AMD Strix Halo 128GB vs NVIDIA Tesla P40 24GB for local LLMs
328 reports on the AMD Strix Halo 128GB and 12 on the NVIDIA Tesla P40 24GB
Which is faster for local LLMs?
On Qwen3.6 35B · 3B active at 4-bit, the AMD Strix Halo 128GB has 60.0 generation tokens/s (7 runs); prompt processing 1005.3 tokens/s (7 runs); the NVIDIA Tesla P40 24GB has 9.7 generation tokens/s (one run). Those are community reports whose engines and settings differ, so the gap is a rough sign, not a controlled measure of the two cards. Memory bandwidth, which caps how fast a card can write, is 35% higher on the NVIDIA Tesla P40 24GB (256 GB/s on the AMD Strix Halo 128GB, 346 on the NVIDIA Tesla P40 24GB). Across 2 shared models with generation measurements on both cards, the AMD Strix Halo 128GB's median is higher on 2 and the NVIDIA Tesla P40 24GB's on 0.
- Memory
- 128GB unified
- Memory bandwidth
- 256 GB/s
- FP16 compute
- 59 TFLOPS
- Reports
- 328
- Memory
- 24GB
- Memory bandwidth
- 346 GB/s
- FP16 compute
- 12 TFLOPS
- Reports
- 12
Measured on both
Generation and prompt processing speeds from plain runs (one device, one request, no speculative decoding, the whole model in the device's memory) of the same model size at the same bit class. Each side shows the engines and reported context behind its median. Context can mean a configured window or prompt depth, so these are community comparisons with differing settings. How to read these
| Model | AMD Strix Halo 128GB | NVIDIA Tesla P40 24GB |
|---|---|---|
| Qwen3.6 35B · 3B active4-bit | 60.07 runsPP 1005.3 (7 runs)llama.cpp, 3 not stated30K context, 6 not stated | 9.7one runengine not statedcontext not stated |
| Ornith1.5 35B · 3B active4-bit | 72.6one runPP 1504.9 (one run)llama.cppcontext not stated | 60.8one runPP 1104.0 (one run)llama.cppcontext not stated |
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. Where a model was also measured above, the measurement wins.
| Model | AMD Strix Halo 128GB | NVIDIA Tesla P40 24GB |
|---|---|---|
| Qwen3.6 35B · 3B active | 12.6 t/sFP16 | 44.4 t/sQ4_K_S |
| Ornith1.5 35B · 3B active | 12.6 t/sFP16 | 44.4 t/sQ4_K_S |
| Qwen3.8 27B | 3.5 t/sFP16 | 12.5 t/sQ4_K_M |
| Qwen3.8 125B · 6B active | 12.6 t/sQ6_K | 16.9 t/sQ2_K, experts in RAM |
| DeepSeek V4 Flash 284B · 13B active | 33.5 t/sQ2_K | doesn't fit at these settings |
| Qwen3.6 27B | 3.5 t/sFP16 | 12.5 t/sQ4_K_M |
| Gemma 4 26B · 4B active | 9.5 t/sFP16 | 28.2 t/sQ5_K_S |
| DeepSeek V4.1 Flash 552B · 16B active | doesn't fit at these settings | doesn't fit at these settings |
Change the context, RAM or model in the calculator: AMD Strix Halo 128GB · NVIDIA Tesla P40 24GB
Every report on each card, with its source: AMD Strix Halo 128GB · NVIDIA Tesla P40 24GB