88789 · Availability: Available · A · Official API
Reported GPU name: NVIDIA RTX PRO 6000 Blackwell MaxQ Works
Reported VRAM: 96 GB
Hourly amount in source scope: USD $7.50/hr · Instance
Source billing unit: Instance-hour
64/128
512 GB
96 GB
Listed whole-server daily price divided by 24 and GPU count. Platform service fees may apply; tax treatment unknown. Stablecoin-denominated USD price (USD-Blockchain).
API record pageRTX PRO 6000 Blackwell MaxQ Works 96GB 4×
TW
USD (stablecoin) • platform fees may apply
$1.88 per GPU-hour
Minimum billable instance cost: USD $7.50/hr · Minimum billing duration unknown
Details & source
SKU: 88789 · Availability: Available · A · Official API
Reported GPU name: NVIDIA RTX PRO 6000 Blackwell MaxQ Works
Reported VRAM: 96 GB
Hourly amount in source scope: USD $7.50/hr · Instance
Source billing unit: Instance-hour
64/128
512 GB
96 GB
Listed whole-server daily price divided by 24 and GPU count. Platform service fees may apply; tax treatment unknown. Stablecoin-denominated USD price (USD-Blockchain).
API record pageL40S 48GB 8×
vCPU: 94
Region not specified by provider
$0.970 per GPU-hour
Minimum billable instance cost: USD $7.76/hr · Minimum billing duration unknown
Details & source
SKU: L40S (48GB) | 8 GPU | 5000 storage · Availability: Unknown · C · Display only
Reported GPU name: L40S (48GB)
Reported VRAM: 48GB
Hourly amount in source scope: USD $7.76/hr · Instance
Source billing unit: Instance-hour
576 GB
48GB
RTX PRO 6000 Blackwell Server Edition 96GB 2×
vCPU: 48 · RAM (GiB): 512
us-east-1
$4.14 per GPU-hour
Minimum billable instance cost: USD $8.29/hr · Initial billing minimum: USD $0.138 / 60 seconds
Details & source
SKU: g7e.12xlarge · Availability: Unknown · A · Official API
Reported GPU name: RTX PRO 6000 Blackwell Server Edition
Reported VRAM: 192 GB
Hourly amount in source scope: USD $8.29/hr · Instance
Source billing unit: Instance-hour
Minimum billing (seconds): 60
192 GB
Linux, shared tenancy, no preinstalled software; storage and data transfer extra.
API record page$1.04 per GPU-hour
Minimum billable instance cost: USD $8.32/hr · Minimum billing duration unknown
Details & source
SKU: 8L40S.160V · Availability: Unknown · A · Official API
Reported GPU name: 8x L40S 48GB
Reported VRAM: 384GB GPU RAM
Hourly amount in source scope: USD $8.32/hr · Instance
Source billing unit: Instance-hour
160 CPU
480GB RAM
384GB GPU RAM
RTX PRO 6000 Blackwell 96GB 4×
vCPU: 60
Region not specified by provider
$2.19 per GPU-hour
Minimum billable instance cost: USD $8.76/hr · Minimum billing duration unknown
Details & source
SKU: RTX PRO 6000 Blackwell (96GB) [Reduced Storage] | 4 GPU | 1400 storage · Availability: Unknown · C · Display only
Reported GPU name: RTX PRO 6000 Blackwell (96GB) [Reduced Storage]
Reported VRAM: 96GB
Hourly amount in source scope: USD $8.76/hr · Instance
Source billing unit: Instance-hour
576 GB
96GB
$1.11 per GPU-hour
Minimum billable instance cost: USD $8.87/hr · Minimum billing duration unknown
Details & source
SKU: Standard_ND96isr_MI300X_v5 · Availability: Unknown · A · Official API
Reported GPU name: MI300X 192GB
Hourly amount in source scope: USD $8.87/hr · Instance
Source billing unit: Instance-hour
A100 SXM4 80GB 4×
Region not specified by provider
$2.42 per GPU-hour
Minimum billable instance cost: USD $9.69/hr · Minimum billing duration unknown
Details & source
SKU: 4A100.88V · Availability: Unknown · A · Official API
Reported GPU name: 4x A100 SXM4 80GB
Reported VRAM: 320GB GPU RAM
Hourly amount in source scope: USD $9.69/hr · Instance
