Kitguru : Nvidia to release three GeForce GTX 800 graphics cards this October

[ShintaiDK]

I can't see how anyone could possibly think 880 won't beat 780Ti. Nvidia is not stupid or incompetent enough for that.

Why would it have to beat it?

I am however quite sure it will beat it in price/performance and performance/watt.

It seems that it may just match a 780ti, give or take.

 

[master_shake_]

I can't see how anyone could possibly think 880 won't beat 780Ti. Nvidia is not stupid or incompetent enough for that.

the 880 will be faster the question is, will the 880 be a fast enough to make upgrading an easy decision.

 

[Grooveriding]

I think it's reasonable to expect as much. 880 slightly slower/parity/slightly faster than 780ti, but, cheaper and using significantly less power to deliver that performance. Could be another 8800GTX to 9800GTX situation where it is a little faster at lower resolutions but slower at high resolutions.

The 880ti nomenclature is still left available for a possible future part down the line. The more rumours come out though, the more it seems 880 will not be an upgrade for 780ti or Titan Black.

 

[CrazyElf]

Sounds like 2560-3200 CUDA cores were just early wishful thinking rumors. Well there goes my prediction that this card will beat 780Ti by 30-35% on avg. The question now is what is NV going to use then to fill in the performance between a 780Ti and GM200/210? If 880 is a 680 successor then how expensive will 780/780Ti successors be? $650 for 780's and $800 for 780Ti's?

So we can safety say that the 430mm^2 rumors are probably false? How big a die are we looking at?


I mean the best case scenario is that we get another 550mm^2 card, but I get the feeling that is not happening until 2015. Then it will be Titan priced - worse if AMD's next card is a turkey. We gotta hope AMD has something. The real variable at this point is, what does AMD have?

In theory, yields ought to be mature on this 28nm process, so we ought to get quite a bit of performance if we get a "giant" 550mm^2 card. How much faster would such a card be compared to the 780Ti?

 

[blackened23]

All due respect you dudes are nuts. (nicest possible joking way) The 870 is outscoring a stock/non overclocked CPU/GPU with GTX 780. Don't know what the heck some are talking about. Bringing up overclocking in a stock comparison. Huh? You can theorize all you want but the basis of comparison is a stock CPU and a stock GPU. And the 870 is faster than the 780 here, if the leak is real.

We're going by stock CPU/GPU scores as a basis of comparison, and the 870 is outscoring the 780. The stock 780 scores around 4500 in 3dmark 11 extreme. The 880 will be faster than the 780ti, because the stock score of the 780ti is around 5000-5100 extreme in 3dmark11. The leaked 880 score was more than 1000 points higher than that in 3dmark 11 extreme. 780ti is not that much faster than the plain 780, it's only 500 points faster in 3dmark 11 e.

And NV would not do a new gen of GPUs if it were not faster. There's no point if it isn't faster. If these leaks are real the 870 is significantly faster than the 770, and the 880 will be significantly faster than the 780ti. 1000 point spread in 3dmark 11 extreme is about the difference between the 780 and 770. So that same difference in speed will apply to the 870 to 880: a 1000 point spread in 3dmark 11 extreme is a big gap. If that is true, it WILL be quite a bit faster than the 780ti. We'll see though next month. (if it is next month)

 

[blackened23]

So no 3DMark 11 links?

The GPU-Z screenshot shows 7GHz Samsung memory.

Guru3d does 3dmark 11 benchmarks on all reviews, you can see their stock 780 and 780ti reviews. 780 is around 4500 extreme and 780ti is around 5100 extreme.

The leaked 870 is higher than the GTX 780 and the leaked 880 is faster than the 780ti. The 880 scored more than 1k points higher than the 780ti IIRC. In fact, guru3d with a stock CPU/GPU had a 780ti score 5114. The stock CPU/GPU 880 in the leak scored 6200ish. It will be faster than the 780ti, but how that translates in game performance, we'll see.

