more Pascal leaks - GTX1080Ti and next Titan

[zMeul]

source: https://www.chiphell.com/thread-1588175-1-1.html

via: http://www.overclock.net/t/1600793/chiphell-gtx-titan-for-pascal-gtx-1080ti-and-gtx-1060-specifications-leaked

 

 

GTX 1080Ti from AIBs: 799 USD

GTX 1080Ti Founders Edition: 849 USD

GTX Titan (Founders Edition only): 999 USD

 

---

 

if the above info is true and the new Titan will use GDDR5X and no FP64 units, that could mean nVidia has two production lines for Pascal GPUs: one for P series Tesla accelerators and one for GTX series of video cards

to note, Titan is supposedly based on a GP102 chip and not GP100 like the Tesla P100 - even the P100 does not use the full GP100 GPU

 

nVidia will milk the shit out of Pascal

 

this could be nVidia's response to the "news" AMD bumped Vega launch to this year

Edited May 21, 2016 by zMeul

[aisle9]

Still cheaper than a Titan X. Jesus, I'm glad I didn't drop $1,200 on one of those.

[agent_x007]

Titan without Double Precision unlocked = Failed info.
I also doubt that 24GB of VRAM running 12-10GHz with 384-bit bus.

[SteveGrabowski0]

That would be truly fucking horrible to see a $1000 card that's not even the big die.

[Aytex]

aint no breaks on the hype train

[awesomeness10120]

Just now, SteveGrabowski0 said:

That would be truly fucking horrible to see a $1000 card that's not even the big die.

The reason this isn't the big die is due to HBM2 and yields. Tesla P100 has more GP100 priority than consumer products. That's why NVIDIA is making an intermediate chip for Pascal.

[SteveGrabowski0]

1 minute ago, awesomeness10120 said:

The reason this isn't the big die is due to HBM2 and yields. Tesla P100 has more GP100 priority than consumer products. That's why NVIDIA is making an intermediate chip for Pascal.

And pricing it like the big die. Nvidia's price inflation was bad enough with the 60 Ti series being promoted to the 80 series and doubled in price with Kepler, but now the Titan isn't even going to be the high end chip? What a joke. 

[Glenwing]

1 minute ago, awesomeness10120 said:

The reason this isn't the big die is due to HBM2 and yields. Tesla P100 has more GP100 priority than consumer products. That's why NVIDIA is making an intermediate chip for Pascal.

If NVIDIA is using a lesser chip to improve yields (reducing cost) then it should no longer cost $1000.

[Godlygamer23]

3 minutes ago, Glenwing said:

If NVIDIA is using a lesser chip to improve yields (reducing cost) then it should no longer cost $1000.

Maybe not, but this is NVIDIA we're talking about. The company that pulled off the $3,000 marketing stunt that is the Titan Z. 

[NateGSR117]

I want GP100 with HBM if they want my hard earn money. I will wait until than or if Vega is better I'll go to AMD for the first time. 

[Glenwing]

Just now, Godlygamer23 said:

Maybe not, but this is NVIDIA we're talking about. The company that pulled off the $3,000 marketing stunt that is the Titan Z. 

I know they have no problem doing it, I'm just saying no one should be defending it.

[patrickjp93]

4 minutes ago, SteveGrabowski0 said:

And pricing it like the big die. Nvidia's price inflation was bad enough with the 60 Ti series being promoted to the 80 series and doubled in price with Kepler, but now the Titan isn't even going to be the high end chip? What a joke. 

It is the high end chip for this generation. 16nm/14nm will be around for Volta, Vega, and likely Navi as well. Dies will get larger later with aid from HDL. An honestly, if Nvidia wants to cut out the 64-bit units from non-pro cards, it can.

[Castdeath97]

2 minutes ago, Godlygamer23 said:

Maybe not, but this is NVIDIA we're talking about. The company that pulled off the $3,000 marketing stunt that is the Titan Z. 

I even wonder if that thing every sold. 

[thexzenon]

So are we even going to see a GP100 consumer card? I for one can't wait for the 1080 TiTi 

1 minute ago, Glenwing said:

If NVIDIA is using a lesser chip to improve yields (reducing cost) then it should no longer cost $1000.

Depends on how much ass Vega will kick I guess. If the difference won't be that huge, they can totally push for $1000 as they'll claim it's the new fastest single-GPU card out there.

[awesomeness10120]

3 minutes ago, SteveGrabowski0 said:

And pricing it like the big die. Nvidia's price inflation was bad enough with the 60 Ti series being promoted to the 80 series and doubled in price with Kepler, but now the Titan isn't even going to be the high end chip? What a joke. 

