llamaperf

NVIDIA Jetson AGX Thor 128GB vs M3 Ultra 512GB for local LLMs

8 reports on the NVIDIA Jetson AGX Thor 128GB and 43 on the M3 Ultra 512GB

Which is faster for local LLMs?

On Qwen3.8 125B · 6B active at 4-bit, the NVIDIA Jetson AGX Thor 128GB has 36.7 generation tokens/s (one run); prompt processing 899.0 tokens/s (one run); the M3 Ultra 512GB has 28.0 generation tokens/s (2 runs); prompt processing 558.8 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 3.0x higher on the M3 Ultra 512GB (273 GB/s on the NVIDIA Jetson AGX Thor 128GB, 819 on the M3 Ultra 512GB). Across 2 shared models with generation measurements on both cards, the NVIDIA Jetson AGX Thor 128GB's median is higher on 1 and the M3 Ultra 512GB's on 1.

NVIDIA Jetson AGX Thor 128GB
Memory
128GB unified
Memory bandwidth
273 GB/s
FP16 compute
258.8 TFLOPS
Reports
8
M3 Ultra 512GB
Memory
512GB unified
Memory bandwidth
819 GB/s
FP16 compute
28.4 TFLOPS
Reports
43

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 Jetson AGX Thor 128GBM3 Ultra 512GB
Qwen3.8 125B · 6B active4-bit36.7one runPP 899.0 (one run)engine not stated256K context28.02 runsPP 558.8 (one run)llama.cpp, oMLX250K context, 1 not stated
Qwen3.8 27B4-bit12.0one runengine not statedcontext not stated31.8one runPP 371.6 (2 runs)MLX, oMLX64K context, 1 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 Jetson AGX Thor 128GBM3 Ultra 512GB
Qwen3.8 125B · 6B active10.0 t/sQ6_K13.6 t/sFP16
Qwen3.8 27B2.8 t/sFP168.0 t/sFP16
Qwen3.6 35B · 3B active10.0 t/sFP1628.7 t/sFP16
DeepSeek V4 Flash 284B · 13B active26.5 t/sQ2_K22.6 t/sQ8_0
DeepSeek V4.1 Flash 552B · 16B activedoesn't fit at these settings30.0 t/sQ4_K_M
Qwen3.6 27B2.8 t/sFP168.0 t/sFP16
Gemma 4 26B · 4B active7.5 t/sFP1621.5 t/sFP16
GLM-5.3 320B · 18B active8.8 t/sQ2_K7.4 t/sQ8_0

Change the context, RAM or model in the calculator: NVIDIA Jetson AGX Thor 128GB · M3 Ultra 512GB

Every report on each card, with its source: NVIDIA Jetson AGX Thor 128GB · M3 Ultra 512GB

More comparisons