>>110297057>From the Anthropic Road Show video:'This has also made us really good at bringing up new chips. As an example, this year, we started working with AMD GPUs for the first time. In the past, bringing on a new chip platform took dozens of engineers months of work. This time we had Mythos try it out. It got Claude running on AMD in just 3 days with basically no human intervention. Claude ported or wrote almost 100% of the kernels.'
>>110297057>with traditional approachthey mean humans?
eh, everything is priced based on the HBM manufacturing bottleneck these days, AMD chips won't get you any crazy savings
>>110297073>It got Claude running on AMDso inference only, not training
>>110297093no, following copyright laws.amd had to give up their human-made cuda clone due to copyright laws. now ai can just launder any code.
>goythos
>>110297057who would anyone give a shit about CUDAesque on cunsumer gank like AMD GPUs? That ship is sailed its SXM and HBM or go home.
>>110297057AMD is so shit they need to bruteforce software via LLMs now. Why can't they just fix their shit?
>>110298133practicaly the same thing, inference is a tiny bit more code but it mostly uses the same math.backprop isn't a lot more code than just mmul
>>110298204no one writes software by hand professionaly anymore.
>>110297057Is it as fast as Nvidia?
>>110298208the issue with AMD GPUs is not individual kernels which are easy to translate, but the libraries
>>110298376doesn't realy matter, ai inference or training isn't that complex.that's why the chineses are not even using cuda and writting raw PTX to squeeze out performance, now it can just be done faster and with less effort.
