https://devblogs.nvidia.com/parallelforall/inside-pascal/
IT'S HAPPENING
Heh, guess /g/ was right, it's Maxwell 3.0 with a HBM memory controller.
>>53883411
Honestly the TFLOP specs are underwhelming, the only thing mighty impressive is the huge die and 300W TDP
>>53883443
Oh yeah, and the astronomical clockspeeds.
No doubt a major contributor to said TDP.
Nice paper laucnh
Maybe this gen I'll upgrade my 9600GT (512MB).
>tfw I've been considering it for pretty much every gen since 4xx (SOMEBODY CALL 911, SEE THAT FERMI BURNIN' ON THE DANCEFLOOR WOHOO)
>always decided against it either because of reasons or memes
>>53883508
Hahaha - rockin the same 9600GT for 8 years now. Looking forward to pascal and polaris. So i can get a MUCH cheaper 390
>Another HBM2 benefit is native support for error correcting code (ECC) funtionality, which provides higher reliability for technical computing applications that are sensitive to data corruption, such as in large-scale clusters and supercomputers, where GPUs process large datasets with long application run times.
>ECC technology detects and corrects single-bit soft errors before they affect the system. In comparison, GDDR5 does not provide internal ECC protection of the contents of memory and is limited to error detection of the GDDR5 bus only: Errors in the memory controller or the DRAM itself are not detected.
Will this have any benefit for average consumers or mainly just for workstation applications?
>>53883720
I don't really see how it'll affect consumers.
>single-bit soft errors
Mostly because I have no idea about what this is.
So my guess is that the average consumer won't notice any difference.
>>53883720
>Will this have any benefit for average consumers or mainly just for workstation applications?
Only workstations, since average consumrs don't buy $1500 Titans.
>>53883720
>>53883787
>>53883793
basically memory having a spontaneous bit flip.
>>53882765
So why do we need to compute things on GRAPHICS processing units when we have CPUs doing it thousands times faster, again?
>>53884045
>So why do we need to compute things on CPU processing units when we have GRAPHICS doing it thousands times faster, again?
fify
The gaming vesion will be the same?
>>53884174
>nearly twice the amount of transistors compared to Maxwell
And AMD is using an even smaller die size. Non-waitfags will get destroyed when the first benchmarks come out.
>>53884781
Except in DX12 where Polaris will wipe the floor of Pascal since Pascal still doesn't have async compute
>>53884831
It's okay Nvidia won't need async because they card won't be idling 50% of the time.
>>53884943
That's basically what Nvidia cards do. Anyway the lack of shader utilization is largely dependent on optimization from devs.
How will Pascal affect 1080P gaming?
>>53885646
Since they didn't launch any desktop cards or anything, we still don't know
>>53883720
>>53883787
Roughly speaking, you have a chance of one bit randomly flipping per GB RAM per month of operation, whether it's system or GPU RAM.
For most people it doesn't matter since the error is so small it'll be either handled by the software, or the work that you're doing isn't critical of errors, or even important.
But if you're doing say molecular simulation and a flipped bit alters the vector of even one particle, it could ruin the whole simulation. Even worse, the simulation might finish, but you'll be given the wrong results with no warning of any errors.
It's always better to have ECC than not have it, primarily as system RAM, but in a GPU it matters less unless you are specifically using yours for scientific compute.
>>53885755
Thank you for a real answer
So is the consumer stuff on ice until 2017 as well or just the Tesla thing?
>>53886471
Nvidia can survive not having a consumer launch in 2016
AMD however needs to launch both high end cpus and gpus this year that impress or they'll legitimately go bankrupt
>>53884831
Games simply won't use asynchronous compute heavily until Nvidia say it's okay. Developers have already said this very clearly. The only games to use it will be ones AMD pays for partnerships with.
>>53886879
Hitman already does, and AMD shits on Nvidia in that game
