llamaperf
Calculator
GPUs
Engines
Guides
Blog
Community
MCP
Advertise
Submit
Sign in
Submit
llamaperf
/
M5 Pro 64GB
/
report
DeepSeek V4 Flash 284B (13B active)
on
M5 Pro 64GB
· ds4
Tone: mixed
Apple hardware
Sep 7, 2026
Share
X
Reddit
LinkedIn
Bluesky
Copy
link
Throughput
2.0 t/s gen
Quant
mixed 4/8-bit
System RAM
64 GB
Summary
User reports 2.04 t/s with an 8 GB expert cache and 1.23 t/s with a 32 GB expert cache, streaming from SSD. Speculative decoding and prefetch both decreased throughput. Codebook 2-bit quantization failed the quality gate.
View source ↗
Feedback