• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
  • The Politics forum has been nuked. Please do not bring political discussion to the rest of the site, or you will be removed. Thanks.

GTX 970/980 replaced by Pascal with GDDR5X at Computex

TSM

Member
Feb 15, 2014
2,566
0
0
Nvidia squeezed like a 50 percent improvement out of maxwell just from a new architecture on the same 28nm node

Now there's a switch from 28nm to 16nm ff+ (and another architecture change, though we don't know how much different it is from maxwell)

It's not going to be 25 percent, 50 percent is probably lowballing it.

I'd also imagine they'd like to stay far enough ahead of AMD to not have to release another card on short notice or cut the prices on these cards dramatically. The 970 was good to Nvidia and I'd imagine they'd like to replicate it's success.
 
Jul 8, 2009
1,245
0
0
I have a GTX780 currently, I also have a Vive coming in the post hopefully by the end of April. Steam VR Perfomance Tool says I'm capable (just shy of green). My question is, do you guys think I should go for a 970/980 right now and hope that I can do the 90 day trade in/difference with EVGA or just wait for Pascal and make due with the 780 and the Vive until then?
 

Hazaro

relies on auto-aim
Jan 21, 2008
32,244
1
1,090
Southern California
I have a GTX780 currently, I also have a Vive coming in the post hopefully by the end of April. Steam VR Perfomance Tool says I'm capable (just shy of green). My question is, do you guys think I should go for a 970/980 right now and hope that I can do the 90 day trade in/difference with EVGA or just wait for Pascal and make due with the 780 and the Vive until then?
Make the decision when your Vive has actually shipped. A 980/980Ti -> Step Up is fine.
 

PnCIa

Member
Nov 26, 2005
2,782
1
1,415
Germany
I'd also imagine they'd like to stay far enough ahead of AMD to not have to release another card on short notice or cut the prices on these cards dramatically. The 970 was good to Nvidia and I'd imagine they'd like to replicate it's success.
I expect AMD to have a compelling 970 of their own this time. Since Nvidia and AMD are both new to their node respectively, they will be on the same ground this time. A good question is how TSMCs 16nm compares to GlobalFoundries 14nm. Exciting times ahead!
 

dr_rus

Member
May 3, 2007
10,983
0
1,350
Moscow, Russia
Nvidia squeezed like a 50 percent improvement out of maxwell just from a new architecture on the same 28nm node

Now there's a switch from 28nm to 16nm ff+ (and another architecture change, though we don't know how much different it is from maxwell)

It's not going to be 25 percent, 50 percent is probably lowballing it.

Depends on what you compare to what. +25% for GP104 when compared to GM204 will be very low in my opinion. +25% for GP104 when compared to GM200 would be very good on the other hand. I feel that we may end up somewhere in between those two figures.
 

NecroGiggles

Member
Nov 4, 2015
80
4
270
USA
Once the new cards hit the market what do you think you could sell a 980 ti for I wonder ? I want a 1080 and I could use some help with the price tag.
 

Freiya

Member
Mar 5, 2014
899
1
405
I'm going to be sad if 980TI drops to under 400 as soon as these cards come out. Reselling 980TI's for the TI equiv of pascal is going to suck ass if that's the case. -_-
 

| Praxis |

Banned
Feb 14, 2010
12,063
1
0
UK
Seems quite the opposite like you should buy it since these are just rebrands with similar performance. Doesn't look like anything will top the 980ti this year.

I'm getting one a month old. Still pretty expensive. Selling my Xbox One and a pair of headphones to cover it though.
 

Oxn

Member
Jul 17, 2011
7,572
5
735
Seems quite the opposite like you should buy it since these are just rebrands with similar performance. Doesn't look like anything will top the 980ti this year.

Please dont give advice if you have no idea what you are talking about?
 

SneakyStephan

Banned
Jan 23, 2011
18,399
0
0
Depends on what you compare to what. +25% for GP104 when compared to GM204 will be very low in my opinion. +25% for GP104 when compared to GM200 would be very good on the other hand. I feel that we may end up somewhere in between those two figures.

I mean +50 percent across the board

gp100 = gm200 + 50 percent, gp104 = gm204 +50 percent , ditto for gp107 whenever it comes out.

You'd assume that's what he meant by a x percent improvement, comparing apples to apples.

ofcourse it would be very good if gp104 was 25 percent faster than gm200 (and then the jump from gm200 to gp100 would be huge)
 

dr_rus

Member
May 3, 2007
10,983
0
1,350
Moscow, Russia
I mean +50 percent across the board

gp100 = gm200 + 50 percent, gp104 = gm204 +50 percent , ditto for gp107 whenever it comes out.

