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

[andybird123]

Where is your evidence then?
The last die shrink we had on GPUs was going from the 580 to the 670/80 which were some pretty big gains

the 780/ti was the real 580 replacement in terms of die size and that was about double

 

[Gabbatek]

I will be very surprised if any Pascal cards are available before June.

When do you think 980ti prices will drop?

 

[TheF34RChannel]

I was on Fudzilla (always good for a laugh) and actually like their naming schemes:

"GeForce PX" for Pascal, "GeForce VX" for Volta

If the ending SKU number indicated the number of GPU shaders, the product lineup would look as follows:

GeForce PX 4096
GeForce PX 5120 Ti
GeForce PX Titan
GeForce PX Titan M

 

[AllBodies]

Where is your evidence then?
The last die shrink we had on GPUs was going from the 580 to the 670/80 which were some pretty big gains

the 780/ti was the real 580 replacement in terms of die size and that was about double

Also the R9 390X is more than twice the speed of the GTX 580 and is also a smaller die. So it's ~2.5x the performance per mm2. Clearly no loss in moving down nodes there.

 

[melmac]

but there is still room there for something a bit faster than a 980ti but at 970/980 type prices, as with every other die shrink previous.

Every other die shrink? The last Die Shrink the mid range *70 part was more expensive than the previous high end *80 card.

Obviously!

If it was more expensive and slower than a 980Ti then what would be the point of it, people would just buy a 980Ti. If it was cheaper and slower it would just be another 970.

We don't need to know the square root of a nanometer to work out any of this! The X70/80 will be less expensive and faster than a 980Ti else it will be completely irrelevant.

Edit: rereading this post, it comes across as a bit condescending... Not intended. What I'm saying is that the mid range will be cheaper and marginally faster than a 980Ti else there will be no point.

I will be amazed if there is card released that's more powerful than then 980Ti at the same price as the 970.

 

[melmac]

Where is your evidence then?
The last die shrink we had on GPUs was going from the 580 to the 670/80 which were some pretty big gains

the 780/ti was the real 580 replacement in terms of die size and that was about double

it took them a year to get that sort of performance. The conversation me and Roff were having was that the the performance gap between one generations high end card and the next released mid range card. That's comparing the 670 to the 580, or the 460 vs 285, etc. I said that gap is getting smaller and he agreed.

Also the R9 390X is more than twice the speed of the GTX 580 and is also a smaller die. So it's ~2.5x the performance per mm2. Clearly no loss in moving down nodes there.

Clearly no loss? How long did it take them to reach that performance? over 3 years after the die shrink!! Will we be waiting 4 or 5 years before we get that increase in performance in the next die shrink?

Die shrinks are becoming more and more difficult and more costly. Are you really expecting a mid range card out at 970 prices that beats the 980ti? Good luck with that.

 

[Dirk Diggler]

I will be amazed if there is card released that's more powerful than then 980Ti at the same price as the 970.

... Are you really expecting a mid range card out at 970 prices that beats the 980ti? Good luck with that.

Consider the new name to be X70. Disregarding price for a minute, let's consider performance. If, like you say, it has no chance of competing with a 980Ti then what does it compete with? A 980? This would make it 10% faster than a 970 which is its direct predecessor. Really? 10%? What would be the point? Why upgrade your 970 for a minor boost that's been available for nearly 2 years?

The perfect strategy would be :

X70 - max £400 and +/- 5% 980Ti
X80 - max £500 and + 20% 980Ti

This begs the question, what price then the 980Ti. The answer is, who cares? It's obsolete. Nvidia won't care what price a GPU it no longer manufactures can fetch. Was anyone paying £550 for a 780ti when the 970 launched? Of course not, but they were paying that shortly before it launched.

Edit: to clarify, those prices are the top range cards as there will be cheaper, less performing variants.

 

[Simon Belmont]

Consider the new name to be X70. Disregarding price for a minute, let's consider performance. If, like you say, it has no chance of competing with a 980Ti then what does it compete with? A 980? This would make it 10% faster than a 970 which is its direct predecessor. Really? 10%? What would be the point? Why upgrade your 970 for a minor boost that's been available for nearly 2 years?

The perfect strategy would be :

X70 - £400 and +/- 5% 980Ti
X80 - £500 and + 20% 980Ti

This begs the question, what price then the 980Ti. The answer is, who cares? It's obsolete. Nvidia won't care what price a GPU it no longer manufactures can fetch. Was anyone paying £550 for a 780ti when the 970 launched? Of course not, but they were paying that shortly before it launched.

