[VC]NVIDIA GeForce GTX 980, GTX 980 SLI, GTX 970, 3DMark performance

[Cloudfire777]

Here is a teaser that was just posted.
I think it is the GTX 970

 

[tviceman]

And on the other side of the coin, those that used the 470/480 sit here and pretend like a few watts suddenly mean the difference between buying a cheaper and faster card when their spons..., I mean brand is slightly more efficient. :thumbsdown:

I've personally moved on from the 1990's 2 foot tall case enclosures with a half dozen fans. SFF can fit plenty of power today and most do reasonable jobs of cooling without having box-fan type airflow. In this regard, I appreciate perf/watt more so now than I did 5 years ago. I don't want to see either company push the TDP envelope further. I'd like to see them get better engineering and reign in TDP a little bit on the higher end, if for nothing else, to leave room for on-air overclocking.

It doesn't matter the issue, the goalpost shifting sucks.

Yep.

I care about performance and price, followed by the other criteria (noise etc.). Everyone has an opinion, it's just annoying when they keep "changing" them to follow the latest trend (and the non stop trying to hype it).

Performance and price are the two most important aspects to me as well. But I'll HAPPILY spend a few more dollars (within reason) for the quieter, lower power using card at equal performance.

Anyways, hopefully the 980 has the performance we crave and a reasonable price. I'm on the fence whether I think it might have some (compression) magic and be another 30% boost over the 780 ti, but on the other hand it could be slower especially at high resolution since there are too many pixels to compress.

I think most people here are going to complain that it's not fast enough, even if/when it's faster than anything else out. That is what people here do. Hopefully it's priced well though.

There is talk on beyond 3D that maxwell is partially tile-based rendering, explaining how it gets away with the same performance at lower bandwidth.

 

[Termie]

Actually, what's interesting about the graph above, if it's true, are the following two observations:

(1) the 970 does not catch up to the 980 with overclocking. Rare situation between number 1 and 2 in Nvidia's lineups

(2) the new cards are 5-10% faster than the cards they replace (780/780Ti).

On a separate note, regarding the discussion of performance/Watt, I can tell you one instance where power absolutely matters - Small Form Factor. Running my 780 Ti in a mini-ITX Silverstone case was a challenge. It's not just power, it's heat. The excess heat from the 290/X and 780 Ti cards is pretty significant in my opinion and makes cooling them in a small case very difficult (and loud).

 

[Face2Face]

Here is a teaser that was just posted.
I think it is the GTX 970

Looks like

Unknown 1 = GTX 970 1382/2003
Unknown 2 = GTX 980 1376/2028

(1) the 970 does not catch up to the 980 with overclocking. Rare situation between number 1 and 2 in Nvidia's lineups

Keep in mind, they are practically at the same clock speeds so its expected the GTX 970 wouldn't catch up.

 

[FatherMurphy]

Where was this graph posted?

 

[Enigmoid]

Looks like

Unknown 1 = GTX 970 1382/2003
Unknown 2 = GTX 980 1376/2028



Keep in mind, they are practically at the same clock speeds so its expected the GTX 970 wouldn't catch up.

Yep, but since they both have identical bandwidth it looks like the chip isn't limited by the 256 bit bus.

 

[tviceman]

If a maxed out 980 can't beat a maxed out 780 TI, that is disappointing.

 

[nvgpu]

http://videocardz.com/52339/nvidia-geforce-gtx-980-has-2048-cuda-cores

2048 cores, 16 SMM, not 15 SMM as many had thought. ~175W TDP.

Also GTX 970 is 148W TDP.

 

[n0x1ous]

If a maxed out 980 can't beat a maxed out 780 TI, that is disappointing.

Be realistic -

same node, smaller die, less cores, less bandwidth, less ROPS.

All the improvement can be is architectural. Nvidia can't break the laws of physics. If it is roughly even with 780ti its impressive work on Nvidia's part. We just need a node shrink very badly to see the type of generational improvement we are used to.

