Can I Run DeepSeek-R1-Distill-Qwen-32B on NVIDIA L40S?
作者： Jakub Rusinowski · 最后更新： 2026年7月21日
Yes, comfortably — DeepSeek R1 Distill Qwen 32B at Q4_K_M needs 22.3 GB at 8K context (19.3 GB weights + 3 GB KV cache/overhead), leaving ~25.7 GB of the NVIDIA L40S's 48 GB, at ~36 tok/s (est.), with room for up to 64K context.
These figures are for DeepSeek-R1-Distill-Qwen-32B, a distill of Qwen2.5-32B — not the full DeepSeek R1. The full DeepSeek R1 (671B) needs about 405 GB of weights at Q4_K_M and is a different model.
See what else this hardware can run →
作为亚马逊联盟成员，我们从符合条件的购买中获得收入。云 GPU 链接为推荐链接——我们可能获得佣金，您无需额外付费。
NVIDIA L40S Specs
| VRAM | 48 GB |
| Memory Bandwidth | 864 GB/s |
DeepSeek R1 Distill Qwen 32B on the NVIDIA L40S: VRAM by quantization
| Quant | VRAM needed | Fits 48 GB? | Max context | Whole PC |
|---|---|---|---|---|
| F16 | 66.9 GB | ✗ No | — | — |
| Q8_0 | 36.9 GB | ✓ Yes | 32K | Build for this |
| Q6_K | 29.2 GB | ✓ Yes | 64K | — |
| Q5_K_M | 25.6 GB | ✓ Yes | 64K | Build for this |
| Q4_K_M | 22.3 GB | ✓ Yes | 64K | Build for this |
| Q3_K_M | 16.6 GB | ✓ Yes | 64K | — |
| Q2_K | 13.5 GB | ✓ Yes | 128K | — |
Assumes an 8K-token context with an f16 KV cache. A longer window needs more; a quantized KV cache needs less. “Max context” is the largest window that still fits in 48 GB. Figures are estimates from parameter count, quantization and memory bandwidth — the analyzer lets you tune KV-cache quant and context.
Context costs VRAM too. At Q4_K_M on the NVIDIA L40S: 8K 22.3 GB · 32K 28.7 GB · 128K 54.5 GB ✗. Past 128K it no longer fits 48 GB — a q8_0 KV cache buys roughly half of that back, which the calculator will price for you.
No compatible GPU? DeepSeek R1 Distill Qwen 32B on 32 GB of system RAM, CPU only: runs comfortably.
DeepSeek R1 Sizes That Fit the NVIDIA L40S (at 8K context)
| DeepSeek R1 Distill Qwen 32B | Q4_K_M · 22.3 GB at 8K context · ~36 tok/s (est.) |
| DeepSeek R1 Distill Qwen 14B | Q4_K_M · 10.6 GB at 8K context · ~73 tok/s (est.) |
| DeepSeek R1 Distill Llama 8B | Q4_K_M · 6.7 GB at 8K context · ~112 tok/s (est.) |
At 2 hrs/day, buying (~$7,499) beats renting at $0.34/hr after about 30.6 years.
Affiliate links — we may earn a commission if you sign up, at no extra cost to you.
Cloud rates verified 2026-07 — estimates, and marketplace prices vary. Buying price is GPU MSRP only, not a full PC.
FAQ
DeepSeek R1 on the NVIDIA L40S — does it fit?
Yes, comfortably — DeepSeek R1 Distill Qwen 32B at Q4_K_M needs 22.3 GB at 8K context (19.3 GB weights + 3 GB KV cache/overhead), leaving ~25.7 GB of the NVIDIA L40S's 48 GB, at ~36 tok/s (est.), with room for up to 64K context.
Which DeepSeek R1 variant fits best on the NVIDIA L40S?
DeepSeek R1 Distill Qwen 32B at Q4_K_M quantization needs 22.3 GB at 8K context (19.3 GB weights + 3 GB KV cache/overhead), estimated ~36 tokens/sec, up to 64K context.
Every DeepSeek R1 size on the NVIDIA L40S
| Size | VRAM at 8K | Verdict | Speed |
|---|---|---|---|
| DeepSeek R1 (671B) | 406.5 GB | Does not fit | — |
| DeepSeek R1 Distill Qwen 14B | 10.6 GB | Runs | ~73 tok/s |
| DeepSeek R1 Distill Llama 8B | 6.7 GB | Runs | ~112 tok/s |
DeepSeek R1 on Other GPUs
- DeepSeek R1 on NVIDIA RTX 6000 Ada Generation
- DeepSeek R1 on Apple M3 Pro
- DeepSeek R1 on NVIDIA GeForce RTX 5090
- DeepSeek R1 on Apple M4
- DeepSeek R1 on Apple M2 Pro
Popular Models on the NVIDIA L40S
VRAM Tier
Troubleshooting
Buying Guide
← Can I Run It? | DeepSeek R1 Model Page | NVIDIA L40S GPU Page | VRAM calculator | Check Your Hardware