NVIDIA Pascal Thread

[MrTeal]

You think they'll double release the 1080/1070 That close together?(gddr5 version then 5x version. That's not the Nvidia I know..

Big if, since I'm not expecting GP104 in May, but if they did release that early with GDDR5 it would make a good refresh similar to Kepler where GP104 gets relaunched as 1170 with GDDR5X in Q1 2017 with GP100 filling in the top end.

 

[railven]

What? Why not? You don't wanna get this chip, then the Titan then the 1080ti?

Money around my household is treated like a government contract. If I don't spend it, she does, and it's gone

So, I'll be upgrading . Interested to see who gets it this year.

 

[tential]

Money around my household is treated like a government contract. If I don't spend it, she does, and it's gone

So, I'll be upgrading . Interested to see who gets it this year.

I won't be surprised if Nvidia releases the 1080 with gddr5. It helps Nvidia be first to market on a new node. If Polaris is faster it won't matter if Nvidia picks up the glut of people who won't wait.

 

[railven]

I won't be surprised if Nvidia releases the 1080 with gddr5. It helps Nvidia be first to market on a new node. If Polaris is faster it won't matter if Nvidia picks up the glut of people who won't wait.

Oh I'll wait, at least a month, like I did for Fury X to arrive (granted that wasn't a month, but I gave Fury X a week or so for stock before I just got the 980 Ti) and hopefully both Polaris 10/11 and GP104 are on the table so I can buy the better one for my needs.

 

[LTC8K6]

I won't be surprised if Nvidia releases the 1080 with gddr5. It helps Nvidia be first to market on a new node. If Polaris is faster it won't matter if Nvidia picks up the glut of people who won't wait.

Plus, NV doesn't seem to be suffering with GDDR5 against HBM. If they can get a little bit more out of GDDR5, it will probably be fine.

 

[dark zero]

Plus, NV doesn't seem to be suffering with GDDR5 against HBM. If they can get a little bit more out of GDDR5, it will probably be fine.

AMD suffered due their poor GPU instead of HBM which looks OK.

Wondering if this chip leaves the GTX 980 Ti behind... if not, is a very bad news.

Also, the 980Ti must be at the levels of the GTX 970 sucessor.

 

[airfathaaaaa]

so now it became a fact that nvidia will launch them on may and thus you changed the header?

 

[moonbogg]

What? Why not? You don't wanna get this chip, then the Titan then the 1080ti?

That's exactly the issue. If I buy these, I'm sure I will like them, but then before too long a titan will come out and I will then wish I had waited, because soon a Ti version would land. I will just stick with the big die cards.
The thing that sucks is that Nvidia now releases mid range cards first. This makes it certain that you can no longer enjoy having the best for an entire generation like I used to could. You enjoy having the best for about a year, and then something new is out to either tempt you or irritate you. I like the old business model much more, and that's why they changed it. They don't want their customers having maximum satisfaction. That is reserved for them. BASTARDS!

 

[Head1985]

Would you guy be interested in something like this?

Geforce 'GTX 1080'
- +$500
- Full GP104 die
- Relatively small die and significant OC headroom (>20%)
- 15-25% faster than Geforce GTX980 Ti
- 256-bit / 512-bit GDDR5, new model @ H2 with GDDR5X

Geforce 'GTX 1070'
- +$400
- Relatively small die and significant OC headroom (>20%)
- Slightly faster than Geforce GTX980 Ti
- 256-bit / 512-bit GDDR5, new model @ H2 with GDDR5X

Nope.Not interested at all.
performance gain need be same as was GTX670/680 vs GTX580.35-40% for GTX680 and 20-25%for 670.(latest Die shrink)GTX670/680 also was mid-range SKU that beat FULL SKU Flagship GTX580 on 40nm node.
35-40% faster than 980TI for 1080
20-25% faster than 980TI for 1070
Or its just fail.We talking about Die shrink first after 5 years.If 1080 its only 20-25% faster than GTX980TI it will be worst performance jump in Die shrink history(jump to new node)

 

[LTC8K6]

http://wccftech.com/nvidia-pascal-geforce-gtx-1080-gp104-gpu/

There is the wccf story, which also seems to add that it will be more efficient, needing only a single 8 pin connector.

Also says that GDDR5X cards might come later.

Maybe they will be Ti models...

 

[MrTeal]

Nope.Not interested at all.
performance gain need be same as was GTX670/680 vs GTX580.35-40% for GTX680 and 20-25%for 670.(latest Die shrink)GTX670/680 also was mid-range SKU that beat FULL SKU Flagship GTX580 on 40nm node.
35-40% faster than 980TI for 1080
20-25% faster than 980TI for 1070
Or its just fail.We talking about Die shrink first after 5 years.If 1080 its only 20-25% faster than GTX980TI it will be worst performance jump in Die shrink history(jump to new node)

Huh? At launch the 680 was only 25% faster than the 580 in TPU's suite, plus it cost close to 30% more than a 580 was selling for. Some games were higher of course, but 35-40% wasn't the average.
25% faster than the previous flagship and for $100 less would actually be a really good launch for GTX1080.

