NVIDIA DGX Spark vs NVIDIA L40S for local LLMs
89 reports on the NVIDIA DGX Spark and 2 on the NVIDIA L40S
Which is faster for local LLMs?
Nobody has reported the same model at the same quant on both cards yet, so there's no measured head-to-head. Memory bandwidth, which caps how fast a card can write, is 3.2x higher on the NVIDIA L40S (273 GB/s on the NVIDIA DGX Spark, 864 on the NVIDIA L40S). The estimates below compare them on the models people run most.
NVIDIA DGX Spark
- Memory
- 128GB unified
- Memory bandwidth
- 273 GB/s
- FP16 compute
- 125 TFLOPS
- Reports
- 89
NVIDIA L40S
- Memory
- 48GB
- Memory bandwidth
- 864 GB/s
- FP16 compute
- 362.1 TFLOPS
- Reports
- 2
Estimated by the calculator
The quant the calculator recommends for each card and its estimated speed at a 32,768-token context with 32 GB of system RAM, from memory bandwidth and the model's shape.
| Model | NVIDIA DGX Spark | NVIDIA L40S |
|---|---|---|
| Qwen3.8 27B | 4.0 t/sFP16 | 17.6 t/sQ8_0 |
| Qwen3.8 125B · 6B active | 14.4 t/sQ6_K | 80.1 t/sQ2_K |
| Qwen3.6 35B · 3B active | 14.5 t/sFP16 | 55.4 t/sQ8_0 |
| DeepSeek V4 Flash 284B · 13B active | 38.3 t/sQ2_K | doesn't fit at these settings |
| Qwen3.6 27B | 4.0 t/sFP16 | 17.6 t/sQ8_0 |
| Gemma 4 26B · 4B active | 10.8 t/sFP16 | 47.5 t/sQ8_0 |
Change the context, RAM or model in the calculator: NVIDIA DGX Spark · NVIDIA L40S
Every report on each card, with its source: NVIDIA DGX Spark · NVIDIA L40S