That said, I think 28nm is limiting the truly amazing gains that would come with a node change with a higher transistor count. AMD anD NV both could realize some amazing performance increases with an increased transistor count, but 28nm is just limited in that respect. That's just unfortunate but it is what it is I suppose - I firmly believe 880 will be at least 20-30% better than the 780ti, but we're at the limits of 28nm past that in all liklihood. Newer nodes cannot get here soon enough IMO.

 

[CrazyElf]

That said, I think 28nm is limiting the truly amazing gains that would come with a node change with a higher transistor count. AMD anD NV both could realize some amazing performance increases with an increased transistor count, but 28nm is just limited in that respect. That's just unfortunate but it is what it is I suppose - I firmly believe 880 will be at least 20-30% better than the 780ti, but we're at the limits of 28nm past that in all liklihood. Newer nodes cannot get here soon enough IMO.

Reportedly 20nm TSMC is only 10% faster than the 28nm at high power. That's probably why they abandoned it.

TSMC claimed that 20nm was 30% faster than 28nm though for their low power process, so I guess that's why they proceeded with that. Combine that with the higher volumes that Apple, Qualcomm, and the others offer.

Like it or not, we're at the point where Moore's Law is dying. Future processes, after the massive capital investments and complexity may prove to be more costly on a per transistor basis than prior nodes, forcing tradeoffs between performance and cost. Supposedly there's going to be a 16nm FinFET process at TSMC for high power. GF I think is partnering up with Samsung (gotta look it up and verify later on), for that too.

The point is, we're reaching the point where well ... where do we go from here? Beyond 16nm, they're going to have to do insane stuff. Double patterning will probably be needed everywhere for 16nm. What happens after that? For 10nm, it's going to be insane, something like quadruple patterning? That pushes prices up even higher. EUV doesn't look like it's happening soon (ex: in the next 5 years). Neither do the 450mm wide wafers. As far as exponential process nodes, this may be the point where it ends. We'll have one-offs in the future though, but they can never substitute for exponential scaling.

That's my guess, unless something truly revolutionary happens.

 

[tollingalong]

If the initial leaks of 870 and 880 are true we're going to see small stock performance increases; 10-15% better performance in new cards. If new Maxwells are anything like the old ones get ready to get another 10-15% overclock on top of that.

I think the main Maxwell boost will be when it's OCed. Stock is just a tiny power saver.

 

[tollingalong]

The point is, we're reaching the point where well ... where do we go from here? Beyond 16nm, they're going to have to do insane stuff. Double patterning will probably be needed everywhere for 16nm. What happens after that? For 10nm, it's going to be insane, something like quadruple patterning? That pushes prices up even higher. EUV doesn't look like it's happening soon (ex: in the next 5 years). Neither do the 450mm wide wafers. As far as exponential process nodes, this may be the point where it ends. We'll have one-offs in the future though, but they can never substitute for exponential scaling.

That's my guess, unless something truly revolutionary happens.

This was a known upcoming problem for years and it's been solved on a POC scale. We have a physics bottleneck. Intel already has optical computers. Folks have worked on quantum networks. The future is available but it's simply the matter of costs.

 

[MathMan]

This was a known upcoming problem for years and it's been solved on a POC scale. We have a physics bottleneck. Intel already has optical computers. Folks have worked on quantum networks. The future is available but it's simply the matter of costs.

I'm afraid that's a bit optimistic: it's one thing to have a proof of concept that shows a couple of equivalent transistors. It's another to have 10B of them. At this time, that's not a matter of cost, it's a matter of not yet being possible at all.

 

[alcoholbob]

If the initial leaks of 870 and 880 are true we're going to see small stock performance increases; 10-15% better performance in new cards. If new Maxwells are anything like the old ones get ready to get another 10-15% overclock on top of that.

I think the main Maxwell boost will be when it's OCed. Stock is just a tiny power saver.

You can do 15-20% on keplar cards pretty easy so. So comparing overclock to overclock its still only 10-15% faster. So were still stuck running at least two of these cards for anything over 1080p.

