[Fudzilla] Quadro P6000 featuring GP100 fully enabled with GDDR5X EDIT: Maybe GP102

[n0x1ous]

http://www.fudzilla.com/news/graphics/41196-quadro-p6000-and-p5000-are-out

Interesting as I thought GP100 would be HBM2 only.

VRWorld is reporting its HBM2 only and guru3d had a link saying the same as fudzilla but its been taken down so unclear on the VRAM at this point.

 

[Stuka87]

My guess is that HBM2 has been delayed more. It certainly should not be a cost issue with those cards costing a fortune anyway.

 

[railven]

Does this mean a chip can essentially have either or memory or is this a variant of the GP100 shown with HBM2?

 

[nurturedhate]

Does this mean a chip can essentially have either or memory or is this a variant of the GP100 shown with HBM2?

I cannot see it having both as that would be a lot of wasted die area. We've seen it 100% with HBM2/1. Could be more likely this is GP102 w/ GDDR5x than GP100.

 

[Flapdrol1337]

I don't see this on the nvidia site.

This being a GP102 product seems much more likely.

 

[Arachnotronic]

It's GP102.

 

[Magee_MC]

Would it be possible to use either GDDR5X or HMB2 with an interposer? Could the interposer be manufactured to accept one or the other so that instead of changing the die, they would change which interposer is used to use different memory.

I have no idea if this is possible technically, or makes any sense, and hopefully someone who does know can weigh in.

 

[dark zero]

I can accept if is on the Titan fue supplies, but in a Flagship? Seriously?

Pascal and Polaris are turning into a massive disappointing generations since the latter is HW constrained and the 2nd has not all the DX 12 features they supposed to have.

Also HBM2 isn't supposed to be ready? Or Intel bought all of them?

 

[Tuna-Fish]

Would it be possible to use either GDDR5X or HMB2 with an interposer? Could the interposer be manufactured to accept one or the other so that instead of changing the die, they would change which interposer is used to use different memory.

I have no idea if this is possible technically, or makes any sense, and hopefully someone who does know can weigh in.

No. The components that need to change are PHYs on the GPU. HBM has lots and lots of tiny PHYs that are built to drive a very low amount of charge relatively slowly, GDDR5X has few large PHYs that are built to drive a huge amount of charge very rapidly. Some memory systems are similar enough on the physical level that you can make a PHY system that works with multiple different memory chips, but HBM and GDDR5X are so dissimilar that it's just not possible.

 

[Magee_MC]

No. The components that need to change are PHYs on the GPU. HBM has lots and lots of tiny PHYs that are built to drive a very low amount of charge relatively slowly, GDDR5X has few large PHYs that are built to drive a huge amount of charge very rapidly. Some memory systems are similar enough on the physical level that you can make a PHY system that works with multiple different memory chips, but HBM and GDDR5X are so dissimilar that it's just not possible.

Thank you for the answer!

So, if I understand this correctly, the chip would either have to be designed to use HBM or GDDR5 from the ground up, which takes up room that could go to the chip itself, or it has to be designed from the ground up for one or the other. If its designed for GDDR5 then they would have to revamp the chip itself to allow it to use HBM which would be an expensive process.

 

[n0x1ous]

GP102 does seem more likely. Its interesting that if its indeed GP102 then we already get confirmation that Titan XP isnt fully enabled. Pathetic if they have fully enabled dies and won't even put them in the $1200 flagship

 

[garagisti]

GP102 does seem more likely. Its interesting that if its indeed GP102 then we already get confirmation that Titan XP isnt fully enabled. Pathetic if they have fully enabled dies and won't even put them in the $1200 flagship

This has been known for about a week. There is probably going to be a refresh with the GP100 next year, and that would probably have hbm2, is what most of the rumours are saying.

Well, it is 780 to 780ti all over again. Sure performance would be better on the GP100 Ti chip compared to Ti chip on the GP102, but damn, only a year between two Titans, on same architecture.

 

[Tuna-Fish]

Thank you for the answer!

So, if I understand this correctly, the chip would either have to be designed to use HBM or GDDR5 from the ground up, which takes up room that could go to the chip itself, or it has to be designed from the ground up for one or the other. If its designed for GDDR5 then they would have to revamp the chip itself to allow it to use HBM which would be an expensive process.

Correct. Lowest-level masks would have to be different, which means not only millions, but time. ~3-4 months minimum from when the decision is made, and that's if everything goes exactly right.

I would not be surprised one bit if the main reason that GP102 exists is as a hedge against a possible HBM2 delay.

 

[tviceman]

GP102 does seem more likely. Its interesting that if its indeed GP102 then we already get confirmation that Titan XP isnt fully enabled. Pathetic if they have fully enabled dies and won't even put them in the $1200 flagship

Considering that the original Titan was cut down, and even more substantial that Quadro will sell for 5x the price, call it pathetic, but it is simply smart business.

 

[n0x1ous]

Considering that the original Titan was cut down, and even more substantial that Quadro will sell for 5x the price, call it pathetic, but it is simply smart business.

Of course it is, but it would just be nice for them to toss us a bone every now and then.