what do you faggits know about Nvidia's Volta?
Pascal is basically Maxwell v2, but Volta is supposed to be a new architecture?
>>51783682
Why the fuck do you think we know anything?
And how the fuck are you so sure Pascal is just a Maxwell die shrink?
>>51783740
it's not a die shrink, but a revised Maxwell design
Volta use TSMC's 16nm FinFET tech
>>51783682
>Pascal is basically Maxwell v2
Confirmed for not knowing shit. Kill yourself
>>51784538
Except he's right.
I'm an AMDfag but you're retarded, Pascal needs an entirely new memory interface, which also means much smaller caches compared to the huge ones Maxwell has, it also needs to stick those FP64 units Nvidia removed from Maxwell.
That is in no way shape or form "Maxwell v2"
>>51784754
>>51784538
Some hardcore grasping at straws ITT
AMDFAGS ON FULL DAMAGE CONTROL
http://www.geeks3d.com/forums/index.php?topic=4247.0
Pascal
-D__CUDA_ARCH__=600 -D__CUDA_ARCH__=610
-D__CUDA_ARCH__=620
Pascal is not Maxwell
AMDFAGS ON SUICIDE WATCH
>>51787539
>>51786194
>dx12
>vulcan
> Volta was originally intended to succeed Nvidia’s 900 series Maxwell GPU architecture in 2016. It was originally going to be the company’s first generation to feature stacked memory. However Volta was designed with HMC , the Hybrid Memory Cube, in mind. Unfortunately however HMC hadn’t matured as quickly as Nvidia had hoped. So a replacement was put in place that makes use of the other major stacked memory standard, High Bandwidth Memory or HBM for short. And thus Pascal was born.
you fags don't know shit. pascal is a stop-gap solution, because volta had to be pushed back. and the most effective way to fill the gap was to modify maxwell.
> hum memory interface
>>51788945
So pretty much buy a 970 now and wait for 3 years?
>>51789321
nah, pascal should be a pretty good chip with very high bandwidth memory. so if you have an extra $500 to blow on gaming hardware, i'd get it. too bad nobody knows anything about volta.
>>51783814
Must be pretty fucking revised if they went from awful FP64 and zero FP16 support to full support for both.
Wait, a company revised something that does the same job as said previous version and gave it a new name? Holy marketing batman. We've got a new guy.
>>51791055
Gosh, it's almost like Nvidia intentionally ripped those things out of Maxwell and could easily add them back to claim them as new features!
>>51791149
Pascal is literally the first GPU ever with native FP16 you pleb.
>>51791149
Desktop GPU at least, mobile GPUs have had sub-32bit floats for a while.
>>51791160
In what way does that contradict what I said, you dumb little kuck?
http://www.3dcenter.org/news/reihenweise-pascal-und-volta-codenamen-aufgetaucht-gp100-gp102-gp104-gp106-gp107-gp10b-gv100
GP100
GP102
GP104
GP106
GP107
GP10B
THANK YOU BASED NVIDIA
>>51791160
can someone explain the use of FP16? my understanding is that most of the heavy lifting in games is done in FP32, but i may be wrong.
>>51791628
That's true currently because there's no benefit to using smaller floats (except saving a tiny bit of memory bandwidth when storing them, who cares). Colors, normals and vectors could be expressed as FP16 without losing any fidelity and Pascall can crunch them faster than FP32s.
Most of the hype over FP16 is in the GPGPU deep learning bandwagon though.
>>51783682
From what I know, Pascal is a new architecture, and Volta is Pascal but greatly improved.
>>51788945
>Nvidia in charge of new memory standards
HBM won over HMC for the JEDEC (iirc thats the group) standard.
