Nvidia Pascal GP100 Die Shot Photo

rl66

2016-08-25 10:26

There is 2 tesla version of pascal, one cut down and one full, the info are on the dedicated website (the cut down have less hbm2 memory too btw).

#5326110

TheDeeGee

2016-08-25 10:27

Dat TDP tho, the same as my entire system :P

#5326117

icedman

2016-08-25 10:51

I'm confused is this supposed to replace the current titan, is it the 1080ti, or both and anyone that already bought a titan has been screwed just like very first titan lol

#5326133

Xionor

2016-08-25 11:45

I'm confused is this supposed to replace the current titan, is it the 1080ti, or both and anyone that already bought a titan has been screwed just like very first titan lol

The only people that actually need a Titan, and have a Titan, have more than enough money for another Titan.

#5326163

DeskStar

2016-08-25 13:14

The only people that actually need a Titan, and have a Titan, have more than enough money for another Titan.

Here......HERE..!!!!

#5326164

Denial

2016-08-25 13:22

I'm confused is this supposed to replace the current titan, is it the 1080ti, or both and anyone that already bought a titan has been screwed just like very first titan lol

GP100 is currently being sold for high performance computing. I don't think it will ever take the form of a Titan personally, or if it does it won't be until next year.

#5326167

Loophole35

2016-08-25 13:24

So when does this GP100 see production? Will it be before Vega? I just hope we see a fully enabled GP102 in GTX form. Did notice that the only difference between GP100 and GP102 is the memory controller.

#5326175

Denial

2016-08-25 13:39

So when does this GP100 see production? Will it be before Vega? I just hope we see a fully enabled GP102 in GTX form. Did notice that the only difference between GP100 and GP102 is the memory controller.

GP100 is already in production, they shipped the first pre-ordered DGX-1 about a week ago to OpenAI, Elon Musk's AI company. They were supposedly shipping engineering sample units to universities prior to that too. GP100 has a number of differences compared to GP102, it uses a completely new FP16v2 CUDA core that's capable of outputting two FP16 instructions per cycle as long as the two operations are the same. GP104/102 has this but they only have one per SM. It also has 8x the FP64 cores per SM. It split the SM too, halving the number of cores/blocks/shared memory but it doubled the number of SMs. GP100 is basically entirely geared towards high performance computing. It actually wouldn't surprise me if Nvidia at some point splits their architecture design, one for HPC and one for everything else.

#5326236

Solfaur

2016-08-25 15:32

I don't think we will see this beast anytime soon in a "consumer" variant. Besides right now nvidia has all segments covered, only low-end need that 1050 and then they're good. After AMD brings out their stuff later this year/early next, well, things will get interesting again (hopefully).

#5326244

rl66

2016-08-25 15:56

I'm confused is this supposed to replace the current titan, is it the 1080ti, or both and anyone that already bought a titan has been screwed just like very first titan lol

from what i have understand it will be for new tesla or computing and for high end quadro (cool 🙂 )... As original Titan were derivated from Quadro: it might be for another Titan too We will see when it will be in shop 🙂

#5326246

rl66

2016-08-25 15:58

It actually wouldn't surprise me if Nvidia at some point splits their architecture design, one for HPC and one for everything else.

as it was in the past 🙂

#5326326

PrMinisterGR

2016-08-25 19:21

GP100 is basically entirely geared towards high performance computing. It actually wouldn't surprise me if Nvidia at some point splits their architecture design, one for HPC and one for everything else.

That's an interesting idea. Since they don't have a whole CPU line to keep alive too, their second arch could be something like that I guess. That would mean split driver teams too.

#5326353