Source billing unit: Instance-hour
88 CPU
480GB RAM
320GB GPU RAM
$1.21 per GPU-hour
Minimum billable instance cost: USD $9.69/hr · Minimum billing duration unknown
Details & source
SKU: 8A100.176V · Availability: Unknown · A · Official API
Reported GPU name: 8x A100 SXM4 80GB
Reported VRAM: 640GB GPU RAM
Hourly amount in source scope: USD $9.69/hr · Instance
Source billing unit: Instance-hour
176 CPU
960GB RAM
640GB GPU RAM
B200 SXM6 180GB 1×
Region not specified by provider
$9.80 per GPU-hour
Minimum billable instance cost: USD $9.80/hr · Minimum billing duration unknown
Details & source
SKU: 1B200.30V · Availability: Unknown · A · Official API
Reported GPU name: 1x B200 SXM6 180GB
Reported VRAM: 180GB GPU RAM
Hourly amount in source scope: USD $9.80/hr · Instance
Source billing unit: Instance-hour
30 CPU
170GB RAM
180GB GPU RAM
$2.70 per GPU-hour
Minimum billable instance cost: USD $10.80/hr · Minimum billing duration unknown
Details & source
SKU: 4H100.80S.176V · Availability: Unknown · A · Official API
Reported GPU name: 4x H100 SXM5 80GB
Reported VRAM: 320GB GPU RAM
Hourly amount in source scope: USD $10.80/hr · Instance
Source billing unit: Instance-hour
176 CPU
680GB RAM
320GB GPU RAM
A100 80GB 8×
vCPU: 120
Region not specified by provider
$1.35 per GPU-hour
Minimum billable instance cost: USD $10.80/hr · Minimum billing duration unknown
Details & source
SKU: A100 (80GB) | 8 GPU | 4096 storage · Availability: Unknown · C · Display only
Reported GPU name: A100 (80GB)
Reported VRAM: 80GB
Hourly amount in source scope: USD $10.80/hr · Instance
Source billing unit: Instance-hour
768 GB
80GB
H100 80GB 4×
vCPU: 64
Region not specified by provider
$2.73 per GPU-hour
Minimum billable instance cost: USD $10.92/hr · Minimum billing duration unknown
Details & source
SKU: H100 (80GB) | 4 GPU | 5000 storage · Availability: Unknown · C · Display only
Reported GPU name: H100 (80GB)
Reported VRAM: 80GB
Hourly amount in source scope: USD $10.92/hr · Instance
Source billing unit: Instance-hour
512 GB
80GB
A100 SXM4 80GB 8×
vCPU: 126
Region not specified by provider
$1.38 per GPU-hour
Minimum billable instance cost: USD $11.04/hr · Minimum billing duration unknown
Details & source
SKU: A100 SXM4 (80GB) | 8 GPU | 5000 storage · Availability: Unknown · C · Display only
Reported GPU name: A100 SXM4 (80GB)
Reported VRAM: 80GB
Hourly amount in source scope: USD $11.04/hr · Instance
Source billing unit: Instance-hour
800 GB
80GB
DGX A100 80GB 8×
vCPU: 120
Region not specified by provider
$1.38 per GPU-hour
Minimum billable instance cost: USD $11.04/hr · Minimum billing duration unknown
Details & source
SKU: DGX A100 (80GB) | 8 GPU | 6000 storage · Availability: Unknown · C · Display only
Reported GPU name: DGX A100 (80GB)
Reported VRAM: 80GB
Hourly amount in source scope: USD $11.04/hr · Instance
Source billing unit: Instance-hour
1800 GB
80GB
H100 NVL 94GB 4×
vCPU: 80
Region not specified by provider
$2.92 per GPU-hour
Minimum billable instance cost: USD $11.68/hr · Minimum billing duration unknown
Details & source
SKU: H100 NVL (94GB) NVLink | 4 GPU | 5000 storage · Availability: Unknown · C · Display only
Reported GPU name: H100 NVL (94GB) NVLink
Reported VRAM: 94GB
Hourly amount in source scope: USD $11.68/hr · Instance
Source billing unit: Instance-hour
512 GB
94GB
RTX PRO 6000 96GB 4×
Region not specified by provider
$2.92 per GPU-hour
Minimum billable instance cost: USD $11.69/hr · Minimum billing duration unknown
Details & source
SKU: 4RTXPRO6000.120V · Availability: Unknown · A · Official API
Reported GPU name: 4x RTX PRO 6000 96GB
