You're viewing a single thread.
Why use commercial graphics accelerators to run a highly limited "AI"-unique work set? There are specific cards made to accelerate machine learning things that are highly potent with far less power draw than 3090's.
Well yeah, but 10x the price....
Not if it's for inference only. What do you think the "AI accelerators" they're putting in phones now are? Do you think they'd be as expensive or power hungry as an entire 3090 for performance if they were putting them in small devices?
Ok,
Show me a PCE-E board that can do inference calculations as fast as a 3090 but is less expensive than a 3090.
I'd be interested (and surprised) too
Yeah show me a phone with 48GB RAM. It’s a big factor to consider. Actually, some people are recommending a Mac Studio cause you can get it with 128GB RAM and more and it’s shared with the AI/GPU accelerator. Very energy efficient, but sucks as soon as you want to do literally anything other than inference
I wouldn’t say it particularly sucks. It could be used as a powerhouse hosting server. Docker makes it very easy to do no matter the os now a days. Really though I’d say its competition is more along the lines of ampere systems in terms of power to performance. It even beats amperes 128 core arm cpu at a power to performance ratio which is extremely impressive in the server/enterprise world. Not to say you’re gonna see them in data centers because price to performance is a thing as well. I just feel like it fits right into the niche it was designed for.
How could you solve the problem of storage expansion? I assume there exists some kind of thunderbolt jbod thing or similar
Because those specific cards are fuckloads more expensive.
What are you recommending, I'd be interested in something that's similar in price to 3090.
It's for inference, not training.
Even better, because those are cheap as hell compared to 3090s.
But can they run Crysis ?
Would you link one? Because the only things I know of are the small coral accelerators that aren't really comparable, and specialised data centre stuff you need to request quotes for to even get a price, from companies that probably aren't much interested in selling one direct to customer.
Huh?
Stuff like llama.cpp really wants a GPU, a 3090 is a great place to start.