Tesla V100 DGXS 32 GB holds 362 of the models in our catalogue and is, in practice, a IQ4_XS card — the largest it takes is Nemotron-H 47B at IQ4_XS.
Neighbours in memory rather than in price: memory decides whether a card can do the job at all, so two cards of the same size at different prices is the comparison you are making. Ordered by memory, then bandwidth — not by the value column, which is worked out from bandwidth and price alone and therefore rewards a cheap card whatever its software stack does to that bandwidth in practice. Read it as one input, not as a ranking.
NVIDIA Tesla V100 DGXS 32 GB — 32 GB VRAM.
- Brand
- NVIDIA
- VRAM
- 32 GB HBM2
- Bandwidth
- 897 GB/s
- FP16 Compute
- 31.3 TFLOPS
- FP32 Compute
- 15.7 TFLOPS
- CUDA Cores
- 5,120
- Tensor Cores
- 640
- TDP
- 250 W
- Architecture
- Volta
With 32 GB VRAM and 897 GB/s bandwidth, this GPU handles models up to 41.9B parameters.
Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~103 tok/s.
| Model | Size | VRAM Q4 | tok/s | Avg |
|---|---|---|---|---|
| Phi-3.5 MoE 42B | 41.9B | 26.1 GB | 109 | 56.7 |
| Falcon 40B | 40B | 24.9 GB | 18 | 20.9 |
| Flint | 39.79B | 24.8 GB | 287 | 77.3 |
| InternVL3 38B | 38B | 23.7 GB | 19 | 78.9 |
| K2 Horizon MoVA 36B-A4B | 37.44B | 23.4 GB | 179 | 45.7 |
| Seed-OSS 36B Instruct | 36B | 22.5 GB | 20 | 54.4 |
| MiniCPM-V 4.7 35B-A3B | 35.21B | 22.0 GB | 239 | 76.4 |
| c4ai-command-r-v01 35B | 35B | 21.9 GB | 21 | 27.5 |
| Qwen 3.5 35B A3B | 35B | 21.9 GB | 239 | 56.1 |
| Qwen 3.6 35B A3B | 35B | 21.9 GB | 239 | 53.3 |
| K2 Horizon 32B | 34.78B | 21.7 GB | 21 | 54.0 |
| Nous Capybara 34B | 34.4B | 21.5 GB | 21 | 42.0 |
| Yi-1.5 34B | 34.4B | 21.5 GB | 21 | 45.3 |
| Falcon-H1 34B | 34B | 21.3 GB | 21 | 58.6 |
| CodeLlama 34B | 34B | 21.3 GB | 21 | 25.4 |
| Nous Hermes 2 34B | 34B | 21.3 GB | 21 | 47.0 |
| Phind CodeLlama 34B | 34B | 21.3 GB | 21 | 68.1 |
| LLaVA-1.6 Yi 34B | 34B | 21.3 GB | 21 | 47.4 |
| WizardCoder Python 34B | 34B | 21.3 GB | 21 | 73.2 |
| Yi 34B | 34B | 21.3 GB | 21 | 33.4 |