Reported VRAM: 384GB GPU RAM
Hourly amount in source scope: USD $11.69/hr · Instance
Source billing unit: Instance-hour
120 CPU
360GB RAM
384GB GPU RAM
RTX PRO 6000 96GB 8×
Region not specified by provider
$1.46 per GPU-hour
Minimum billable instance cost: USD $11.69/hr · Minimum billing duration unknown
Details & source
SKU: 8RTXPRO6000.240V · Availability: Unknown · A · Official API
Reported GPU name: 8x RTX PRO 6000 96GB
Reported VRAM: 768GB GPU RAM
Hourly amount in source scope: USD $11.69/hr · Instance
Source billing unit: Instance-hour
240 CPU
720GB RAM
768GB GPU RAM
H100 NVL 4×
vCPU: 72
Region not specified by provider
$3.11 per GPU-hour
Minimum billable instance cost: USD $12.44/hr · Minimum billing duration unknown
Details & source
SKU: H100 NVL | 4 GPU | 3000 storage · Availability: Unknown · C · Display only
Reported GPU name: H100 NVL
Hourly amount in source scope: USD $12.44/hr · Instance
Source billing unit: Instance-hour
512 GB
H200 SXM5 141GB 2×
Region not specified by provider
$6.91 per GPU-hour
Minimum billable instance cost: USD $13.81/hr · Minimum billing duration unknown
Details & source
SKU: 2H200.141S.88V · Availability: Unknown · A · Official API
Reported GPU name: 2x H200 SXM5 141GB
Reported VRAM: 282GB GPU RAM
Hourly amount in source scope: USD $13.81/hr · Instance
Source billing unit: Instance-hour
88 CPU
340GB RAM
282GB GPU RAM
$3.45 per GPU-hour
Minimum billable instance cost: USD $13.81/hr · Minimum billing duration unknown
Details & source
SKU: 4H200.141S.176V · Availability: Unknown · A · Official API
Reported GPU name: 4x H200 SXM5 141GB
Reported VRAM: 564GB GPU RAM
Hourly amount in source scope: USD $13.81/hr · Instance
Source billing unit: Instance-hour
176 CPU
680GB RAM
564GB GPU RAM
A100 SXM4 40GB 8×
Region not specified by provider
$1.73 per GPU-hour
Minimum billable instance cost: USD $13.86/hr · Minimum billing duration unknown
Details & source
SKU: 8A100.40S.176V · Availability: Unknown · A · Official API
Reported GPU name: 8x A100 SXM4 40GB
Reported VRAM: 320GB GPU RAM
Hourly amount in source scope: USD $13.86/hr · Instance
Source billing unit: Instance-hour
176 CPU
960GB RAM
320GB GPU RAM
$3.51 per GPU-hour
Original hourly price: EUR 12.771/hr · Instance
Minimum billable instance cost: USD $14.03/hr · Minimum billing duration unknown
Details & source
SKU: /instance/server/h100_sxm_4_80g/fr-par-2 · Availability: Unknown · A · Official API
Reported GPU name: H100-SXM
Hourly amount in source scope: USD $14.03/hr · Instance
Source billing unit: Instance-minute
Original hourly price: EUR 12.771 · USD conversion rate: 1.0989 · ECB 2026-11-17
Intel Xeon Platinum 8452Y (2.0 GHz), x64, vCPUs: 64
480 GiB
H200 NVL 141GB 4×
vCPU: 62
Region not specified by provider
$3.62 per GPU-hour
Minimum billable instance cost: USD $14.48/hr · Minimum billing duration unknown
Details & source
SKU: H200 NVL (141GB) | 4 GPU | 3000 storage · Availability: Unknown · C · Display only
Reported GPU name: H200 NVL (141GB)
Reported VRAM: 141GB
Hourly amount in source scope: USD $14.48/hr · Instance
Source billing unit: Instance-hour
720 GB
141GB
$2.04 per GPU-hour
Minimum billable instance cost: USD $16.29/hr · Initial billing minimum: USD $0.271 / 60 seconds
Details & source
SKU: g5.48xlarge · Availability: Unknown · A · Official API
Reported GPU name: A10G
Reported VRAM: 192 GB
Hourly amount in source scope: USD $16.29/hr · Instance
Source billing unit: Instance-hour
Minimum billing (seconds): 60
192 GB
Linux, shared tenancy, no preinstalled software; storage and data transfer extra.