 

[Head1985]

At launch it was around 30-35% for GTX680 and 20% for 670.

I expecting even more from pascal because GTX980TI its just cutdown version.Its slower than TITANX by 5-8%.
SO 35-40% from 1080 vs cutdown 980TI
And 20-25% from 1070 vs cutdown 980TI.

 

[iiiankiii]

Nope.Not interested at all.
performance gain need be same as was GTX670/680 vs GTX580.35-40% for GTX680 and 20-25%for 670.(latest Die shrink)GTX670/680 also was mid-range SKU that beat FULL SKU Flagship GTX580 on 40nm node.
35-40% faster than 980TI for 1080
20-25% faster than 980TI for 1070
Or its just fail.We talking about Die shrink first after 5 years.If 1080 its only 20-25% faster than GTX980TI it will be worst performance jump in Die shrink history(jump to new node)

Well, it used to be like you said. Too bad node shrinks are taking longer and longer. They have to switch up strategy. I think both AMD and Nvidia will slow roll this node shrink. 15-25% increase sounds like the going rate for future releases. They need to make each node shrink last longer. I expect AMD and Nvidia to be stuck on 14nm for the next 3-5 years.

 

[tential]

That's exactly the issue. If I buy these, I'm sure I will like them, but then before too long a titan will come out and I will then wish I had waited, because soon a Ti version would land. I will just stick with the big die cards.
The thing that sucks is that Nvidia now releases mid range cards first. This makes it certain that you can no longer enjoy having the best for an entire generation like I used to could. You enjoy having the best for about a year, and then something new is out to either tempt you or irritate you. I like the old business model much more, and that's why they changed it. They don't want their customers having maximum satisfaction. That is reserved for them. BASTARDS!

Actually it's slightly worse than that. Not only is the 1080 Going to be faster it's going to excel in specific areas maxwell didn't.

Do you seriously think the next round of game works will be optimized so that maxwell and pascal do well?
With Hawaii already looking like it has legs it shouldn't, I see Nvidia full steam ahead with pascal. They want people like you to upgrade and when pascal does well and crushes in Nvidia sponsor ed games people will upgrade if they're first to market. Enough will anyway.

 

[Head1985]

Well, it used to be like you said. Too bad node shrinks are taking longer and longer. They have to switch up strategy. I think both AMD and Nvidia will slow roll this node shrink. 15-25% increase sounds like the going rate for future releases. They need to make each node shrink last longer. I expect AMD and Nvidia to be stuck on 14nm for the next 3-5 years.

They can do pretty much same thing like they did in 28nm.
Gp104 around 300mm2 GPU 35-40%faster than 980TI
Then year later GP100 500-550mm2 with 100% faster than 980TI
Year and half later new architekture and same node.
GV104(volta) around 400mm2 GPU beat GP100 at 8% same as GTX980 did vs 780TI at launch.
Then year later GV100(volta) 600mm2 40%faster than GP100.

 

[MrTeal]

At launch it was around 30-35% for GTX680 and 20% for 670.
I expecting even more from pascal because GTX980TI its just cutdown version.Its slower than TITANX by 5-10%.
SO 35-40% from 1080 vs cutdown 980TI
And 20-25% from 1070 vs cutdown 980TI.

It all depends on your test suite, of course.

TPU measured ~25%, Guru3D measured 31%. That's still a ways off 35-40% faster than a 580, plus as I said it cost 30% more than a 580. If you were to get 980Ti + 25% and that big a price drop, it would be a great value and they'd sell a ton of them.

Keep in mind, along with the last new node came a pretty big shrink in die sizes. GF110 was 520mm² and GP104 was 294mm², and the transition also had the benefit of moving from a DP compute heavy die to a DP light die. Here we're moving from a very large 601mm² die that's already extremely light on on DP compute, to an unknown size for GP104. An equivalent size ratio would be 340mm².

I'd love to see a GTX at 980Ti + 40% (or Titan X + 35% if you want to use full dies) performance for $500-$550, but I'd be pretty shocked if it happened.

 

[Genx87]

Actually it's slightly worse than that. Not only is the 1080 Going to be faster it's going to excel in specific areas maxwell didn't.

Do you seriously think the next round of game works will be optimized so that maxwell and pascal do well?
With Hawaii already looking like it has legs it shouldn't, I see Nvidia full steam ahead with pascal. They want people like you to upgrade and when pascal does well and crushes in Nvidia sponsor ed games people will upgrade if they're first to market. Enough will anyway.

This will be the norm going forward as we use close to the metal APIs. What works on one arch may not work so well on another. I do expect Nvidia and AMD to support at least one generation back. But given the complexity of coding closer to the metal. I will be shocked if they will support 3-4 or more generations back besides basic compatibility fixes.