This^ exactly happened from 700 series to 900 series. I expect 900 series to Pascal would be the same? Tuesday after 5pm hope we get the answer?

 

[melmac]

Consider the new name to be X70. Disregarding price for a minute, let's consider performance. If, like you say, it has no chance of competing with a 980Ti then what does it compete with? A 980? This would make it 10% faster than a 970 which is its direct predecessor. Really? 10%? What would be the point? Why upgrade your 970 for a minor boost that's been available for nearly 2 years?

The perfect strategy would be :

X70 - £400 and +/- 5% 980Ti
X80 - £500 and + 20% 980Ti

Nvidia have surprised me before, I guess so has AMD. So maybe anything is possible and I am been a bit negative

What would you say to

X70 - £350 midway between the 980 and the 980Ti
X80 - £450 a little faster than the 980Ti.

The new cards come with 6GB or 8Gb ram as standard and with extra compute power to help with Dx12.

Just after seeing your edit!! I am presuming that Nvidia will have the same lineup with Pascal, a Titan, Ti, X80, X70 etc.

 

[melmac]

This^ exactly happened from 700 series to 900 series. I expect 900 series to Pascal would be the same? Tuesday after 5pm hope we get the answer?

except the 900 series came at the end of 28nm where things were cheaper and allowed Nvidia to sell at a really good price. There is going to be extra cost with Pascal from the die shrink.

 

[Dirk Diggler]

Nvidia have surprised me before, I guess so has AMD. So maybe anything is possible and I am been a bit negative.


Just after seeing your edit. I am taking that Nvidia will have the same line up, a titan, a Ti, an X80 and an X70.
What would you say to

X70 - £350 midway between the 980 and the 980Ti
X80 - £450 a little faster than the 980Ti.

The new cards come with 6GB or 8Gb ram as standard and with extra compute power to help with Dx12.

Yes, I think this reduced price is indeed feasible also. Why not? Both scenarios facilitate a vibrant market for the new cards and a period of time to shift the older generation Maxwell cards for SLI configs and system builders.

The new cards can't be too slow else there will be no market for them, and they can't be too fast else there will be a heap of Maxwell cards that can't be shifted.

6 months down the line and boom, the titan and Ti appears and who knows what price or performance they bring.

 

[tyler_jrb]

Even without the die shrink if the last gen is anything to go by.

980 with 2048 cores beat the 780Ti with 2880.
970 with 1664 cores beat the 780 with 2304.

That was without a die shrink or gddr5x memory. This time we have a 28-16nm reduction. Which equates to about 43%.

Depends how well nvidia take advantage of this. Hopefully will find out soon .

 

[bru]

The perfect strategy would be :

X70 - max £400 and +/- 5% 980Ti
X80 - max £500 and + 20% 980Ti

This^ exactly happened from 700 series to 900 series. I expect 900 series to Pascal would be the same? Tuesday after 5pm hope we get the answer?

Er no it didn't, my MSI 970 was £281.99 3 days after launch, nowhere near £400

 

[AllBodies]

Clearly no loss? How long did it take them to reach that performance? over 3 years after the die shrink!! Will we be waiting 4 or 5 years before we get that increase in performance in the next die shrink?

Die shrinks are becoming more and more difficult and more costly. Are you really expecting a mid range card out at 970 prices that beats the 980ti? Good luck with that.

Well first of all the 390X is just a 290X with twice the RAM. And second of all you were saying there was a slowdown of performance per node, not per launch.

Just because they've changed strategy to release the smaller dies first to help profit and wait for maturity doesn't mean the nodes are worse. The GTX 580 didn't launch on 40nm either...

If you want to compare properly you'd compare the fastest equivalent card ever made on each node. So GTX 580 vs 780 Ti or 390X (perhaps even Fury X). And then look at the performance per mm2 you're getting from the node.

If you're insinuating we should compare the GTX 580 to the 680 to determine 28nm's performance, then that's ludicrous.

EDIT: And you wouldn't compare the 980 Ti or Titan X to the 580, because they had the DP hardware stripped out, so they're not equivalent.

 

[melmac]

Well first of all the 390X is just a 290X with twice the RAM. And second of all you were saying there was a slowdown of performance per node, not per launch.