 

[RussianSensation]

All due respect you dudes are nuts. (nicest possible joking way) The 870 is outscoring a stock/non overclocked CPU/GPU with GTX 780. Don't know what the heck some are talking about. Bringing up overclocking in a stock comparison. Huh? You can theorize all you want but the basis of comparison is a stock CPU and a stock GPU. And the 870 is faster than the 780 here, if the leak is real.

I agree that comparing factory overclocked 780/780Ti to a stock 870/880 makes little sense. At the same time those preliminary scores are somewhat underwhelming so far. In the last 2 weeks I went from being pretty excited about these cards to almost not caring anymore. Why?

AMD path:
Moving from HD5870 $370 card in Sept 2009 to R9 290X ($550) in December 2013 meant that in 3 years, we got almost 3x the performance increase:
http://www.computerbase.de/2013-12/grafikkarten-2013-vergleich/10/

NV path:
Moving from March 2010 GTX480 $500 (using 7850 as a reference) to May 2013 GTX780 $650 meant that in 3 years, we got 2x the performance increase.
http://www.computerbase.de/2013-05/nvidia-geforce-gtx-780-test/3/

I personally went from HD4890 which I got in August 2009 to HD7970 which I purchased in June 2012, which is also at least a 2.5x increase in performance in less than 3 years and triple the VRAM (!), and nearly 3x the performance when looking at 7970 OC vs. 4890 OC. You would get identical results if you upgraded from GTX260 216/275 to 680 instead of 4890 to 7970; so really it doesn't matter if you mixed and matched AMD/NV upgrades along the way:
http://forums.anandtech.com/showthread.php?t=2298406

vs. Now we are supposed to get excited about buying a $400-450 880 that's maybe 5-10% faster than 780Ti or in effect barely 55% faster than 7970Ghz but nearly 3 years later? That's seriously disappointing. Ok so they focused on lowering the power consumption and are constrained by the 28nm node -- I guess more waiting for 20nm GM200 then. I am now slowly becoming convinced that the 880 will not meet my historical upgrade criteria of being 1.75-2x faster than my existing card. After being used to getting 75%-100%+ performance increases with most of my GPU upgrades every 2.5-3 years, I no longer find a next gen card that's 50% faster for $400-450 as anything but disappointing, regardless if it's coming from AMD or NV. Personally, it's just not good enough after nearly 3 years of waiting for a serious upgrade.

Maybe NV will surprise with this one and here is hoping that the leaks are for lower end 860Ti cards. I think I'd rather spend $650 on a GM200 that's 50-60% faster than 780Ti rather than $400-450 on a card that's only 5-10% faster. Taking emotions out of the equation, even the mathematics back this up:

Upgrading to 880:
Base case = R9 280X/7970Ghz/680 OC/770 ~ 100%
880 => project 10% faster than 780Ti = 1.42x1.10 = 156% for $400 (56% faster or paying $7.14 for each 1% increase in performance over Base case)

Waiting for GM200:
GM200 => project 50% faster than 780Ti = 1.42x1.50 = 213% for $650 (113% faster or paying $5.75 for each 1% increase in performance over Base case) <<<winner>>>
http://www.computerbase.de/2014-08/grafikkarten-2014-im-vergleich-sommerausgabe/2/

If 880 is not even faster than 780Ti, then it gets even worse!

 

[sontin]

Your comparision doesnt make sense. The GTX480 is more future proof than the 5870. So the difference is much smaller than between the 5870 and 290x.

On the other hand: AMD went from 180W to >300W while nVidia stayed at ~260W.

 

[Fastx]

RS, any opinion on how long after the 880 release would it be before the 880Ti is released?

No plans/thoughts to get 880Ti unless pricing is good, but I am curious per above question? It's funny I almost went with the 780 ($650) when it was first released but after reading one of your post on pricing back (at $650) then, I decided not to. I CF my 7950 instead and never regretted it, and only sold due to the mining craze for a healthy profit.

Thanks for that post that made me decide against the 780 at $650.00!

