People are indeed using cerebras in relatively large volumes but usually those are big corporate clients that quietly do their work compared to most used FOSS project. Cerebras isn't still openly available for individual users whereas nvidia has already created a massive distribution from last 10-15 years and that's why it may feel that way. Also, lots of HPC libraries and framework have been well adopted for nvidia cards whereas in Cerebras if you want to do a scientific simulation then you'll need to modify your codebase as per the Cerebras SDK which is again bit little tedious. Cerebras is indeed more cost and energy efficient at large scale compared to nvidia in ml training.
Why isn't there much demand for WSE-2 then?
People are indeed using cerebras in relatively large volumes but usually those are big corporate clients that quietly do their work compared to most used FOSS project. Cerebras isn't still openly available for individual users whereas nvidia has already created a massive distribution from last 10-15 years and that's why it may feel that way. Also, lots of HPC libraries and framework have been well adopted for nvidia cards whereas in Cerebras if you want to do a scientific simulation then you'll need to modify your codebase as per the Cerebras SDK which is again bit little tedious. Cerebras is indeed more cost and energy efficient at large scale compared to nvidia in ml training.
@microbe_rz31 I think it's more of trying to hypothesis rather than overthinking which does include in it in some form.
Hoping to collaborate with your company but hoped it could go in my product alongside lord willing our data center
Bravo...why nobody talking about this miracle