You'd assume that's what he meant by a x percent improvement, comparing apples to apples.

ofcourse it would be very good if gp104 was 25 percent faster than gm200 (and then the jump from gm200 to gp100 would be huge)

Well, 50% increase on average may be too high but in heavy modes and in modern loads it can be even bigger than that. Then again this is a long overdue process change so it may be better than we expect as we've essentially skipped 20nm.

It's also worth to remember that while GM200 was a pure gaming behemoth GP100 is likely to be target at HPC and gaming like GK110 was which means that pure gaming increases for GP100 are likely to be lower than you'd expect.

So a 50% increase for GP104 going off GM204 is very possible in my opinion. A 50% gaming increase for GP100 going off GM200 is less likely imo. But who knows?
 

SneakyStephan

Banned
Jan 23, 2011
18,399
0
0
Well, 50% increase on average may be too high but in heavy modes and in modern loads it can be even bigger than that. Then again this is a long overdue process change so it may be better than we expect as we've essentially skipped 20nm.

It's also worth to remember that while GM200 was a pure gaming behemoth GP100 is likely to be target at HPC and gaming like GK110 was which means that pure gaming increases for GP100 are likely to be lower than you'd expect.

So a 50% increase for GP104 going off GM204 is very possible in my opinion. A 50% gaming increase for GP100 going off GM200 is less likely imo. But who knows?

You think more cuda cores again in gp100?
I hope not, gimme pure gaming performance (also more cuda cores would mean them pushing more physx and flex stuff again, which always had shitty meaningless implementations in games like the clutter in borderlands 2)

And yeah who knows indeed:p

Right before a new gpu architecture comes out is always the most fun (speculation)
 

dr_rus

Member
May 3, 2007
10,983
0
1,350
Moscow, Russia
You think more cuda cores again in gp100?
I hope not, gimme pure gaming performance (also more cuda cores would mean them pushing more physx and flex stuff again, which always had shitty meaningless implementations in games like the clutter in borderlands 2)

And yeah who knows indeed:p

Right before a new gpu architecture comes out is always the most fun (speculation)

CUDA core is just NV's brand naming for a base processing unit in a SIMD cluster aka shader processor (SP). Thus more CUDA cores = more SPs = more performance overall. And of course there will be more SPs. That's the main benefit of a new production process - you can pack more than twice more units in the same chip size on it.
 

grumble

Member
Dec 6, 2008
7,899
0
0
I hope for X80 to be at least 20% faster than 980Ti on average because this way I'll be able to justify the upgrade. Anything less and I'll just wait for GP100 parts.



8GBs for all GP104 parts most likely.

There is a decent chance this will be the biggest jump in a few years, since the lack die shrink. It could be big for AMD and nvidia.
 

Odrion

Banned
Aug 31, 2005
19,688
1
1,130
I don't think Nvidia is the generous type to offer 980ti power at 970 level just because they can. They will if AMD is putting that sort of power at that sort of price, but that's about it.

You'll probably see it shy of 10~15% under 980ti, and the X80 will be 20% more powerful than the 980ti. Juuuust enough performance for people to be as happy as they can be about a $400 card and the X80 is juuuuust powerful enough for people eager for the newness to ditch their 980ti and upgrade (and then ditch the X80 once the X80ti comes out.)

edit: I could also see Nvidia being absolute dicks and making the jump from 970 to X70 still be roughly in 30% improvement range with the argument being "Yeah but now the card is smaller and uses less electricity!"
 

Durante

Member
Oct 1, 2006
48,836
1
0
peter.metaclassofnil.com
I don't think Nvidia is the generous type to offer 980ti power at 970 level just because they can. They will if AMD is putting that sort of power at that sort of price, but that's about it.

You'll probably see it shy of 10~15% under 980ti, and the X80 will be 20% more powerful than the 980ti. Juuuust enough for people to be happy with a $400 Pascal card and the X80 is juuuuust powerful enough for people eager for the newness to ditch their 980ti and upgrade (and then ditch the X80 once the X80ti comes out.)
That's exactly what I would have said (and in fact did say) before the 970 was released.

Perhaps they figure that they make more money by getting people to upgrade more readily at that price point. Or perhaps it was just a one-off experiment. Both seem at least somewhat possible at this point.
 

