Nvidia GTC16 Webcast Live Stream@5pm/9am PDT *Jen looks great in leather.*

[KompuKare]

Details specs of P100 on Nvidia website:

https://devblogs.nvidia.com/parallelforall/inside-pascal/

300W TDP!!!!!!!!!!!!!!!!!!!!!!!!!!!!

That's pretty high for Nvidia. With the clocks they gave (1328 MHz, Boost: 1480 MHz) might possible be outside the comfort zone for the card to be 300W.

What is the DP performance of Intel's Knight Landing which they have to match/beat? Because otherwise I would have thought they would have keept the clocks down a bit and got it to fit into 250W.

Still, if 1.3GHz+ is possible on this max size chip, maybe this bodes well for the smaller Nvidia ones? (Was going to say bodes well for Polaris as well but that's not on TSMC's 16nm so still an unknown.)

 

[Orangey]

Tesla P100 accelerators have four 4-die HBM2 stacks, for a total of 16 GB of memory, and 720 GB/s peak bandwidth

The combination of HBM2 stacks, GPU die, and silicon interposer are packaged in a single 55mm x 55mm BGA package

 

[DiscoDave]

So how does this stack up against amd's offering?

Had the itch to upgrade my LGA 1366/7970 setup for ages now.. I just end up waiting more and more.

 

[AllBodies]

Ouch, so they've only managed 1.884x increase in Transistor density.

Samsung/GloFo LPP is 'up to' 2.3x according to Samsung.

 

[Rroff]

Still, if 1.3GHz+ is possible on this max size chip, maybe this bodes well for the smaller Nvidia ones? (Was going to say bodes well for Polaris as well but that's not on TSMC's 16nm so still an unknown.)

Hence my post before which DM seems to have missed the point of heh.

 

[GFX-Kid]

16 GB of HBM2

 

[leon4112]

Great hes talking about Knight-rider cars now whos the first car that has a red led in the front!!!

 

[CAT-THE-FIFTH]

Thats for the DGX1, a server super computer.

The whole conference is basically about AI and Deep learning, whatever that is.

It appears that is the price for one with two Xeons and a 8 P100 based graphics cards - still makes them quite expensive??

$10000 each??

Ouch, so they've only managed 1.884x increase in Transistor density.

Samsung/GloFo LPP is 'up to' 2.3x according to Samsung.

It could be since it has such a high TDP,they have spaced the internal components apart,to prevent heat spots?? Bulldozer chips the like the FX8150 kind of did that. It might explain the largish die.

 

[mcalladine]

lol all the people wake up and take pictures when pascal is shown

 

[stulid]

woohoo dave net confirmed.

 

[AllBodies]

It could be since it has such a high TDP,they have spaced the internal components apart,to prevent heat spots?? Bulldozer chips the like the FX8150 kind of did that. It might explain the largish die.

I suppose that's possible, but also it points to not so great performance per watt increase.

Since it's only 87.5% more transistors (than Titan X) AND it has HBM2 (less power than GDDR5), but it has 50W more TDP :/

 

[madmonkey]

Walks into thread looking for news on Pascal

Walks out of thread to look at 980ti prices

 

[pmc25]

Ouch, so they've only managed 1.884x increase in Transistor density.

Samsung/GloFo LPP is 'up to' 2.3x according to Samsung.

I find it extraordinary that they've gone for 600mm2.

No doubt AMD beat them to market, and likely with a much denser (and therefore cheaper) chip in Vega.

I suspect a perfect storm of having to add back in all the hardware they took out, new memory controller, all the NVLink stuff (which will still be present but inactive in GP100 Titan / 1080Ti), having less than 2 years of design time on Pascal and new process have seriously gimped it.

How the hell are the Titan / 1080Ti going to be competitive at their likely price and probably inactive core count (due to insane defectivity)?

It's possible NVIDIA aren't too far behind on the GP104, depending on when AMD ultimately end up launching, but P100 looks like Fermi all over again ... or worse.

 

[CAT-THE-FIFTH]

I suppose that's possible, but also it points to not so great performance per watt increase.

Since it's only 87.5% more transistors (than Titan X) AND it has HBM2 (less power than GDDR5), but it has 50W more TDP :/

Technically we should be comparing it to the Kepler based GK110 and GK210 cards and not the GM200 since Maxwell had most of DP compute stripped out.

 

[stulid]

woohoo that's it confirmed

 

[markoboyo11]

what a load of ****

 

[Smrtka]

what the hell ?

 

[mcalladine]

woohoo that's it confirmed

roborace ?

 

[deadfolk]

What are they odds of them announcing anything during the rest of the conf? Should I just order my 980ti now?

 

[easyrider]

Roborace...whats the point!