Judging by iphone 6 pre-orders, I dont expect there to be any 20nm wafters available to anyone but Apple anytime soon.

 

[FatherMurphy]

Well, that explains the unexpectedly large difference in projected performance between 970 and 980....

If 980 is the full die, and the 970 has 3 SMMs disabled, I imagine we'll be seeing something in between (ala 975 or 970ti) eventually, right?

 

[tviceman]

Be realistic -

same node, smaller die, less cores, less bandwidth, less ROPs.

Based on what GM107 delivered in all areas, I am being quite realistic in my expectations.

 

[n0x1ous]

Based on what GM107 delivered in all areas, I am being quite realistic in my expectations.

GM107 is a bigger die than the one it replaces.

GM204 is bigger than it what it technically replaces GK104 but everyone seems to be comparing to the much larger GK110 instead.

 

[Gloomy]

Extremely disappointing progress-wise. A 970 at almost 1400Mhz can't match a 1200Mhz 290. But to be expected, with 32 ROPs + 256bit @ 1440p + 4xAA, it's surprisingly decent when that's considered. Maybe it'll do better at 1080p.

More interested in these chips making their way into laptops. As far as desktop performance is concerned... nah, not appealing at all IMO. But then I already have a 290, so. :|

I read a prediction elsewhere that Nvidia was going to phase out the 780 and 780Ti, replace them with the 980/970 and leave the Titan line intact for a while. As that would be the best way to maximize margins while competing with Hawaii (which is after all, a much smaller chip than GK110). I shrugged it off, but I think it's understandable from Nvidia's standpoint. And maybe they will be more willing to compete price-wise with these chips, which is good.

Honestly I found it weird that the 780 and 780Ti was priced so far out of line. IMO so far that it can't really be explained by "Nvidia tax". I mean, the 780's just recently coming down below 500 dollars. Most of the time since the 290 launch it was 150 dollars above 290 MSRP. And I don't think it's coming down in price now because they suddenly want to compete with the 290. I think it's because it's being phased out.

Anyway, this looks like decent enough cards as long as you don't own anything faster than a 290 already.

 

[AtenRa]

If those numbers are true this is very nice performance from those Chips. The naming is not the best, much like AMDs Tonga, but the performance from the die size is very good.

Im really looking forward for the official reviews.

 

[96Firebird]

I mean, the 780's just recently coming down below 500 dollars. Most of the time since the 290 launch it was 150 dollars above 290 MSRP. And I don't think it's coming down in price now because they suddenly want to compete with the 290. I think it's because it's being phased out.

The 780 can be had for $400 right now, the 780 Ti is still priced too high. And the 290 wasn't even near its MSRP for a while after its release due to the mining crazy, so the 780's price didn't need to be lowered.

But its drop now is almost certainly because the new cards are arriving soon.

 

[Termie]

Looks like

Unknown 1 = GTX 970 1382/2003
Unknown 2 = GTX 980 1376/2028



Keep in mind, they are practically at the same clock speeds so its expected the GTX 970 wouldn't catch up.

No, I mean the OC'd 970 doesn't beat the stock 980. As far as I can remember, that's never happened in an Nvidia lineup. An OC'd 670 beat the 680, an OC'd 780 beat the 780 Ti.

Anyway, for all we know, this graph could be fake, but generally speaking, it's probably a pretty good approximation of performance (in other words, we could have all made it up).

Also, I'll just note that those are absolutely monstrous overclocks on the 290/X and 780 /Ti. Not at all common without ultra-high-end models with voltage tweaking. I assume the 970/980 numbers, if real, were based on overclocks of reference models without a lot of tools available to tweak them.

 

[Face2Face]

No, I mean the OC'd 970 doesn't beat the stock 980. As far as I can remember, that's never happened in an Nvidia lineup. An OC'd 670 beat the 680, an OC'd 780 beat the 780 Ti.

