If Vega is in 5-10% range of the Titan, then it won't be priced under $700 at all.
I think
I know AMD isn't charity but it being priced $500 would be killer, would literally hoborape the entire Nvidia overpriced lineup.
Then again if the die is small they could get away with it.
>>58275898
And NVIDIA would still outsell them because they have superior mindshare.
>>58276555
That's a different bag of worms, and to stop such lunacy it would require for /v/'s kind to grow a brain.
A most difficult undertaking.
>>58275898
It won't be, there is no insentive to do that as Nvidia has a position where they wpuld just counter the price point, with a half year lead production will be smoth and the bigest chunk of R&D will be covered next to AMD with a new chip and 100% of R&D costs outstanding. This is why duopolies don't work any better than monopolies.
>>58275863
Nvidia already has a 16gb hbm2 card out?
>>58277088
Haha
>>58277088
I think they have a Tesla, wasn't GP100 a HBM chip?
>>58275863
I thought the move to 16gb hbm meant ~2 teraflops bandwidth?
>>58277131
There's no point to so much bandwidth when the GPU can't hope to use it.
>>58277178
Still, thought it was unavoidable for some reason related to number of channels or something. Guess not.
>>58277131
No, 1TB/s was the originally advertised bandwidth for Gen 2. It seems both Nvidia and AMD have realised 1TB/s was overkill with their current micro-architectures, so have opted for lower bandwidth solutions. GP100 has the memory downlocked, to reach 720GB/s. Whereas Vega will use a 2 stack 2048bit bus instead of a four stack 4096bit bus design, achieving 512MB/s.
I think AMD has the right idea though, using only two stacks means being able to have a smaller and less complex interposer, which should help reduce costs and improve margins.