Intel Arc Pro B70 vs M2 Max 96GB for local LLMs
14 reports on the Intel Arc Pro B70 and 7 on the M2 Max 96GB
Which is faster for local LLMs?
On Qwen3 8B at 4-bit, the Intel Arc Pro B70 runs at 77.7 tokens per second (one run) and the M2 Max 96GB at 43.0 (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 48% higher on the Intel Arc Pro B70 (608 GB/s on the Intel Arc Pro B70, 410 on the M2 Max 96GB).
- Memory
- 32GB
- Memory bandwidth
- 608 GB/s
- FP16 compute
- 183.5 TFLOPS
- Reports
- 14
- Memory
- 96GB unified
- Memory bandwidth
- 410 GB/s
- FP16 compute
- 13.6 TFLOPS
- Reports
- 7
Measured on both
Median tokens per second 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 quant level. Engines and context lengths can still differ between the runs. How to read these
| Model | Intel Arc Pro B70 | M2 Max 96GB |
|---|---|---|
| Qwen3 8B4-bit | 77.7one run | 43.0one run |
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 | Intel Arc Pro B70 | M2 Max 96GB |
|---|---|---|
| Qwen3 8B | 20.9 t/sFP16 | 17.6 t/sFP16 |
| Qwen3.8 27B | 15.2 t/sQ6_K | 5.2 t/sFP16 |
| Qwen3.8 125B · 6B active | 18.6 t/sQ2_K, experts in RAM | 36.2 t/sQ3_K_M |
| DeepSeek V4 Flash 284B · 13B active | doesn't fit at these settings | doesn't fit at these settings |
| Qwen3.6 35B · 3B active | 55.3 t/sQ5_K_M | 37.6 t/sQ8_0 |
| Qwen3.6 27B | 15.2 t/sQ6_K | 5.2 t/sFP16 |
| Gemma 4 26B · 4B active | 38.8 t/sQ6_K | 12.5 t/sFP16 |
Change the context, RAM or model in the calculator: Intel Arc Pro B70 · M2 Max 96GB
Every report on each card, with its source: Intel Arc Pro B70 · M2 Max 96GB