llamaperf
Calculator
GPUs
Engines
Guides
Blog
Community
MCP
Advertise
Submit
Sign in
Submit
llamaperf
/
M4 Max 128GB
/
report
Muse 30B Glimmer
on
M4 Max 128GB
· WebGPU
Apple hardware
Sep 7, 2026
Share
X
Reddit
LinkedIn
Bluesky
Copy
link
Throughput
25.0 t/s gen
Quant
UD-Q2_K_XL (GGUF)
System RAM
128 GB
Summary
User reports running a model in-browser with custom WebGPU kernels, at speed comparable to llama.cpp.
View source ↗
Feedback