Thread 106258380

>5060 ti ai performance but with 48gb for $1200
is it worth it?

ai is only optimized for Nvidia. the extra VRAM is wasted because of that and we sadly won't ever see any real competition that forces Nvidia to give us their best shit that they reserve for $15,000 per unit enterprises. we need an Ryzen moment like with the CPU market that dethroned Intel, but its unlikely to ever happen.

I'm not sure that is a US price. This is the listed distributor for maxsun in the US, and the price listed here is $3k. (I don't think it is worth $3k).

https://www.hydratechbuilds.com/product-page/intel-arc-pro-b60-dual-48g-turbo

>>106258597
>$2,999.00

Bah! Taking a page right out of ngreedias book I see. Aint no way this thing is any more faster than a 4070. If it were the speed of the 4090 with the 48gb, $1200 would be reasonable.

>>106258380 (OP)
>designed, engineered, assembled in the usa
>that'll be 4.999,99 USD plus tip

>>106258380 (OP)
is that a intel card? Does intel have a cuda alternative that works and isn't a pain in the ass to setup like with amd?

Waste of sand. All that GPU will be any good for is running llama.cpp, and that's a stretch. It'll be useless for image or video gen.

If you have $3K but a 4090D 48GB, at least you can actually make use of your 48GB with that.

>>106258742
never played with intel gpus but comfyui ui seems to work on window

>>106258742
https://www.intel.com/content/www/us/en/developer/tools/oneapi/base-toolkit-download.html

>>106258522
AI isn't optimized at all

>>106258827
neither are my bowel movements.

>>106258797
and thats supported by pytorch and co. ?

>>106258933
>pytorch
That I'm not sure about. I just happen to be at the beginning of this particular rabbit hole.

https://pytorch.org/blog/intel-gpu-support-pytorch-2-5/

>>106258380 (OP)
can you put two of them together and have a 96GB system?

>>106258756
gpus are good for more than AI slop. 8 or ten of those in an air cooled gpu chassis is a lot of compute.

>>106258933
dpc++/sycl is low level gpu programming using high level c++. it is pretty slick just for what it is.

Intel's one api is built on that, and I'm pretty sure the pytorch interface would be built on top of that.

>>106258380 (OP)
How good is the Blender rendering performance and is it good for multi-GPU setups?

>>106258597
>$2,999.00
>GDDR6
Nothing newer for all that money, huh?

>>106260057
My guess is they are still trying to figure out how to price it. Intel does not seem to know how to behave when they are not a market leader. They need to price this very competitively until the market has any kind of confidence in them. I like the dpc++ stuff, and have a lot of code developed around it, but I wouldn't pay $3k for this. $1.2k? maybe/probably.

>>106260057
>>106260256
Intel isn't pricing this.
This is Maxsun's creation and given that it is somewhat novel, it's hard to price.
You want as much money as possible but who wants it is wholly dependent on pricing.

>>106258380 (OP)
>$1200 for intel arc with absolutely shotty support via ipex-llm (that is almost always months behind and can't even support the latest models)

>>106258742
Yes, but software has to have Intel cards support. Some popular has, but for the rest you have to modify a lot of code by yourself.