API record pageRTX PRO 6000 Blackwell Server Edition 96GB 4×
vCPU: 96 · RAM (GiB): 1024
us-east-1
$4.14 per GPU-hour
Minimum billable instance cost: USD $16.57/hr · Initial billing minimum: USD $0.276 / 60 seconds
Details & source
SKU: g7e.24xlarge · Availability: Unknown · A · Official API
Reported GPU name: RTX PRO 6000 Blackwell Server Edition
Reported VRAM: 384 GB
Hourly amount in source scope: USD $16.57/hr · Instance
Source billing unit: Instance-hour
Minimum billing (seconds): 60
384 GB
Linux, shared tenancy, no preinstalled software; storage and data transfer extra.
API record pageRTX PRO 6000 Blackwell 96GB 8×
vCPU: 120
Region not specified by provider
$2.19 per GPU-hour
Minimum billable instance cost: USD $17.52/hr · Minimum billing duration unknown
Details & source
SKU: RTX PRO 6000 Blackwell (96GB) [Reduced Storage] | 8 GPU | 2800 storage · Availability: Unknown · C · Display only
Reported GPU name: RTX PRO 6000 Blackwell (96GB) [Reduced Storage]
Reported VRAM: 96GB
Hourly amount in source scope: USD $17.52/hr · Instance
Source billing unit: Instance-hour
1152 GB
96GB
$4.53 per GPU-hour
Original hourly price: CNY 120.72/hr · Instance
Minimum billable instance cost: USD $18.11/hr · Minimum billing duration unknown
Details & source
SKU: bcc.gn5.c56m480.4A100-80g · Availability: Unknown · C · Display only
Reported GPU name: 4*Nvidia A100 Nvswitch-80g
Reported VRAM: 320GB
Hourly amount in source scope: USD $18.11/hr · Instance
Source billing unit: Instance-hour
Original hourly price: CNY 120.72 · USD conversion rate: 0.150012 · ECB 2026-11-17
56核
480GB
320GB
$2.27 per GPU-hour
Minimum billable instance cost: USD $18.17/hr · Minimum billing duration unknown
Details & source
SKU: Standard_ND96isr_H100_v5 · Availability: Unknown · A · Official API
Reported GPU name: H100 SXM 80GB
Hourly amount in source scope: USD $18.17/hr · Instance
Source billing unit: Instance-hour
B200 SXM6 180GB 2×
Region not specified by provider
$9.79 per GPU-hour
Minimum billable instance cost: USD $19.59/hr · Minimum billing duration unknown
Details & source
SKU: 2B200.60V · Availability: Unknown · A · Official API
Reported GPU name: 2x B200 SXM6 180GB
Reported VRAM: 360GB GPU RAM
Hourly amount in source scope: USD $19.59/hr · Instance
Source billing unit: Instance-hour
60 CPU
340GB RAM
360GB GPU RAM
$4.90 per GPU-hour
Minimum billable instance cost: USD $19.59/hr · Minimum billing duration unknown
Details & source
SKU: 4B200.120V · Availability: Unknown · A · Official API
Reported GPU name: 4x B200 SXM6 180GB
Reported VRAM: 720GB GPU RAM
Hourly amount in source scope: USD $19.59/hr · Instance
Source billing unit: Instance-hour
120 CPU
680GB RAM
720GB GPU RAM
B200 CC SXM6 180GB 2×
Region not specified by provider
$9.99 per GPU-hour
Minimum billable instance cost: USD $19.98/hr · Minimum billing duration unknown
Details & source
SKU: 2B200.60V.CC.AMD · Availability: Unknown · A · Official API
Reported GPU name: 2x B200 CC SXM6 180GB
