Have you tried with a single CPU to get rid of the NUMA penalty? I understand this likely means halving the memory but I am interested in how much of a difference it makes
I have (192GB machine with two CPUs), pretty much does the trick. It just runs some small models used for embedding, etc. and has those on one CPU / memory node and all the Docker containers on the other one.c
To me context means everything.
Tokens per second is a great metric but in the real world context window is the deal breaker when a real use case is on the table.
Is it just me or does this post not mention how much RAM they had? I would love to know - I have a dual-Xeon 1U screamer with 96GB of DDR4 RDIMM just sitting around...
FWIW I'm getting a hardware max of 20 tok/s (approx topping out the GPU's compute) on my custom local diffusiongemma port running on an M3.
Author here. The short version: a viral post ran Gemma 4 on a 2016 Xeon; my Xeons are 2013, and the fork it used assumes AVX2, which Ivy Bridge doesn't have. The build failure was easy. The fun bug was the silent one: two MoE graph ops with no dispatch case on non-AVX2 builds, so every expert FFN output was uninitialized memory. Deterministic, NaN-free, fluent-looking multilingual gibberish.
The fix is open upstream as PR #2138 (https://github.com/ikawrakow/ik_llama.cpp/pull/2138), awaiting review. Fair warning on the AI angle: the patch was written by Claude at my direction. The post is explicit about which parts were me and which weren't. Happy to answer questions about either the bug or the workflow.
Here's the thing: life also imitates art. If you invert your load-bearing assumption, it could be that he just reads too much slop. But my honest take? You might be right.
Here's my report running several different models on a dual Xeon with 256 GB of DDR4 and no GPU.
https://gist.github.com/hparadiz/f3596d00a62d8ebb2dadcc46ee5...
Have you tried with a single CPU to get rid of the NUMA penalty? I understand this likely means halving the memory but I am interested in how much of a difference it makes
I have (192GB machine with two CPUs), pretty much does the trick. It just runs some small models used for embedding, etc. and has those on one CPU / memory node and all the Docker containers on the other one.c
Thank you for sharing!
That's quite slow I'm getting 8-12 t/s on a 13 year old CPU. (Speed varies by context size and other settings who knows)
https://news.ycombinator.com/item?id=48354801
Thank you for sharing / linking!
He's shown me his set up in his basement. It's sick! Talk about your 3d printer next!
Author here, it looks like my original comment was flagged for some reason. The fix is open upstream as PR #2138 (https://github.com/ikawrakow/ik_llama.cpp/pull/2138)
Need to run this on my Xeons with AMX
To me context means everything. Tokens per second is a great metric but in the real world context window is the deal breaker when a real use case is on the table.
Is it just me or does this post not mention how much RAM they had? I would love to know - I have a dual-Xeon 1U screamer with 96GB of DDR4 RDIMM just sitting around...
FWIW I'm getting a hardware max of 20 tok/s (approx topping out the GPU's compute) on my custom local diffusiongemma port running on an M3.
hey, I’m the author. That box has 384gb, but loading the model “only” uses about 80gb.
Related:
A 10 year old Xeon is all you need
https://news.ycombinator.com/item?id=48353348
Author here. The short version: a viral post ran Gemma 4 on a 2016 Xeon; my Xeons are 2013, and the fork it used assumes AVX2, which Ivy Bridge doesn't have. The build failure was easy. The fun bug was the silent one: two MoE graph ops with no dispatch case on non-AVX2 builds, so every expert FFN output was uninitialized memory. Deterministic, NaN-free, fluent-looking multilingual gibberish.
The fix is open upstream as PR #2138 (https://github.com/ikawrakow/ik_llama.cpp/pull/2138), awaiting review. Fair warning on the AI angle: the patch was written by Claude at my direction. The post is explicit about which parts were me and which weren't. Happy to answer questions about either the bug or the workflow.
This reads as pretty clearly AI-generated text, which is against HN guidelines.
The PR? He said it was AI in the comment you replied to...
I don't think the post itself reads like AI at all, but that's just me.
Here's the thing: life also imitates art. If you invert your load-bearing assumption, it could be that he just reads too much slop. But my honest take? You might be right.
Truly amazing. This gives a peek into the future for what's possible.
Sorry for asking here but literally nobody knows:
Android studio connected to a local model disconnects automatiacally after 10 minutes. How set this limit to 12 hours or remove it completely?
I could run my LM studio model all night... but I cant, since Android studio times out after a hard limit of 10M.
This is not related to number of tokens. I do 130 sec.