DeepSeek-V4.1-Flash-1017
3recorded runs
190.0best tok/s
3verified
What the DeepSeek-V4.1-Flash-1017 numbers mean
The fastest verified result on our board is 190.0 output tok/s on an DGX Spark x4 at EXL3 3.5bpw with a 300000-token context. That is about 5.3 ms per generated token — 500 tokens takes roughly 2.6 seconds. Across the 3 records we hold for this model the range is 76.3–190.0 tok/s on 2 GPUs, because quantisation, context length and the engine build all move the number.
How much VRAM does DeepSeek-V4.1-Flash-1017 need?
DeepSeek-V4.1-Flash-1017 has 552.0B parameters. These estimates use the same formula as the VRAM calculator: weights, plus about 15% for a moderate-context KV cache and runtime overhead, plus a small engine floor.
| Quantisation | Estimated VRAM | Smallest card that fits |
|---|---|---|
| Q4_K_M (4-bit) | 357.9 GB | multi-GPU or CPU offload |
| Q8_0 (8-bit) | 675.3 GB | multi-GPU or CPU offload |
| FP16 (16-bit) | 1270.4 GB | multi-GPU or CPU offload |
Fastest GPUs for this model (best tok/s)
- DGX Spark x4190.0
- DGX Spark x3152.9
Models with a similar footprint
| Model | Params | Q4_K_M weights |
|---|---|---|
| DeepSeek-V4.1-Flash | 552B | 310.5 GB |
| GLM-5.2 | 753B | 423.6 GB |
| GLM-5.3-Flash | 321B | 180.6 GB |
| DeepSeek-V4-Flash-Vision-Exp | 305B | 171.6 GB |
| keys-DeepSeekV4Flash-Vision-EXP-ablit | 305B | 171.6 GB |
| DeepSeek-V4-Flash-0731 | 304B | 171.0 GB |
| GPU | Model | Quant | Backend | Ctx | tok/s | prompt tok/s | VRAM GB | Source |
|---|---|---|---|---|---|---|---|---|
| DGX Spark x4 · 128 GB | DeepSeek-V4.1-Flash-1017 | EXL3 3.5bpw | vllm bot-lab-21's EXL3 build + drowzeys' … | 300000 | 190.0 ✓ | — | — | src |
| DGX Spark x3 · 128 GB | DeepSeek-V4.1-Flash-1017 | EXL3 3.5bpw | vllm bot-lab-21's EXL3 build + drowzeys' … | 300000 | 152.9 ✓ | — | — | src |
| DGX Spark x4 · 128 GB | DeepSeek-V4.1-Flash-1017 | MXFP4+MXFP8 | vllm dsv41-feat (commit e47aa780b) | 300000 | 76.3 ✓ | — | — | src |