Odrion

Banned
Aug 31, 2005
19,688
1
1,130
That's exactly what I would have said (and in fact did say) before the 970 was released.

Perhaps they figure that they make more money by getting people to upgrade more readily at that price point. Or perhaps it was just a one-off experiment. Both seem at least somewhat possible at this point.

I guess but isn't the 970 only a bit more powerful than the 780? And the 770 was a bit more powerful than the 680 and so on...

I thought the love for the 970 was 1.) the $330 price and 2.) the "next-gen" bump from the consoles being so muted that the 970 ran everything new at high or ultra at 1080p/60fps.
 

Durante

Member
Oct 1, 2006
48,836
1
0
peter.metaclassofnil.com
I guess but isn't the 970 only a bit more powerful than the 780? And the 770 was a bit more powerful than the 680 and so on...

I thought the love for the 970 was 1.) the $330 price and 2.) the "next-gen" bump from the consoles being so muted that the 970 ran everything new at high or ultra at 1080p/60fps.
I just looked back at the Computerbase benchmarks at the 970 launch, and in those it was only 3% behind the 780ti in average performance. If anything, I'd assume it to be even closer if you include modern games.

Its price/performance ratio certainly exceeded what I would have ever expected them to offer. Though as I said above, that certainly doesn't mean that this will happen again.
 

Beerman462

Member
Aug 12, 2011
8,422
138
845
I just looked back at the Computerbase benchmarks at the 970 launch, and in those it was only 3% behind the 780ti in average performance. If anything, I'd assume it to be even closer if you include modern games.

Its price/performance ratio certainly exceeded what I would have ever expected them to offer. Though as I said above, that certainly doesn't mean that this will happen again.

I feel like the GPU market has been frozen for a year+. I am really confused my the 970/390 market. The 390 was cheaper than the 970 for awhile and now it's more expensive.
 

x3sphere

Member
Sep 24, 2006
4,897
0
1,020
AZ
gamercards.exophase.com
I wouldn't be surprised if the Pascal equivalent of 970 ends up very close to the 980 TI. My thinking is, both GPU manufactures will want to make a big splash with these releases since it's a node shrink, which typically has brought huge performance gains.

And, it's a great time to incentivize people to upgrade because of all the hype surrounding VR.
 

Daante

Member
Sep 19, 2007
2,078
0
0
Megaton?

http://wccftech.com/nvidia-ultra-low-level-gameworks-api/

"What Nvidia has created is probably the first fully concurrent low level API. From what we can gather, the API will be able to operate on an “ultra” low level concurrently (and under) DirectX12"


"True SLI: This particular feature is the first “true” multi-GPU solution in the world. When enabled, the resources of the GPU will be merged using the SLI bridge and the PCI-E interface. Drivers will effectively see only one GPU with pooled hardware specifications. (Two 4GB cards in SLI will show up as one GPU with 8 GB of vRAM)."
 

Kezen

Banned
Jul 28, 2014
8,919
0
375
France
Megaton?

http://wccftech.com/nvidia-ultra-low-level-gameworks-api/

"What Nvidia has created is probably the first fully concurrent low level API. From what we can gather, the API will be able to operate on an “ultra” low level concurrently (and under) DirectX12"


"True SLI: This particular feature is the first “true” multi-GPU solution in the world. When enabled, the resources of the GPU will be merged using the SLI bridge and the PCI-E interface. Drivers will effectively see only one GPU with pooled hardware specifications. (Two 4GB cards in SLI will show up as one GPU with 8 GB of vRAM)."

Look at the date.
 

icecold1983

Member
Feb 28, 2010
6,270
0
780
id say 50% is the top range of what to expect when looking at replacements in each gpus respective tier. id be shocked if nvidia offers more than that, for business reasons alone if nothing else. rumors also say yields are awful for nvidia and amd.
 

Theonik

Member
Aug 14, 2010
24,245
0
810
kininarima.su
Nvidia squeezed like a 50 percent improvement out of maxwell just from a new architecture on the same 28nm node

Now there's a switch from 28nm to 16nm ff+ (and another architecture change, though we don't know how much different it is from maxwell)

It's not going to be 25 percent, 50 percent is probably lowballing it.
There is two things we know of that might bring this to doubt. First of all, Maxwell managed the boost it did by downgrading double precision float performance for the sake of more game oriented single precision float performance. We already know that Pascal is again increasing DP float performance so it should be a regression in that respect. They should still be able to compensate due to the shrink by adding more computational units.