Just because they've changed strategy to release the smaller dies first to help profit and wait for maturity doesn't mean the nodes are worse. The GTX 580 didn't launch on 40nm either...

If you want to compare properly you'd compare the fastest equivalent card ever made on each node. So GTX 580 vs 780 Ti or 390X (perhaps even Fury X). And then look at the performance per mm2 you're getting from the node.

If you're insinuating we should compare the GTX 580 to the 680 to determine 28nm's performance, then that's ludicrous.

EDIT: And you wouldn't compare the 980 Ti or Titan X to the 580, because they had the DP hardware stripped out, so they're not equivalent.

Wait, you can shout theory to me all night long. It doesn't change anything.

When there is a die shrink, the gap between the last generations high end card and the next gen mid range card has been decreasing.

Read it again, 9800 to 260, 285 to 470, 580 to 670. The 780Ti to 970 is a new process but not a die shrink. So hard to make a call on that one.

I am not talking about anything else, because we were discussing what the performance of what the Pascal's mid range will be. Not the high end card, not a card that's might be released 2 years from now. No, just the mid range X70 card that's supposedly been released first.

So maybe it's my fault a few of my posts have been a bit ambiguous. The discussion started comparing mid range cards to previous high end cards and all my posts were about that only. I wasn't talk about the life of a new die shrink, just the cards first released on a new node.

 

[AllBodies]

Wait, you can shout theory to me all night long. It doesn't change anything.

When there is a die shrink, the gap between the last generations high end card and the next gen mid range card has been decreasing.

Read it again, 9800 to 260, 285 to 470, 580 to 670. The 780Ti to 970 is a new process but not a die shrink. So hard to make a call on that one.

I am not talking about anything else, because we were discussing what the performance of what the Pascal's mid range will be. Not the high end card, not a card that's might be released 2 years from now. No, just the mid range X70 card that's supposedly been released first.

So maybe it's my fault a few of my posts have been a bit ambiguous. The discussion started comparing mid range cards to previous high end cards and all my posts were about that only. I wasn't talk about the life of a new die shrink, just the cards first released on a new node.

That's fair enough if that's what you were getting at.

At the same time though, what I'm talking about is kind of related and you should take it somewhat into account.

What I mean is, I believe it's agreed upon Pascal is meant to be 2x the performance per watt of Maxwell (AMD claim Polaris is 2.5x Fiji). And this kind of also means it's twice the performance per mm2.

So if we hopefully get told the die size of the X70 and X80, we should have a better idea of how it should perform. If they're more than 300mm2 then the X70 should match the 980 Ti.

 

[melmac]

That's fair enough if that's what you were getting at.

At the same time though, what I'm talking about is kind of related and you should take it somewhat into account.

What I mean is, I believe it's agreed upon Pascal is meant to be 2x the performance per watt of Maxwell (AMD claim Polaris is 2.5x Fiji). And this kind of also means it's twice the performance per mm2.

So if we hopefully get told the die size of the X70 and X80, we should have a better idea of how it should perform. If they're more than 300mm2 then the X70 should match the 980 Ti.

That's exactly what I have been saying. Just talking about the mid range.

Over the life of a node, well, who knows what will happen? But performance out the gate on the 28nm process wasn't as big as people expected. It took time for both camps to get up to speed.

Do you think that it's going to be different for this launch? I am trying to be optimistic, but I think it will be worse in the beginning. It's way more complex.

And then you have to consider my point about power. The blurb from both camps is power saving. It's a global problem now. Sure there is that 2 times performance per watt of Maxwell, but will they use that to lower power consumption or bring the max performance? MY guess is that saving power will be the primary focus.

Ignore my figures, but say you have a card that is using 300 watts now with 100% performance. The next gen card could get twice the performance for the same 300 watts or it could use 200 watts but get a lower performance increase.

I just have a feeling that it's going to be the latter, that it's all about the power use.

 

[easyrider]

I just have a feeling that it's going to be the latter, that it's all about the power use.

Me too.

 

[Dicehunter]

So any real news yet or is it still BS and lies made up by forum goers ?

 

[Dirk Diggler]

A company that markets a 12GB card for $1000 is going to be focused on efficiency over performance in order to get the enthusiast to part with another $1000?

I disagree. In mobile applications maybe, but the discrete desktop market? No way.