It's either NVIDIA being overconfident and thinking AMD is not a threat and price gouging the hell out of their GPUs, or the cost the manufacture has skyrocketed to the point where they need to up prices. I wanna see a new 4870. 

[patrickjp93]

5 minutes ago, Glenwing said:

If NVIDIA is using a lesser chip to improve yields (reducing cost) then it should no longer cost $1000.

Let the market decide that. Nvidia will either lose sales or do equally well.

[Thiback]

1 minute ago, Castdeath97 said:

I even wonder if that thing every sold. 

Not many units were sold, but still, some people bought it (probably to show off).

[Godlygamer23]

1 minute ago, NateGSR117 said:

I want GP100 with HBM if they want my hard earn money. I will wait until than or if Vega is better I'll go to AMD for the first time. 

GP100 is going to be used on Tesla, and that's only a compute card. Unless NVIDIA plans on pushing that GPU to some sort of Quadro based product, and even then, the drivers are going to be different. They're going to be built to handle mostly professional and compute work, rather than gaming based workloads. 

[Godlygamer23]

5 minutes ago, Castdeath97 said:

I even wonder if that thing every sold. 

People who reviewed it(like Guru3D) had to have it purchased(I believe Guru did it through a sponsor) because NVIDIA wouldn't provide review samples, so they at least sold one.  

 

Edit: There are actually eight listings for them on eBay.

http://www.ebay.com/sch/i.html?_from=R40&_trksid=p2050601.m570.l1313.TR0.TRC0.H0.XTitan+Z.TRS0&_nkw=Titan+Z&_sacat=0

Edited May 21, 2016 by Godlygamer23

[patrickjp93]

3 minutes ago, awesomeness10120 said:

It's either NVIDIA being overconfident and thinking AMD is not a threat and price gouging the hell out of their GPUs, or the cost the manufacture has skyrocketed to the point where they need to up prices. I wanna see a new 4870. 

No, it's an effort to get good yields. If Nvidia is slicing out the 64-bit compute from consumer cards to get smaller dies and better yields while keeping the same SP count as the GP 100 chip (which was also confirmed to be 3840 SPs total when JSH announced the P100 Tesla), then what is there to complain about?

[Tech_Dreamer]

still no HBM2?

[Thiback]

2 minutes ago, awesomeness10120 said:

It's either NVIDIA being overconfident and thinking AMD is not a threat and price gouging the hell out of their GPUs, or the cost the manufacture has skyrocketed to the point where they need to up prices. I wanna see a new 4870. 

What do you mean by wantingt a new HD 4870 ?

[awesomeness10120]

Just now, patrickjp93 said:

No, it's an effort to get good yields. If Nvidia is slicing out the 64-bit compute from consumer cards while keeping the same SP count as the GP 100 chip (which was also confirmed to be 3840 SPs total when JSH announced the P100 Tesla), then what is there to complain about?

Absolutely nothing on the hardware side. People are salty about prices. GTX 500 series pricing would be a slam dunk.

[patrickjp93]

1 minute ago, awesomeness10120 said:

Absolutely nothing on the hardware side. People are salty about prices. GTX 500 series pricing would be a slam dunk.

From a business perspective, maybe not. AMD has historically not done well by playing the cheapness card. Recent changes in marketshare are also minuscule, and I await trends, but history says lowering prices doesn't bode well for IC makers.

[Curufinwe_wins]

4 minutes ago, Godlygamer23 said:

GP100 is going to be used on Tesla, and that's only a compute card. Unless NVIDIA plans on pushing that GPU to some sort of Quadro based product, and even then, the drivers are going to be different. They're going to be built to handle mostly professional and compute work, rather than gaming based workloads. 

This has the exact same cuda configuration as P100 compared to GP102. I think this leak is just blatantly wrong (or using incorrect labels for dies).

 

repeating my comments from the other thread:

 

yea I gotta say that doesn't make sense...

 

Three high end pascal dies (all GK110 gpus were labeled as GK110 even with/without DP) and no lower end one?

 

The 1070 is already a 3/4 cutdown of the 1080, they would NEVER do a 1/2 cutdown. That is an INSANE waste. You just don't do that with GPU dies. No one does, no one ever has (compare 970m to GM204 is the largest cutdown done in recent times, 965m is a GM206 gpu)

 

Also G"P100" is already known to be the top end gpu name unless this one is just using the wrong label there. The core count matches so it seems likely to just be wrongly labeled. But HBM2 seems like a prereq on these cards.