Reported VRAM: 360GB GPU RAM
Hourly amount in source scope: USD $19.98/hr · Instance
Source billing unit: Instance-hour
60 CPU (AMD EPYC)
340GB RAM
360GB GPU RAM
B200 CC SXM6 180GB 4×
Region not specified by provider
$5.00 per GPU-hour
Minimum billable instance cost: USD $19.98/hr · Minimum billing duration unknown
Details & source
SKU: 4B200.120V.CC.AMD · Availability: Unknown · A · Official API
Reported GPU name: 4x B200 CC SXM6 180GB
Reported VRAM: 720GB GPU RAM
Hourly amount in source scope: USD $19.98/hr · Instance
Source billing unit: Instance-hour
120 CPU (AMD EPYC)
680GB RAM
720GB GPU RAM
$2.74 per GPU-hour
Minimum billable instance cost: USD $21.96/hr · Initial billing minimum: USD $0.366 / 60 seconds
Details & source
SKU: p4d.24xlarge · Availability: Unknown · A · Official API
Reported GPU name: A100
Reported VRAM: 320 GB HBM2
Hourly amount in source scope: USD $21.96/hr · Instance
Source billing unit: Instance-hour
Minimum billing (seconds): 60
320 GB HBM2
Linux, shared tenancy, no preinstalled software; storage and data transfer extra.
API record pageH100 NVL 94GB 8×
vCPU: 140
Region not specified by provider
$2.92 per GPU-hour
Minimum billable instance cost: USD $23.36/hr · Minimum billing duration unknown
Details & source
SKU: H100 NVL (94GB) NVLink | 8 GPU | 10000 storage · Availability: Unknown · C · Display only
Reported GPU name: H100 NVL (94GB) NVLink
Reported VRAM: 94GB
Hourly amount in source scope: USD $23.36/hr · Instance
Source billing unit: Instance-hour
1500 GB
94GB
RTX PRO 6000 96GB 8×
Region not specified by provider
$2.92 per GPU-hour
Minimum billable instance cost: USD $23.38/hr · Minimum billing duration unknown
Details & source
SKU: 8RTXPRO6000.240V · Availability: Unknown · A · Official API
Reported GPU name: 8x RTX PRO 6000 96GB
Reported VRAM: 768GB GPU RAM
Hourly amount in source scope: USD $23.38/hr · Instance
Source billing unit: Instance-hour
240 CPU
720GB RAM
768GB GPU RAM
H100 NVL 8×
vCPU: 140
Region not specified by provider
$3.11 per GPU-hour
Minimum billable instance cost: USD $24.88/hr · Minimum billing duration unknown
Details & source
SKU: H100 NVL | 8 GPU | 6000 storage · Availability: Unknown · C · Display only
Reported GPU name: H100 NVL
Hourly amount in source scope: USD $24.88/hr · Instance
Source billing unit: Instance-hour
1024 GB
H100 SXM 80GB 8×
vCPU: 126
Region not specified by provider
$3.14 per GPU-hour
Minimum billable instance cost: USD $25.12/hr · Minimum billing duration unknown
Details & source
SKU: H100 SXM5 (80GB) | 8 GPU | 2800 storage · Availability: Unknown · C · Display only
Reported GPU name: H100 SXM5 (80GB)
Reported VRAM: 80GB
Hourly amount in source scope: USD $25.12/hr · Instance
Source billing unit: Instance-hour
976 GB
80GB
B300 CC SXM6 268GB 4×
Region not specified by provider
$6.52 per GPU-hour
Minimum billable instance cost: USD $26.09/hr · Minimum billing duration unknown
Details & source
SKU: 4B300.120V.CC.AMD · Availability: Unknown · A · Official API