 

[MrTeal]

Actually it's slightly worse than that. Not only is the 1080 Going to be faster it's going to excel in specific areas maxwell didn't.

Do you seriously think the next round of game works will be optimized so that maxwell and pascal do well?
With Hawaii already looking like it has legs it shouldn't, I see Nvidia full steam ahead with pascal. They want people like you to upgrade and when pascal does well and crushes in Nvidia sponsor ed games people will upgrade if they're first to market. Enough will anyway.

New gameworks feature: Moonworks. Accurately textures and shadows the moon in outdoor areas to minute deal. Exclusive use of FP16. Approximately 20% FPS hit to GP104, 50+% hit to Maxwell/Polaris.

 

[DooKey]

Whoever is first to market gets my money. I want a new shiny.

 

[moonbogg]

New gameworks feature: Moonworks. Accurately textures and shadows the moon in outdoor areas to minute deal. Exclusive use of FP16. Approximately 20% FPS hit to GP104, 50+% hit to Maxwell/Polaris.

That's right bichtes.

 

[Head1985]

It all depends on your test suite, of course.
TPU measured ~25%, Guru3D measured 31%. That's still a ways off 35-40% faster than a 580, plus as I said it cost 30% more than a 580. If you were to get 980Ti + 25% and that big a price drop, it would be a great value and they'd sell a ton of them.

Keep in mind, along with the last new node came a pretty big shrink in die sizes. GF110 was 520mm² and GP104 was 294mm², and the transition also had the benefit of moving from a DP compute heavy die to a DP light die. Here we're moving from a very large 601mm² die that's already extremely light on on DP compute, to an unknown size for GP104. An equivalent size ratio would be 340mm².

I'd love to see a GTX at 980Ti + 40% (or Titan X + 35% if you want to use full dies) performance for $500-$550, but I'd be pretty shocked if it happened.

Yep around 30% vs GTX580.But GTX980TI is not flagship.Its only cutdown SKU.Its slover than TITANX by 5-8%.If you compare 1080 vs cutdown GTX980TI you need add 5-8% to that.
I expecting same performance gain from 1080/1070 like 680/670 bring.
So yeah 30% vs TITANX
35-40% vs 980TI
BTW both GTX580 and 680 have at launch MSRP 500USD.
https://www.techpowerup.com/gpudb/270/geforce-gtx-580.html
https://www.techpowerup.com/gpudb/342/geforce-gtx-680.html

 

[xorbe]

New gameworks feature: Moonworks. Accurately textures and shadows the moon in outdoor areas to minute deal. Exclusive use of FP16. Approximately 20% FPS hit to GP104, 50+% hit to Maxwell/Polaris.

Joke is so how NV operates though, lol.

 

[swilli89]

At launch it was around 30-35% for GTX680 and 20% for 670.

I expecting even more from pascal because GTX980TI its just cutdown version.Its slower than TITANX by 5-8%.
SO 35-40% from 1080 vs cutdown 980TI
And 20-25% from 1070 vs cutdown 980TI.

Amazing that anyone, looking at that chart.. Look at that 7970, apparently much slower than a 680, then three years later MUCH faster. Incredible the change there.

 

[tential]

From latest Tomb Raider game patch notes:

Adds NVIDIA VXAO Ambient Occlusion technology. This is the world’s most advanced real-time AO solution, specifically developed for NVIDIA Maxwell hardware.

Just saying, don't expect to NOT upgrade to Pascal the second it drops if you're an Nvidia user who values Nvidia features. You will need to upgrade to the 1070/1080 (and then again if you're a person who wants a big chip), if you want to stay current with Nvidia features.

And I mean, if you look at games that employ GAmeworks, you see a SIGNIFICANT drop in Kepler performance.
I'm just saying, I know a 980Ti user will find it unacceptable they have the performance of a R9 390x when Pascal hits.

It'll be interesting to see if the market rewards Nvidia again or not.


I don't know what more proof you need than Nvidia only cares about the current gen user of cards...

 

[DooKey]

From latest Tomb Raider game patch notes:


Just saying, don't expect to NOT upgrade to Pascal the second it drops if you're an Nvidia user who values Nvidia features. You will need to upgrade to the 1070/1080 (and then again if you're a person who wants a big chip), if you want to stay current with Nvidia features.

And I mean, if you look at games that employ GAmeworks, you see a SIGNIFICANT drop in Kepler performance.
I'm just saying, I know a 980Ti user will find it unacceptable they have the performance of a R9 390x when Pascal hits.

It'll be interesting to see if the market rewards Nvidia again or not.


I don't know what more proof you need than Nvidia only cares about the current gen user of cards...

Don't be obtuse. Sometimes only the current gen has the hardware to do certain things. However, the market has spoken and it appears NV has a market that is willing to upgrade often versus the market AMD holds that likes to buy fire sale merchandise and only upgrade when they get more fire sale merchandise. I know which market I would shoot for.