Moreover, it is hard to do both a shrink and major optimisations and going from 28nm to 16nm might be really costly which means nVidia will try to maximize their financial gain first.

You think more cuda cores again in gp100?
I hope not, gimme pure gaming performance (also more cuda cores would mean them pushing more physx and flex stuff again, which always had shitty meaningless implementations in games like the clutter in borderlands 2)

And yeah who knows indeed:p

Right before a new gpu architecture comes out is always the most fun (speculation)
Like it was mentioned before, Cuda cores are nothing more than a way to express the computing performance of the card. More Cuda cores means more performance including more gaming performance.
 

Hasney

Member
Dec 6, 2012
46,467
2
0
Linus on LTT said he was recieving something GPU wise on the WAN show in the next couple of months, but has no idea what asit's under NDA on the side of whoever is sending it, so it could be the window for something new.
 

Theonik

Member
Aug 14, 2010
24,245
0
810
kininarima.su
Damn, you guys switch cards fast.
I think I'll keep my 980Ti till the 1080Ti. Don't expect the 1080 to be fast enough to justify the upgrade, then again resell value might be better if I go 980Ti -> 1080 ->1080Ti. I'm more interested in new features and connectivity as I want to move to high framerate and HDR by 2017.
 

ALLseeingEYE

Member
Mar 26, 2014
1,393
0
0
Vancouver B.C.
I think it will shake out.
1060 - 4GB ;slightly below 970 $229
1070 - 6GB ;slightly below 980ti/titanx $329
1080 - 8GB ;20-30% above 980ti/titanx $499

Fixed.

At least, that's what I'm expecting. I'm hearing Pascal will be a nice upgrade though, would be awesome to see the 1070 be more powerful than a 980ti. It would basically make 4K gaming much cheaper, and very viable.

I think I'll keep my 980Ti till the 1080Ti. Don't expect the 1080 to be fast enough to justify the upgrade, then again resell value might be better if I go 980Ti -> 1080 ->1080Ti. I'm more interested in new features and connectivity as I want to move to high framerate and HDR by 2017.

If you can sell your 980ti, or trade it straight up for a 1080, I'd personally go for it. A 20-30% power increase isn't insignificant. It could basically be the difference between running modern games at max settings at either 30-40fps on your 980ti and running them at a solid 60fps on the 1080. Hopefully they release the 1080 ti around the same time though. Can't wait to check out these cards.
 

mosdl

Member
Dec 3, 2014
448
0
0
Is there any chance the new cards will do a similar thing as the 970 VRAM situation (that slow .5 gigs)?
 

jax

Banned
Jun 2, 2013
7,968
0
0
Ugh. Guess that means I'm sticking with the 770 for the Vive until this drops. This wait is gonna suck. I wish the 980 and 970 would go on sale or something? Ugh.
 

Theonik

Member
Aug 14, 2010
24,245
0
810
kininarima.su
If you can sell your 980ti, or trade it straight up for a 1080, I'd personally go for it. A 20-30% power increase isn't insignificant. It could basically be the difference between running modern games at max settings at either 30-40fps on your 980ti and running them at a solid 60fps on the 1080. Hopefully they release the 1080 ti around the same time though. Can't wait to check out these cards.
It's not like I won't be tempted. It all depends on timing and what the 1080 enables me to do that I can't do with my 980Ti. A 1080Ti or a pascal Titan will probably not be out till late this year but probably early next year. I don't expect the 1080 is going to really matter to me till then. By the time my 980Ti feels old and I feel like I need the performance the replacement I want will be out, and what I want in terms of features will not be available in terms of monitors till around when the card I want is.

So really I don't see 10-20% pushing me over the edge right now.

Is there any chance the new cards will do a similar thing as the 970 VRAM situation (that slow .5 gigs)?
There is no evidence either way. I would imagine nVidia wouldn't want this clusterfuck again. Moreover there is something else to determine first.
What will the prices be and how gimped would the 1070 be. The problem with the 970 was that it was too much value and would canibalise the 980. nVidia decided to do this to differentiate the two.
The x70 and x80 tiers are usually more strongly separated, the 970 is an outlier, unless the 1070 is another value beast nVidia won't bother doing that.
 

neurosisxeno

Member
Dec 2, 2013
8,143
0
440
Vermont
The x70 and x80 tiers are usually more strongly separated, the 970 is an outlier, unless the 1070 is another value beast nVidia won't bother doing that.

