>tfw fell for 2x32GB DDR5-8000 RAM meme
wrong pic ig
>>109250005>8000hzhow stable is it on a scale of 1 to 10?
>tfw fell for the 4x128gb ram meme
>>109250005How so you can sell half of it to more then you paid money back?
>>1092500056000Mhz is all you need
>>109250005>tfw fell for the 4GB RAM meme
>>109250005>he fell for the 512Gb of ram meme...
>>109251751must be pretty good for llm inference.
>>109251754Nta, it isn't. Low ram bandwidth
>>109251760Not sure if just wrong or droidjeet. Macs are great at it because the memory is fast. Let me know when you can run Deepseek 3.2 8bit at 17 t/s (that's two of those clustered together with RDMA because that model is 768GB in size)
>>109251751shiieeeetttt this makes my macbook pro m4 pro with 24GB RAM look like an infant.must feel good, no?
>>109251792It's pretty sweet. EXO is still a little buggy so I'm using it for much atm, but I plan to build some stuff out to make it much more useful in an agent sense. Also that model is overkill, I just loaded it for fun.Curious when Apple will drop the IChadUltraM4 chips
>>109251760isn't it like 800GB/s ?that'd be plenty for a moe.
>>109251754Nah, Mac inference speed is kinda shit, it just let's you run fuck off huge models(which are even slower). Would be kinda shit for models over 32b just due to speed. Maybe a 70b. 200b wpuld crawl.
