what do you think?
If they talk about the tradtional stacked memory , it is great technology that helps the GPU performance by a lot. The first generation (AMD Fiji) allowed up to 4GB memory on GPU, but the second one (Quadro GP100, probably Volta and Vega later) have 16GB (and maybe more in the future).
I’ve tested GP100, and it is by far the fastest GPU I’ve seen (about 40%-50% faster than Quadro P6000).
Best,
Blago.
The GP100 sounds like a beast.
Blago, with the news that Vega will be coming out within the week, will we see any benefits of this card paired with RT? It looks to be in direct competition with Nvidia’s deep learning cards, and touting that it is a good card for pro users. Having HBM2 with 16 GB of ram sounds like it could be a good piece of hardware. Is this going to be a card we could look forward to using with Vray? It would be great to know. Thanks.
Best,
Jeremiah
I can’t tell until we have tested it. So far we haven’t had the chance to get one.
Best,
Blago.
since you gus have so cleverly written a cuda driver for cpu in-house, any possibility you could do a cuda driver for amd cards too? is it even possible?
So far OpenCL is the best way of running raytracing on AMD GPU.
Best,
Blago.
i appreciate that, i was just thinking if you could get CUDA running youd only have a single target to code for and we could mix and match cpu’s gpus and not worry about compatibility. anyway it was more of a question of interest, wether it would be technically possible for a third party, such as yourselves to get cuda running on AMD cards. i know AMD could do it if they wanted, but probably never will due to politics.
Well I hope too hear what happens when you do get one. The price point of these cards are said to be pretty decent for price vs. performance.
Cheers.