Reported GPU name: 4x B300 CC SXM6 268GB
Reported VRAM: 1072GB GPU RAM
Hourly amount in source scope: USD $26.09/hr · Instance
Source billing unit: Instance-hour
120 CPU (AMD EPYC)
1020GB RAM
1072GB GPU RAM
H200 SXM5 141GB 4×
Region not specified by provider
$6.91 per GPU-hour
Minimum billable instance cost: USD $27.62/hr · Minimum billing duration unknown
Details & source
SKU: 4H200.141S.176V · Availability: Unknown · A · Official API
Reported GPU name: 4x H200 SXM5 141GB
Reported VRAM: 564GB GPU RAM
Hourly amount in source scope: USD $27.62/hr · Instance
Source billing unit: Instance-hour
176 CPU
680GB RAM
564GB GPU RAM
$3.45 per GPU-hour
Minimum billable instance cost: USD $27.62/hr · Minimum billing duration unknown
Details & source
SKU: 8H200.141S.176V · Availability: Unknown · A · Official API
Reported GPU name: 8x H200 SXM5 141GB
Reported VRAM: 1128GB GPU RAM
Hourly amount in source scope: USD $27.62/hr · Instance
Source billing unit: Instance-hour
176 CPU
1360GB RAM
1128GB GPU RAM
$3.56 per GPU-hour
Minimum billable instance cost: USD $28.51/hr · Initial billing minimum: USD $0.475 / 60 seconds
Details & source
SKU: g7.48xlarge · Availability: Unknown · A · Official API
Reported GPU name: RTX PRO 4500
Reported VRAM: 256 GB
Hourly amount in source scope: USD $28.51/hr · Instance
Source billing unit: Instance-hour
Minimum billing (seconds): 60
256 GB
Linux, shared tenancy, no preinstalled software; storage and data transfer extra.
API record pageH200 NVL 141GB 8×
vCPU: 120
Region not specified by provider
$3.62 per GPU-hour
Minimum billable instance cost: USD $28.96/hr · Minimum billing duration unknown
Details & source
SKU: H200 NVL (141GB) | 8 GPU | 6000 storage · Availability: Unknown · C · Display only
Reported GPU name: H200 NVL (141GB)
Reported VRAM: 141GB
Hourly amount in source scope: USD $28.96/hr · Instance
Source billing unit: Instance-hour
1760 GB
141GB
$3.91 per GPU-hour
Original hourly price: CNY 208.67/hr · Instance
Minimum billable instance cost: USD $31.30/hr · Minimum billing duration unknown
Details & source
SKU: bcc.gn5.c112m896.8A100-40g · Availability: Unknown · C · Display only
Reported GPU name: 8*Nvidia A100 Nvswitch-40g
Reported VRAM: 320GB
Hourly amount in source scope: USD $31.30/hr · Instance
Source billing unit: Instance-hour
Original hourly price: CNY 208.67 · USD conversion rate: 0.150012 · ECB 2026-11-17
112核
896GB
320GB
RTX PRO 6000 Blackwell Server Edition 96GB 8×
vCPU: 192 · RAM (GiB): 2048
us-east-1
$4.14 per GPU-hour
Minimum billable instance cost: USD $33.14/hr · Initial billing minimum: USD $0.552 / 60 seconds
Details & source
SKU: g7e.48xlarge · Availability: Unknown · A · Official API
Reported GPU name: RTX PRO 6000 Blackwell Server Edition
Reported VRAM: 768 GB
Hourly amount in source scope: USD $33.14/hr · Instance
Source billing unit: Instance-hour
Minimum billing (seconds): 60
768 GB
Linux, shared tenancy, no preinstalled software; storage and data transfer extra.