Anyway, for all we know, this graph could be fake, but generally speaking, it's probably a pretty good approximation of performance (in other words, we could have all made it up).

Also, I'll just note that those are absolutely monstrous overclocks on the 290/X and 780 /Ti. Not at all common without ultra-high-end models with voltage tweaking. I assume the 970/980 numbers, if real, were based on overclocks of reference models without a lot of tools available to tweak them.

Gotcha!

Looks like the overclocked benches were done with non-reference cards, but like you said, the graph should be taken with a grain until we get some real reviews

 

[Gloomy]

The 780 can be had for $400 right now, the 780 Ti is still priced too high. And the 290 wasn't even near its MSRP for a while after its release due to the mining crazy, so the 780's price didn't need to be lowered.

But its drop now is almost certainly because the new cards are arriving soon.

You're right, that was probably a big part of it.

 

[tviceman]

GM107 is a bigger die than the one it replaces.

GM204 is bigger than it what it technically replaces GK104 but everyone seems to be comparing to the much larger GK110 instead.

Doesn't change the metrics I quoted. If Gm204 has the same perf/w and perf/mm2 then at ~400mm2 it should beat a 780 TI by 10-15% at gtx770 power consumption.

 

[Silverforce11]

I's hypocritically funny that when Fermi 1.0 came out, it was blasted nonstop and continuously for it's inferior perf/w and temperatures. People lambasted how a gtx480 would heat an entire room after an hour or two of gaming. The complaints continued until gtx580 upped the performance, dropped the thermals, and slightly lowered power consumption.

Absolutely none of those same people complain about the exhaust temps or power consumption of the r290x and r290. People will forever continue to spin and justify a particular brand or vendor no matter the situation.

This is untrue. I complaint about R290/X regarding its ridiculous temps and power use AND noise. I bashed the reference design almost as much as blackened.

 

[CakeMonster]

Nothing is surprising here. They will have huge margins on this, being much cheaper to produce. They are milking us for cash at this point.

What we can only hope for is a real high performance version, which I'll guess will be a 384bit 980 "Ti" or "Titan" with 6Gb a bit later. But we all know what kind of price tag it will have.

And before you accuse me of whining about NV, I would expect AMD to do the exact same thing quite soon. Keep costs extremely low, just grab the money for a miniscule performance increase (if any).

20nm can't get here soon enough....

 

[Silverforce11]

Gibbo over at OCUk already pretty much revealed that the 980 will be ~780ti, while the 970 will be ~780.

This is still a good result on the same node, with less power use, so they got both good perf/w as well as perf/mm2 gains. Definitely Maxwell is a GREAT architecture.

 

[poohbear]

Nothing is surprising here. They will have huge margins on this, being much cheaper to produce. They are milking us for cash at this point.

What we can only hope for is a real high performance version, which I'll guess will be a 384bit 980 "Ti" or "Titan" with 6Gb a bit later. But we all know what kind of price tag it will have.

And before you accuse me of whining about NV, I would expect AMD to do the exact same thing quite soon. Keep costs extremely low, just grab the money for a miniscule performance increase (if any).

20nm can't get here soon enough....[/QUOTE9]
For those of us upgrading from a 5××× or even 6××× series card its not a miniscule increase in performance. Not everyone here upgrades every gen.

 

[96Firebird]

Definitely Maxwell is a GREAT architecture.

This makes me think, why couldn't Maxwell had come out instead of Kepler? Nothing has really changed since Kepler was released, except 28nm has gotten more mature. I realized these architectures take years to develop, but it seems like Maxwell is just so much better than Kepler on perf/watt considering they are both on the same node. They are both running DX11, VRAM hasn't changed much besides memory speed.

Guess I need to read up on why graphics architectures get more advanced over time.

 

[Genx87]

Maxwell is clearly designed for a smaller process. That is why Maxwell wasnt released when Kepler came out.