M3 Ultra 96GB vs AMD Radeon 890M for local LLMs
3 reports on the M3 Ultra 96GB and 1 on the AMD Radeon 890M
Which is faster for local LLMs?
On Qwen3.5 35B · 3B active at 4-bit, the M3 Ultra 96GB runs at 109.7 tokens per second (one run) and the AMD Radeon 890M at 14.9 (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
- 96GB unified
- Memory bandwidth
- 819 GB/s
- FP16 compute
- 21.3 TFLOPS
- Reports
- 3
- Memory
- shared memory unified
- Memory bandwidth
- not on record
- FP16 compute
- not on record
- Reports
- 1
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 | M3 Ultra 96GB | AMD Radeon 890M |
|---|---|---|
| Qwen3.5 35B · 3B active4-bit | 109.7one run | 14.9one 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 | M3 Ultra 96GB | AMD Radeon 890M |
|---|---|---|
| Qwen3.5 35B · 3B active | 57.1 t/sQ8_0 | not in the calculator yet |
| Qwen3.8 27B | 7.9 t/sFP16 | not in the calculator yet |
| Qwen3.8 125B · 6B active | 55.1 t/sQ3_K_M | not in the calculator yet |
| DeepSeek V4 Flash 284B · 13B active | doesn't fit at these settings | not in the calculator yet |
| Qwen3.6 35B · 3B active | 57.1 t/sQ8_0 | not in the calculator yet |
| Qwen3.6 27B | 7.9 t/sFP16 | not in the calculator yet |
| Gemma 4 26B · 4B active | 19.0 t/sFP16 | not in the calculator yet |
Change the context, RAM or model in the calculator: M3 Ultra 96GB · AMD Radeon 890M
Every report on each card, with its source: M3 Ultra 96GB · AMD Radeon 890M