API record page$4.53 per GPU-hour
Original hourly price: CNY 241.43/hr · Instance
Minimum billable instance cost: USD $36.22/hr · Minimum billing duration unknown
Details & source
SKU: bcc.gn5.c112m960.8A100-80g · Availability: Unknown · C · Display only
Reported GPU name: 8*Nvidia A100 Nvswitch-80g
Reported VRAM: 640GB
Hourly amount in source scope: USD $36.22/hr · Instance
Source billing unit: Instance-hour
Original hourly price: CNY 241.43 · USD conversion rate: 0.150012 · ECB 2026-11-17
112核
960GB
640GB
$5.77 per GPU-hour
Original hourly price: EUR 42/hr · Instance
Minimum billable instance cost: USD $46.15/hr · Initial billing minimum: USD $0.769 / 60 seconds
Details & source
SKU: h200-1920-eph.consumption · Availability: Unknown · A · Official API
Reported GPU name: H200
Reported VRAM: 141 GB/GPU
Hourly amount in source scope: USD $46.15/hr · Instance
Source billing unit: Instance-hour
Minimum billing (seconds): 60
Original hourly price: EUR 42 · USD conversion rate: 1.0989 · ECB 2026-11-17
1920 GB
141 GB/GPU
FR subsidiary base Linux public-cloud catalog; exact region and capacity unspecified; local/3-AZ prices not inferred.
API record page$6.00 per GPU-hour
Minimum billable instance cost: USD $48.00/hr · Minimum billing duration unknown
Details & source
SKU: Standard_ND96isr_MI300X_v5 · Availability: Unknown · A · Official API
Reported GPU name: MI300X 192GB
Hourly amount in source scope: USD $48.00/hr · Instance
Source billing unit: Instance-hour
B300 CC SXM6 268GB 4×
Region not specified by provider
$13.04 per GPU-hour
Minimum billable instance cost: USD $52.18/hr · Minimum billing duration unknown
Details & source
SKU: 4B300.120V.CC.AMD · Availability: Unknown · A · Official API
Reported GPU name: 4x B300 CC SXM6 268GB
Reported VRAM: 1072GB GPU RAM
Hourly amount in source scope: USD $52.18/hr · Instance
Source billing unit: Instance-hour
120 CPU (AMD EPYC)
1020GB RAM
1072GB GPU RAM
$6.88 per GPU-hour
Minimum billable instance cost: USD $55.04/hr · Initial billing minimum: USD $0.917 / 60 seconds
Details & source
SKU: p5.48xlarge · Availability: Unknown · A · Official API
Reported GPU name: H100
Reported VRAM: 640 GB HBM3
Hourly amount in source scope: USD $55.04/hr · Instance
Source billing unit: Instance-hour
Minimum billing (seconds): 60
640 GB HBM3
Linux, shared tenancy, no preinstalled software; storage and data transfer extra.
API record pageH200 SXM5 141GB 8×
Region not specified by provider
$6.90 per GPU-hour
Minimum billable instance cost: USD $55.23/hr · Minimum billing duration unknown
Details & source
SKU: 8H200.141S.176V · Availability: Unknown · A · Official API
Reported GPU name: 8x H200 SXM5 141GB
Reported VRAM: 1128GB GPU RAM
Hourly amount in source scope: USD $55.23/hr · Instance
Source billing unit: Instance-hour
176 CPU
1360GB RAM
1128GB GPU RAM
GB300 SXM6 288GB 4×
Region not specified by provider
$14.71 per GPU-hour
Minimum billable instance cost: USD $58.83/hr · Minimum billing duration unknown
Details & source
SKU: 4GB300.128V · Availability: Unknown · A · Official API
Reported GPU name: 4x GB300 SXM6 288GB
Reported VRAM: 1152GB GPU RAM
Hourly amount in source scope: USD $58.83/hr · Instance
Source billing unit: Instance-hour
128 CPU
900GB RAM
1152GB GPU RAM
$10.60 per GPU-hour
Minimum billable instance cost: USD $84.80/hr · Minimum billing duration unknown
Details & source
SKU: Standard_ND96isr_H200_v5 · Availability: Unknown · A · Official API
Reported GPU name: H200 141GB
Hourly amount in source scope: USD $84.80/hr · Instance
Source billing unit: Instance-hour
B300 SXM6 268GB 8×
Region not specified by provider
$12.79 per GPU-hour
Minimum billable instance cost: USD $102.30/hr · Minimum billing duration unknown
Details & source
SKU: 8B300.240V · Availability: Unknown · A · Official API
Reported GPU name: 8x B300 SXM6 268GB
Reported VRAM: 2144GB GPU RAM
Hourly amount in source scope: USD $102.30/hr · Instance
Source billing unit: Instance-hour
240 CPU
2040GB RAM
2144GB GPU RAM
$14.00 per GPU-hour
Minimum billable instance cost: USD $112.00/hr · Minimum billing duration unknown
Details & source
SKU: B110978:BM.GPU.B200.8 · Availability: Unknown · A · Official API
Reported GPU name: OCI - Compute - GPU - B200
Hourly amount in source scope: USD $112.00/hr · Instance
Source billing unit: Instance-hour
128 OCPUs
4096 GB
Global public list rate; region and inventory unspecified. Shape CPU and memory included.
