** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **

[Orangey]

"NVIDIA Confidential" lol.

 

[Muzy]

More "leaks" coming out at the moment.

Source
http://videocardz.com/58625/nvidia-geforce-x80ti-and-geforce-x80-titan-specs-hit-the-web-or-do-they

fake

 

[Jono8]

More "leaks" coming out at the moment.

Source
http://videocardz.com/58625/nvidia-geforce-x80ti-and-geforce-x80-titan-specs-hit-the-web-or-do-they

Would be awesome if true and would make the X80 card significantly more powerful than a 980Ti.

 

[DmitryD]

"Specs" are wishful thinking/guesswork.

 

[smogsy]

Depending on the % diff, might get the smaller pascal, Coming from 980s.

cant wait for pascal

 

[Fishpan]

What's with the discrepancy in the Pascal series with some cards using HBM2 and others using GDDR5? Wasn't HBM2 was a selling point of Pascal?

 

[Rroff]

What's with the discrepancy in the Pascal series with some cards using HBM2 and others using GDDR5? Wasn't HBM2 was a selling point of Pascal?

HBM2 availability will have some implications - 8GB won't be available for atleast a few weeks yet possibly months depending on how close the reality is to Samsung's PR and 16GB isn't looking like being available until the end of the year or early next. 4GB on a new flagship would be a bit meh.

I'll be surprised if both memory techs are supported on the same design though.

 

[Mtom]

HBM2 availability will have some implications - 8GB won't be available for atleast a few weeks yet possibly months depending on how close the reality is to Samsung's PR and 16GB isn't looking like being available until the end of the year or early next. 4GB on a new flagship would be a bit meh.

I'll be surprised if both memory techs are supported on the same design though.

There seems to be some misunderstanding here. The HBM2 production started with 4GB capacity modules. They can put 4 of those on a 4096bit bus to reach a total of 16GB.
Second half this year they will start manufacturing the 8GB modules, with 4of those on a 4096bit busthe total memory capacity can be 32GB

 

[Kaapstad]

HBM2 availability will have some implications - 8GB won't be available for atleast a few weeks yet possibly months depending on how close the reality is to Samsung's PR and 16GB isn't looking like being available until the end of the year or early next. 4GB on a new flagship would be a bit meh.

I'll be surprised if both memory techs are supported on the same design though.

My Titan X runs smoother on the ROTTR bench @2160p using 12gb of GDDR5 than my Fury X does @1080p using 4gb of HBM1.

4gb of HBM on any new Pascal/Polaris cards would be a disaster.

 

[ICDP]

My Titan X runs smoother on the ROTTR bench @2160p using 12gb of GDDR5 than my Fury X does @1080p using 4gb of HBM1.

4gb of HBM on any new Pascal/Polaris cards would be a disaster.

While I agree 100% that 4GB VRAM on the new Polaris/Pascal GPUs would be a terrible idea using that test case as proof is fallacious. That's an indication there's a problem with that specific game rather than an issue with 4GB VRAM at 1080p.

4GB VRAM should be fine for 1080p.

 

[Kaapstad]

While I agree 100% that 4GB VRAM on the new Polaris/Pascal GPUs would be a terrible idea using that test case as proof is fallacious. That's an indication there's a problem with that specific game rather than an issue with 4GB VRAM at 1080p.

4GB VRAM should be fine for 1080p.

That is the whole point though.

The 12gb on a TX can handle all the rubbish game devs want to throw at it.

 

[Rroff]

There seems to be some misunderstanding here. The HBM2 production started with 4GB capacity modules. They can put 4 of those on a 4096bit bus to reach a total of 16GB.
Second half this year they will start manufacturing the 8GB modules, with 4of those on a 4096bit busthe total memory capacity can be 32GB

What you say is correct however in terms of what we'll see on GPUs with HBM2 it doesn't look like we'll be seeing the bigger configurations any time soon. (amongst other things ~40mm2 HBM v ~90mm2 HBM2 per package likely means in most cases 2 rather than 4 stacks on a typical GPU).

 

[theMilitant]

Would be slightly underwhelmed by 6GB VRAM on the x80 card if that spec turns out to be correct. Hampers its chances of being a true 4K card for a few years to come.

 

[Mauller]

Apparently a more genuine Pascal leak of information will be released on WCCFtech sometime soon. Khalid is saying that people will not be too fond of what the info is.

 

[mahius]

Apparently a more genuine Pascal leak of information will be released on WCCFtech sometime soon. Khalid is saying that people will not be too fond of what the info is.

No hard feelings mate, but the words 'geniune', 'leak' and 'WCCFtech' don't go together well. Knowing that site, it'll be yet more baseless conjecture.

Either that or Nvidia is delaying Pascal because the Polaris 10 demo had them worried about selling their next 'Titan' at 1.5k buckaroos lol

Nvidia is looking a bit empty-handed right now, AMD have now shown off both Polaris 11 and 10, whereas Nvidia have shown jack. It's got me worried that the Pascal releases (especially x80ti) will be much later than we think. Perhaps not even anything in 2016.

 

[ICDP]

That is the whole point though.

The 12gb on a TX can handle all the rubbish game devs want to throw at it.

Really, are you saying the only reason we should get more VRAM is because it is a counter to the fact some devs are rubbish?

 

[Mtom]

What you say is correct however in terms of what we'll see on GPUs with HBM2 it doesn't look like we'll be seeing the bigger configurations any time soon. (amongst other things ~40mm2 HBM v ~90mm2 HBM2 per package likely means in most cases 2 rather than 4 stacks on a typical GPU).

If you only put two ram chips on the interposer you only use half the bandwith.

 

[bru]

Nvidia is looking a bit empty-handed right now, AMD have now shown off both Polaris 11 and 10, whereas Nvidia have shown jack. It's got me worried that the Pascal releases (especially x80ti) will be much later than we think. Perhaps not even anything in 2016.

To be honest I am quite surprised that AMD have shown as much of Polaris as they have, normally we get nothing official at all from either side right up until announcement day. I just hope this trickle of info from AMD doesn't come back to bite them in the backside somehow.

 

[pmc25]

No hard feelings mate, but the words 'geniune', 'leak' and 'WCCFtech' don't go together well. Knowing that site, it'll be yet more baseless conjecture.

Either that or Nvidia is delaying Pascal because the Polaris 10 demo had them worried about selling their next 'Titan' at 1.5k buckaroos lol

Nvidia is looking a bit empty-handed right now, AMD have now shown off both Polaris 11 and 10, whereas Nvidia have shown jack. It's got me worried that the Pascal releases (especially x80ti) will be much later than we think. Perhaps not even anything in 2016.

Titan and x80ti haven't been thought to be a '16 release for a long, long time. Nearly a year I'd say.

 

[Orangey]

If you only put two ram chips on the interposer you only use half the bandwith.

2048 bit bus and 512GB/s is still plenty for mid-range. Besides any chunks not spread across multiple stacks are limited to one stack's b/w anyway.