NVIDIA GP100 Silicon Moves to Testing Phase

[nvgpu]

http://www.techpowerup.com/216203/nvidia-gp100-silicon-moves-to-testing-phase.html

 

[Arachnotronic]

http://www.techpowerup.com/216203/nvidia-gp100-silicon-moves-to-testing-phase.html

Awesome; give me some of that 16nm FinFET GPU action!

 

[ShintaiDK]

Patience is a virtue 😉

 

[tviceman]

I don't see where there are any indications that this is GP100, and given that Nvidia has led with mid-range and small die processors with Kepler and Maxwell, I'm inclined to believe this is not big-die Pascal until shown more information to persuade me otherwise.

 

[Ajay]

I don't see where there are any indications that this is GP100, and given that Nvidia has led with mid-range and small die processors with Kepler and Maxwell, I'm inclined to believe this is not big-die Pascal until shown more information to persuade me otherwise.

I tend to agree - though maybe HBM2 only makes sense on high-end GFX cards (esp. if NV wants to start w/8GB). Interesting times.

 

[tviceman]

I tend to agree - though maybe HBM2 only makes sense on high-end GFX cards (esp. if NV wants to start w/8GB). Interesting times.

Supposedly there is more efficient and faster GDDR5, dubbed GDDR5X, inbound (though not HBM efficient). If that is case, then mid-range and low end Pascal can definitely go without HBM. But I'm not sure of the timing, as the article states "2016" which doesn't sound like early 2016.

http://www.tomshardware.com/news/micron-launches-8gb-gddr5-modules,29974.html

 

[alcoholbob]

So insured value of GP100 die is ~$3,125. In comparison GM200 insured value was $335 for the die back in July 2014.

If this launches in 2016 it's going to Tesla only. If by some small miracle this launches in Geforce it will most likely 1) Months after the Tesla launch 2) Be in a $1,000 Titan only for several months 3) The Titan will be an cut-down version of the chip.

 

[jpiniero]

If this launches in 2016 it's going to Tesla only. If by some small miracle this launches in Geforce it will most likely 1) Months after the Tesla launch 2) Be in a $1,000 Titan only for several months 3) The Titan will be an cut-down version of the chip.

nVidia badly needs to refresh Tesla, so that's the main goal but there will probably be a Titan too. I'm sure it has GDDR5, but considering how small the die will be HBM1 is theoretically possible.

 

[x3sphere]

I wouldn't be surprised if they try to push an even higher price with the Titan this time. May have not worked in the past (Titan Z), but the market dynamics seem right for it this time... dieshrink, huge VRAM increase and likely at least double the performance. They can probably get away with $1.5K.

 

[Ajay]

nVidia badly needs to refresh Tesla, so that's the main goal but there will probably be a Titan too. I'm sure it has GDDR5, but considering how small the die will be HBM1 is theoretically possible.

The Top of Line 16FF based compute card will probably need HBM2 to compete with Intel's 14nm Xeon Phi with HMC.

 

[alcoholbob]

I wouldn't be surprised if they try to push an even higher price with the Titan this time. May have not worked in the past (Titan Z), but the market dynamics seem right for it this time... dieshrink, huge VRAM increase and likely at least double the performance. They can probably get away with $1.5K.

They can get away with it if they launch before AMD on 16FF, definitely.

If they can make a card as fast or nearly as fast as 980Ti SLI they can charge $1500 for the Single Card crown since two 980 Tis is $1350.

But it will basically be a price drop as soon as AMD launches again, repeat of Titan X -> 980Ti I'm sure.