API record page$14.24 per GPU-hour
Minimum billable instance cost: USD $113.93/hr · Initial billing minimum: USD $1.90 / 60 seconds
Details & source
SKU: p6-b200.48xlarge · Availability: Unknown · A · Official API
Reported GPU name: B200
Reported VRAM: 1432 GB HBM3e
Hourly amount in source scope: USD $113.93/hr · Instance
Source billing unit: Instance-hour
Minimum billing (seconds): 60
1432 GB HBM3e
Linux, shared tenancy, no preinstalled software; storage and data transfer extra.
API record page$17.80 per GPU-hour
Minimum billable instance cost: USD $142.42/hr · Initial billing minimum: USD $2.37 / 60 seconds
Details & source
SKU: p6-b300.48xlarge · Availability: Unknown · A · Official API
Reported GPU name: B300
Reported VRAM: 2144 GB HBM3e
Hourly amount in source scope: USD $142.42/hr · Instance
Source billing unit: Instance-hour
Minimum billing (seconds): 60
2144 GB HBM3e
Linux, shared tenancy, no preinstalled software; storage and data transfer extra.
API record pageWhat to check before renting
- Test the exact model, precision, context or audio length and batch size; leave room for caches and temporary allocations.
- Confirm driver, CUDA compute capability or ROCm gfx target, framework and exact quantization kernels.
- Confirm CPU architecture: x86_64 versus aarch64, including container and wheel availability.
- Allow host RAM and disk for downloading and loading the weights, plus temporary files.
- For N > 1, confirm tensor-parallel engine support and GPU interconnect/topology; aggregate VRAM alone is insufficient.
- Check GPU count, minimum billable instance, billing duration, tax, FX, storage and egress. Plan request retries for interruptible Spot capacity.
- Recheck region access, quota and actual stock with the provider; an unknown availability label is not a reservation.
Official sources
- Hugging Face: official model card · https://huggingface.co/Qwen/Qwen2.5-72B-Instruct
- Hugging Face: official model configuration · https://huggingface.co/Qwen/Qwen2.5-72B-Instruct/blob/main/config.json
- Hugging Face: official BF16 weight index · https://huggingface.co/Qwen/Qwen2.5-72B-Instruct/blob/main/model.safetensors.index.json
- Qwen: speed and GPU-memory benchmark · https://qwen.readthedocs.io/en/v2.5/benchmark/speed_benchmark.html
- Qwen: benchmark memory calculation (bytes / 1024³) · https://github.com/QwenLM/Qwen2.5/blob/main/examples/speed-benchmark/speed_benchmark_transformers.py
- Hugging Face: memory estimator and its inference caveats · https://huggingface.co/docs/accelerate/usage_guides/model_size_estimator
- vLLM: GPU installation and software floor · https://docs.vllm.ai/en/stable/getting_started/installation/gpu/
- vLLM: tensor-parallel serving · https://docs.vllm.ai/en/stable/serving/parallelism_scaling/
- PyTorch: stable CUDA release support matrix · https://github.com/pytorch/pytorch/blob/main/RELEASE.md#pytorch-cuda-support-matrix
Know the full job cost before you rent: GPU Budget Kit ($12)