llamaperf

NVIDIA CMP 170HX 64GB (unlocked) vs AMD Strix Halo 128GB for local LLMs

15 reports on the NVIDIA CMP 170HX 64GB (unlocked) and 348 on the AMD Strix Halo 128GB

Which is faster for local LLMs?

On Qwen3.8 27B at 8-bit, the NVIDIA CMP 170HX 64GB (unlocked) has 50.0 generation tokens/s (one run); prompt processing 3468.0 tokens/s (one run); the AMD Strix Halo 128GB has 7.5 generation tokens/s (3 runs); prompt processing 257.8 tokens/s (2 runs). 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 5.8x higher on the NVIDIA CMP 170HX 64GB (unlocked) (1,493 GB/s on the NVIDIA CMP 170HX 64GB (unlocked), 256 on the AMD Strix Halo 128GB). Across 2 shared models with generation measurements on both cards, the NVIDIA CMP 170HX 64GB (unlocked)'s median is higher on 2 and the AMD Strix Halo 128GB's on 0.

NVIDIA CMP 170HX 64GB (unlocked)
Memory
64GB
Memory bandwidth
1,493 GB/s
FP16 compute
not on record
Reports
15
AMD Strix Halo 128GB
Memory
128GB unified
Memory bandwidth
256 GB/s
FP16 compute
59 TFLOPS
Reports
348

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

ModelNVIDIA CMP 170HX 64GB (unlocked)AMD Strix Halo 128GB
Qwen3.8 27B8-bit50.0one runPP 3468.0 (one run)vLLM, 1 not statedcontext not stated7.53 runsPP 257.8 (2 runs)llama.cpp, 1 not statedcontext not stated
Qwen3.6 27B4-bit38.0one runPP 907.0 (one run)llama.cppcontext not stated12.0one runPP 322.0 (one run)engine not statedcontext 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.

ModelNVIDIA CMP 170HX 64GB (unlocked)AMD Strix Halo 128GB
Qwen3.8 27B30.4 t/sQ8_03.5 t/sFP16
Qwen3.6 27B30.4 t/sQ8_03.5 t/sFP16
Qwen3.8 125B · 6B active127.0 t/sQ3_K_S12.6 t/sQ6_K
Qwen3.6 35B · 3B active109.5 t/sQ8_012.6 t/sFP16
DeepSeek V4 Flash 284B · 13B activedoesn't fit at these settings33.5 t/sQ2_K
DeepSeek V4.1 Flash 552B · 16B activedoesn't fit at these settingsdoesn't fit at these settings
Gemma 4 26B · 4B active82.1 t/sQ8_09.5 t/sFP16
GLM-5.3 320B · 18B activedoesn't fit at these settings11.1 t/sQ2_K

Change the context, RAM or model in the calculator: NVIDIA CMP 170HX 64GB (unlocked) · AMD Strix Halo 128GB

Every report on each card, with its source: NVIDIA CMP 170HX 64GB (unlocked) · AMD Strix Halo 128GB

More comparisons