Skip to content

Add M3 Ultra (512 GB) Flash benchmark. - #779

Open
daemon2k3 wants to merge 1 commit into
antirez:mainfrom
daemon2k3:bench-m3-ultra-flash
Open

Add M3 Ultra (512 GB) Flash benchmark.#779
daemon2k3 wants to merge 1 commit into
antirez:mainfrom
daemon2k3:bench-m3-ultra-flash

Conversation

@daemon2k3

Copy link
Copy Markdown

ds4-bench, DeepSeek V4 Flash (q2 resident), ctx frontier 2048..65536: prefill 596 -> 391 t/s, steady decode 44.1 -> 34.5 t/s. Server-class launch, DSpark off for the raw-engine curve.

ds4-bench, DeepSeek V4 Flash (q2 resident), ctx frontier 2048..65536:
prefill 596 -> 391 t/s, steady decode 44.1 -> 34.5 t/s.
Server-class launch, DSpark off for the raw-engine curve.
@GiorgioOppo

Copy link
Copy Markdown

Hi @daemon2k3 , would you like to try our implementation #621 with a slightly different GGUF file on your Mac?

@daemon2k3

Copy link
Copy Markdown
Author

@GiorgioOppo - to be honest, my target here was to improve prefill as much as possible :). Posted this just because there weren't benches in the repo. But hey, I'll keep this in mind for a later run.

@GiorgioOppo

Copy link
Copy Markdown

@daemon2k3, Thanks anyway for your availability

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants