take it with a teaspoon of salt: Pascal up to 17 Billion Transistors, 32GB HBM2

[zMeul]

source: Fudzilla via PCPer

 

Pascal is the successor to the Maxwell Titan X GM200 and we have been tipped off by some reliable sources that it will have  more than a double the number of transisters. The huge increase comes from  Pascal's 16 nm FinFET process and its transistor size is close to two times smaller.

Pascal has 17 billion transistors and it will be significantly smaller silicon than the Maxwell 28nm based GM200.

Nvidia will use second generation HBM for its Pascal GPU to get to a 32GB on the highest end card, This is 2.7 times more than the already impressive 12GB used on Titan X. The second generation HBM or HBM 2.0 will enable 8Gb per DRAM die, 2Gbps speed per pin and 256 GB per second Bandwidth/ stack.

---

17bil transistors, that huge, it's little more than double the count for Titan X with it's 7,1 billion, with a B

any bets? who's gonna be the 1st to market with a HBM2 product: AMD or nVidia?

also, 32GB of VRAM? yeah .. that's not gonna be a desktop product, rather a Tesla accelerator or a Quadro

[KamiKatze]

Sounds like the specs for the ne Titan P (P for Pascal you know; gotta keep the Titan naming)

[Eugene Yap]

I'm talking with a silo of salt

[patrickjp93]

I think Nvidia will be our first but have the lower binned, slower chips, not that it will matter at these ludicrous bandwidth levels.

[syoumans54]

Well I'm officially no longer happy with my current setup.... Thanks Obama.

[thekeemo]

[FunkyFeatures]

who's gonna be the 1st to market with a HBM2 product: AMD or nVidia?

I believe recent news said AMD had some kind of first time exclusive on HBM2, meaning they would have to make a card before nVidia were allowed to.

[Husky]

Well there goes AMD... lol

[zMeul]

I believe recent news said AMD had some kind of first time exclusive on HBM2, meaning they would have to make a card before nVidia were allowed to.

that's a lot of BS since:

1. nVidia taped out 1st
2. nVidia announced Pascal with HBM2 - otherwise they wouldn't've bothered
3. AMD doesn't control HBM2, SK-Hynix does; and ultimately, not even since it's JDEC standard
4. it's a rumor with absolutely no confirmation; also coming from WCCF Tech who also said there's no problem with HBM supply .. duh!!

[ChineseChef]

I believe recent news said AMD had some kind of first time exclusive on HBM2, meaning they would have to make a card before nVidia were allowed to.

 

I was under the impression that was more about supply than product release.  But it can be assumed that if AMD gets the first X many units that they will likely be first to market with them.  Unless NVidia is just ready to plug and play, which is very unlikely.

[patrickjp93]

I believe recent news said AMD had some kind of first time exclusive on HBM2, meaning they would have to make a card before nVidia were allowed to.

Hynix could do nothing about it if Micron or Samsung or Elpida jumped in and started making it. It's a JEDEC standard after all, and Micron has the TSVs and experience, not to mention a will to snub Hynix after the stunt it pulled buying the votes to keep HMC out of the JEDEC standards. Intel stopped using Hynix's memory in its SSDs after that too. Hynix may come to regret that move more than its current CEO can imagine.

[patrickjp93]

I was under the impression that was more about supply than product release. But it can be assumed that if AMD gets the first X many units that they will likely be first to market with them. Unless NVidia is just ready to plug and play, which is very unlikely.

Why would it be unlikely? Nvidia taped out 2 months before AMD did.

[ChineseChef]

Why would it be unlikely? Nvidia taped out 2 months before AMD did.

 

I meant that it isn't like they can just wait for their turn at production, then just plug in the chips into ready made cards and start selling.  Assembly is slightly more complicated than that.

 

But, that doesn't mean that they won't have it ready in record time for that sweet sweet money.

[FunkyFeatures]

that's a lot of BS since:

• nVidia taped out 1st
• nVidia announced Pascal with HBM2 - otherwise they wouldn't've bothered
• AMD doesn't control HBM2, SK-Hynix does; and ultimately, not even since it's JDEC standard
• it's a rumor with absolutely no confirmation; also coming from WCCF Tech who also said there's no problem with HBM supply .. duh!!

"I believe". I didnt say they had 100% exclusivity to anything. Upon reseach, i found a link, saying that it is an unconfirmed rumor like you say

[Shaade]

Real or not,

 

[evilarceus]

inb4 128GB VRAM total in builds

[jos]

Going for 8k?

[deviant88]

Not real.

But if it would be real it means at least 2 things in my book:

1. 4K will get rekt by the Pascal.

2. AMD will get rekt by nvidia.

 

If AMD doesnt have some fat dongers for CPU and GPU next year they should start looking to sell its stuff or they will go bankrupt and shutdown.

[Notional]

 

that's a lot of BS since:

1. nVidia taped out 1st
2. nVidia announced Pascal with HBM2 - otherwise they wouldn't've bothered
3. AMD doesn't control HBM2, SK-Hynix does; and ultimately, not even since it's JDEC standard
4. it's a rumor with absolutely no confirmation; also coming from WCCF Tech who also said there's no problem with HBM supply .. duh!!

 

 

1. Pretty sure AMD taped out first on 14nm FF (which should include Greenland, but that is not confirmed though).
2. AMD announced Greenland with HBM2.
3. SK-Hynix controls all HBM2 production they themselves produce. And AMD got exclusive rights to HBM1 they helped develop, and as rumours go, also get dibs on HBM2 modules.
4. So is everything Pascal related.

[patrickjp93]

I meant that it isn't like they can just wait for their turn at production, then just plug in the chips into ready made cards and start selling. Assembly is slightly more complicated than that.

But, that doesn't mean that they won't have it ready in record time for that sweet sweet money.

Have the does ship to Nvidia and then when the HBM and interposer supplies come in ship it all to assembly. Done deal. If AMD doesn't have the product volume to begin assembly and move to market without causing supply lag and incurring extra expenses at stalled supply points, the Nvidia could very well move in first.

[patrickjp93]

• Pretty sure AMD taped out first on 14nm FF (which should include Greenland, but that is not confirmed though).
• AMD announced Greenland with HBM2.
• SK-Hynix controls all HBM2 production they themselves produce. And AMD got exclusive rights to HBM1 they helped develop, and as rumours go, also get dibs on HBM2 modules.
• So is everything Pascal related.

Pascal taped out in April. Greenland taped out late June/early July given GloFo's own risk production timelines.

[Jaybird]

so are chips still 2d? they seem very flat. I dont know why manufactures havent gone vertical like a skyscraper. 

[PudgyAstronaut]

Well there goes AMD... lol

 

Eh it's okay the Pascal Titan will cost $2500. I don't know how they would even begin to argue that 32GB of VRAM makes any sense at all for games. "Once you load a texture, it just stays there... ready... forever (until you restart)"

[patrickjp93]

so are chips still 2d? they seem very flat. I dont know why manufactures havent gone vertical like a skyscraper.

Because building everything into the die would destroy yields. Also, engineering a cooling system would be extremely difficult.

[themaniac]

am I the only one that care more about the lower end gpu's like the xx70 series