V100 32GB
nvidia · volta · 32 GB HBM2 · 300W TDP
VRAM
32 GB
BF16 TFLOPS
28.3
Bandwidth
900 GB/s
From
$0.19/hr
Spec Sheet
Pricing by Provider
| Provider | On-Demand | Reserved | Spot | Badge |
|---|---|---|---|---|
| vast_ai | $0.35/hr | - | $0.19/hr | Cheapest |
| tensordock | $0.29/hr | - | $0.19/hr | |
| runpod | $0.49/hr | - | $0.29/hr | |
| aws | $3.06/hr | $1.96/hr | $0.92/hr |
Compatible Models (260)
Single GPU (176 models)
Multi-GPU (84 models)
Training Capabilities
Estimated GPU count for full fine-tuning (AdamW, BF16) and QLoRA
| Model Size | Full Fine-Tune | QLoRA |
|---|---|---|
| 7B model | 5 GPUs | 1 GPU |
| 13B model | 8 GPUs | 1 GPU |
| 70B model | 42 GPUs | 2 GPUs |
Energy Efficiency
Estimated tokens/second per Watt for popular models
Similar GPUs
| GPU | VRAM | BF16 TFLOPS | BW (GB/s) | From |
|---|---|---|---|---|
| V100 16GB | 16 GB | 28.3 | 900 | $0.15/hr |
| RTX 5090 | 32 GB | 210 | 1792 | $0.89/hr |
| Instinct MI100 | 32 GB | 184.6 | 1229 | $0.40/hr |
| TPU v4 | 32 GB | 275 | 1200 | $2.25/hr |
| TPU v6e (Trillium) | 32 GB | 460 | 1640 | $1.75/hr |
Methodology Note
Performance estimates for the V100 32GBare based on InferenceBench's roofline performance model with CUDA kernel-level optimization including FlashAttention v2 and PagedAttention. Memory calculations account for model weights (32 GB HBM2 available), KV-cache allocation, and activation memory. Throughput predictions use the V100 32GB's rated 900 GB/s memory bandwidth and 28.3 BF16 TFLOPS compute capacity as roofline ceilings, with empirical correction factors per GPU architecture (volta). See our full methodology.
Frequently Asked Questions
How many AI models can run on V100 32GB?
The V100 32GB can run 260 AI models from our database within a single node. Compatible models range across various parameter sizes depending on the quantization precision (BF16, FP8, INT4). Smaller models fit on a single GPU while larger models may require multi-GPU setups up to 8x V100 32GB.
What is the V100 32GB inference throughput?
The V100 32GB delivers 28.3 BF16 TFLOPS and 28.3 FP8 TFLOPS with 900 GB/s memory bandwidth. Actual inference throughput (tokens/sec) depends on the model size, precision, and batch size. Use our calculator for model-specific throughput estimates.
How much does V100 32GB cost per hour?
The V100 32GB is available starting from $0.19/hour via vast_ai. Prices vary by provider and pricing tier (on-demand, reserved, spot). Compare pricing across all providers in the table above.