Since Kepler they have been really close. I bought a 670 because a factory OCed 670 at launch was on par with a 680, but was $200 cheaper. The 970 was a smaller gap true, but it's not like the difference between the 2 Kepler series' was massive by comparison, just slightly wider. The difference between the 570 and 580 was pretty large though.
 

Theonik

Member
Aug 14, 2010
24,245
0
810
kininarima.su
Since Kepler they have been really close. I bought a 670 because a factory OCed 670 at launch was on par with a 680, but was $200 cheaper. The 970 was a smaller gap true, but it's not like the difference between the 2 Kepler series' was massive by comparison, just slightly wider. The difference between the 570 and 580 was pretty large though.
Ultimately this is exactly what nVidia wants to avoid. They don't want to straight out do that though As in, they can make money if lots of people buy the x70 part but they lose money when people buy the 970 over the 980 because they are too close. So nVidia wants to control performance and pricing across tiers. Where they place the balance is entirely on them.
 
Dec 11, 2010
34,632
2
0
I don't think Nvidia is the generous type to offer 980ti power at 970 level just because they can. They will if AMD is putting that sort of power at that sort of price, but that's about it.

Nvidia is happy to offer huge increases if they believe that bigger increases are around the corner. Even in the absense of competition, if it is "easy" for them to make significantly better processors, they will be happy to offer those increases since it incentivises upgrade. The new normal of holding back the "Big" chip until 8-12 months after the "little" chip launches is based on the slowing of die shrinks and the increased difficulties in performance increases, generally.
 

neurosisxeno

Member
Dec 2, 2013
8,143
0
440
Vermont
Nvidia is happy to offer huge increases if they believe that bigger increases are around the corner. Even in the absense of competition, if it is "easy" for them to make significantly better processors, they will be happy to offer those increases since it incentivises upgrade. The new normal of holding back the "Big" chip until 8-12 months after the "little" chip launches is based on the slowing of die shrinks and the increased difficulties in performance increases, generally.

Basically this. 20nm (or was it 22/24nm?) not panning out resulted in an extra cycle and a half where we got more 28nm chips while they readied 16/14nm to be the next process shrink. This forced Nvidia to essentially section up their offerings. Maxwell was slated to be the first 16nm GPU Nvidia offered, but because of delays from TSMC and Samsung they were forced to refashion it as a 28nm GPU (that's what the 750Ti was). If Nvidia could offer 50% improvements with each revision they absolutely would, because people would upgrade more frequently, but low yields and slowed implementation from their partners has forced them to slow their progress.
 

Odrion

Banned
Aug 31, 2005
19,688
1
1,130
So when would be the suppose deadline to sell off your Maxwell cards before they drop in value?
 

EscapeArtist

Banned
Dec 18, 2014
686
147
445
Midwest
I legitimately put a 970 in my Amazon cart today because I need a GPU upgrade (I have a 760 4gb right now) but never actually checked out because I feel like I should wait to see what the next generation of GPUs offer. As much as I hate to I should probably just hold off until they announce release dates and pricing.
 

kaskade

Member
Oct 21, 2007
9,325
0
0
I was going to buy a 970 for a vive. I ended up selling my old card a few days ago so now I'm just running my PC from my processor. Anyway, I was thinking of waiting until my vive ships before I buy a card so I can potentially do that step up program from EVGA if something comes out within the 90 days. Are we thinking these will be out sometime in the summer?
 
Apr 8, 2011
10,630
6
0
I was going to buy a 970 for a vive. I ended up selling my old card a few days ago so now I'm just running my PC from my processor. Anyway, I was thinking of waiting until my vive ships before I buy a card so I can potentially do that step up program from EVGA if something comes out within the 90 days. Are we thinking these will be out sometime in the summer?

Seems like a gamble that I wouldn't take.
 

dr_rus

Member
May 3, 2007
10,983
0
1,350
Moscow, Russia
Is there any chance the new cards will do a similar thing as the 970 VRAM situation (that slow .5 gigs)?

Will depend on how the memory will be configured on these both in the chip and on the board. Considering that we're likely looking at 8GB cards with 8GBs comprised of 8 1GB chips I'd say this is kinda unlikely as they'll be able to cut down more cost if they'll simply remove one 1GB chip completely and leave only 7GBs of VRAM on the card. But it's even more probable that we'll get the full 8GBs on all GP104 boards after the 970 fiasco.