Will Volta use HBM2?
>>58744850
HMC by Cuckron.
>>58744850
probably, in the highest end model at least.
384b/512b GDDR5X/6 would still work, but it would make bigger cards quite toasty.
>>58744850
The fuck is Volta at this point even. I remember awhile ago Nvidia was saying it'd have Hybrid Memory Cubes, an ARM processor, and some other junk I forgot. HMC is virtually dead, and Nvidia's arm cores are trash. Then Pascal popped up in the roadmap. GP100 was delayed so much (only limited samples out currently), that I really doubt the meme of it coming out this year will happen. Especially if the TSMC 12nm meme is true, that's not scheduled to be in volume till well into 2018. Even then it still uses the 20nm BEOL, so die scaling won't increase.
Honestly, the fuck is it even meant to be. Vega 20 is at least clear in purpose, a "7nm" die shrink (14nm back end), of Vega 10, with FP64.
>>58745648
The fuck
Nvidia has a high performing ARM core
t. amdfag
>>58744850
Most unlikely
Both from volta and vega
>>58745648
From what I've read, Volta is a Pascal refresh but with HBM.
>>58745977
oh fuck that would be funny as hell, pascal was literally maxwell but on a smaller node, its the 8800 all over again
Yes.
The question is: will it be available for consumers this year? Rumors say it's a late 2017 launch for supercomputers. No rumor points to a consumer launch this year.
>>58745843
Both are confirmed HBM2.
Why would you need it? 4k is a meme, 1440p covered with 1070/1080, 7nm is not gonna happen in the next 5 years and there is 50% chance that it's not gonna happen AT ALL. Silicon is a kill, buy now 1070 and you will be set for the next 10 years if not more.
Volta confirmed to use HBM2 long ago
Consumer Volta will not use HBM2, whats the point, just look at this blunder of the century
https://videocardz.com/65649/sk-hynix-updates-memory-product-catalog-hbm2-available-in-q1-2017
>See even Hynix plans can change quite often. The current plan is to release 1.6 GHz HBM2 stack with 4GB memory, and since we already know Vega has two stacks, this gives us 8GB and 409,6 GB/s bandwidth. That’s obviously lower than Fiji, in fact, it’s even lower than Pascal GP102 (480 GB/s)
You get more memory capacity & bandwidth from GDDR5X and upcoming GDDR6 than HBM2
HBM2 was a mistake and AMD is shit