 

[blackened23]

I think (due to the leaks, if accurate) it will be more than the 5-10% you suggest RS. There is 1500 point spread in 3dmark 11 extreme between the 870 and 880 per the leaks. And a 1000 point spread between the 780ti and the 880 leaked score. As a point of reference, there's around a 1200-1300 point spread between the GTX 770 and 780 in that same test. And the 780 is obviously muich faster.. That alone suggests the card (880) will be very much faster than the 870, and the 870 is faster than the GTX 780 per the leak.. This all depends on accurate leaks, but we've had accurate 3dmark11 leaks in the past with prior GPUs.

I guess we'll wait and see, though. From the leaks i'm thinking 25% or so. But if that's the case I highly doubt a 400$ price tag. I'm thinking 550ish. Just my guesstimate - i'm more curious about game performance. I dunno. At a minimum i'd see a 480 > 580 type of difference (15%) between the 780ti and the 880 with a similar price (450-500$), or 25% or more performance over the 780ti with a 550-600$ price. Just my random thoughts and guesstimates. Who knows where it will actually land..

 

[blackened23]

http://www.fudzilla.com/home/item/35488-nvidia-maxwell-geforce-800-comes-in-september

Now the company is ready for the Geforce 800 series and it looks like this chip will be power efficient and powerful, too. For Nvidia it was all about getting the 20nm manufacturing process to the point where it makes financial sense, while gaining a performance uplift substantial to excite customers and get them to buy new cards.

Fudzilla saying it (all of the 800 series) is 20nm? I'm not sure if they're speaking strictly of the mobile Maxwell gpu's for next gen or the desktop, but i'm thinking...no way it's true..

 

[n0x1ous]

GM200 => project 50% faster than 780Ti

Is this assuming 28nm still? I don't see them making this chip on 28nm because they don't have the ability to make the die much bigger. Assuming 20nm and traditional Nvidia flagship performance upgrades between nodes, I would venture more like 75%+ from 780ti

 

[tviceman]

AMD path:
Moving from HD5870 $370 card in Sept 2009 to R9 290X ($550) in December 2013 meant that in 3 years, we got almost 3x the performance increase:
http://www.computerbase.de/2013-12/grafikkarten-2013-vergleich/10/

NV path:
Moving from March 2010 GTX480 $500 (using 7850 as a reference) to May 2013 GTX780 $650 meant that in 3 years, we got 2x the performance increase.
http://www.computerbase.de/2013-05/nvidia-geforce-gtx-780-test/3/

That's an interesting comparison there.

The time frame for AMD's 5870 to R9 290x is more than 4 years. The time frame from GTX480 to GTX780 is a full year less. Also, AMD's power usage has shot way up in those comparisons, while Nvidia's actually went down.

I know you're comparing "flagship" products prices, but AMD went from having a significant lead time in architecture advancement and way more efficient numbers, to actually falling behind in efficiency and even being passed up architecture replacements. Does that trend stop or reverse? If not, are we looking at AMD entirely relegating (and accepting) themselves to second fiddle in the GPU market in 18 months?

 

[Grooveriding]

I agree that comparing factory overclocked 780/780Ti to a stock 870/880 makes little sense. At the same time those preliminary scores are somewhat underwhelming so far. In the last 2 weeks I went from being pretty excited about these cards to almost not caring anymore. Why?

AMD path:
Moving from HD5870 $370 card in Sept 2009 to R9 290X ($550) in December 2013 meant that in 3 years, we got almost 3x the performance increase:
http://www.computerbase.de/2013-12/grafikkarten-2013-vergleich/10/

NV path:
Moving from March 2010 GTX480 $500 (using 7850 as a reference) to May 2013 GTX780 $650 meant that in 3 years, we got 2x the performance increase.
http://www.computerbase.de/2013-05/nvidia-geforce-gtx-780-test/3/

