llamaperf

M3 Ultra 256GB vs AMD RX 9070 XT 16GB for local LLMs

7 reports on the M3 Ultra 256GB and 10 on the AMD RX 9070 XT 16GB

Which is faster for local LLMs?

On Qwen3.5 9B at 4-bit, the M3 Ultra 256GB runs at 107.5 tokens per second (one run) and the AMD RX 9070 XT 16GB at 7.9 (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 bandwidth, which caps how fast a card can write, is 28% higher on the M3 Ultra 256GB (819 GB/s on the M3 Ultra 256GB, 640 on the AMD RX 9070 XT 16GB).

M3 Ultra 256GB
Memory
256GB unified
Memory bandwidth
819 GB/s
FP16 compute
28.4 TFLOPS
Reports
7
AMD RX 9070 XT 16GB
Memory
16GB
Memory bandwidth
640 GB/s
FP16 compute
195 TFLOPS
Reports
10

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

ModelM3 Ultra 256GBAMD RX 9070 XT 16GB
Qwen3.5 9B4-bit107.5one run7.9one 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.

ModelM3 Ultra 256GBAMD RX 9070 XT 16GB
Qwen3.5 9B23.9 t/sFP1639.1 t/sQ8_0
Qwen3.8 27B8.0 t/sFP1641.7 t/sQ2_K
Qwen3.8 125B · 6B active27.3 t/sQ8_0doesn't fit at these settings
Qwen3.6 35B · 3B active28.7 t/sFP16138.0 t/sQ2_K
DeepSeek V4 Flash 284B · 13B active36.0 t/sQ4_K_Mdoesn't fit at these settings
Qwen3.6 27B8.0 t/sFP1641.7 t/sQ2_K
Gemma 4 26B · 4B active21.5 t/sFP1693.0 t/sQ3_K_S

Change the context, RAM or model in the calculator: M3 Ultra 256GB · AMD RX 9070 XT 16GB

Every report on each card, with its source: M3 Ultra 256GB · AMD RX 9070 XT 16GB

More comparisons