I personally went from HD4890 which I got in August 2009 to HD7970 which I purchased in June 2012, which is also at least a 2.5x increase in performance in less than 3 years and triple the VRAM (!), and nearly 3x the performance when looking at 7970 OC vs. 4890 OC. You would get identical results if you upgraded from GTX260 216/275 to 680 instead of 4890 to 7970; so really it doesn't matter if you mixed and matched AMD/NV upgrades along the way:
http://forums.anandtech.com/showthread.php?t=2298406

vs. Now we are supposed to get excited about buying a $400-450 880 that's maybe 5-10% faster than 780Ti or in effect barely 55% faster than 7970Ghz but nearly 3 years later? That's seriously disappointing. Ok so they focused on lowering the power consumption and are constrained by the 28nm node -- I guess more waiting for 20nm GM200 then. I am now slowly becoming convinced that the 880 will not meet my historical upgrade criteria of being 1.75-2x faster than my existing card. After being used to getting 75%-100%+ performance increases with most of my GPU upgrades every 2.5-3 years, I no longer find a next gen card that's 50% faster for $400-450 as anything but disappointing, regardless if it's coming from AMD or NV. Personally, it's just not good enough after nearly 3 years of waiting for a serious upgrade.

Maybe NV will surprise with this one and here is hoping that the leaks are for lower end 860Ti cards. I think I'd rather spend $650 on a GM200 that's 50-60% faster than 780Ti rather than $400-450 on a card that's only 5-10% faster. Taking emotions out of the equation, even the mathematics back this up:

Upgrading to 880:
Base case = R9 280X/7970Ghz/680 OC/770 ~ 100%
880 => project 10% faster than 780Ti = 1.42x1.10 = 156% for $400 (56% faster or paying $7.14 for each 1% increase in performance over Base case)

Waiting for GM200:
GM200 => project 50% faster than 780Ti = 1.42x1.50 = 213% for $650 (113% faster or paying $5.75 for each 1% increase in performance over Base case) <<<winner>>>
http://www.computerbase.de/2014-08/grafikkarten-2014-im-vergleich-sommerausgabe/2/

If 880 is not even faster than 780Ti, then it gets even worse!

Yeah I think at best we are looking at maybe 10% faster for the 880 over 780ti, possibly not even that. Very likely it will be slower at 4K and 2560x1440/2560x1600.

The marketing will push efficiency and price and 1080p benchmarks. That coolaler leak, which are usually genuine, really puts a very bad look on performance.

 

[sontin]

nVidia announced a new K5200 card with 2304 cores and 150W.
So, is this the first sign of GM204?
http://www.nvidia.com/object/quadro-desktop-gpus.html

 

[mv2devnull]

No. Only the K620 and K2200 are GM*.

 

[96Firebird]

No. Only the K620 and K2200 are GM*.

What could it be then? I don't know of a Kepler chip with 2304 cores.

Whoops, the 780 does. But then how did they get 8GB with 384-bit bus.

 

[n0x1ous]

96Firebird: 780 has 2304 cores

Have to agree with Sontin. 2304 initially made me think binned 780, but 150w would be too low even for a binned 780 and 8GB suggest it couldnt be a 384-bit bus so it smells like GM204

 

[exar333]

Yeah I think at best we are looking at maybe 10% faster for the 880 over 780ti, possibly not even that. Very likely it will be slower at 4K and 2560x1440/2560x1600.

The marketing will push efficiency and price and 1080p benchmarks. That coolaler leak, which are usually genuine, really puts a very bad look on performance.

I am not disagreeing with you (in fact, I think you are spot-on) but this boggles my mind. With everyone going to higher-resolution displays on Desktop AND mobile, this seems very short-sighted to cut bandwidth. If this is true, my next card will most likely be AMD.

 

[n0x1ous]

I am not disagreeing with you (in fact, I think you are spot-on) but this boggles my mind. With everyone going to higher-resolution displays on Desktop AND mobile, this seems very short-sighted to cut bandwidth. If this is true, my next card will most likely be AMD.

Its already been shown that maxwell can punch above its weight class with lower bandwidth due to the higher amount of cache. I wouldnt make this assumption yet....