Possible GTX 880 benched in 3DMark Firestrike Extreme!

[ocre]

<snip>
I see double standards all over our forum. People bashed 7970 for being $550 when 580 1.5GB was $450 and 580 3GB was $550 and 7970 overclocked beat 580 overclocked by 40-80%:
http://m.hardocp.com/article/2012/0...verclocking_performance_review/7#.U9NdwYkazCQ

I think 20-30% faster for a next generation architecture priced at $600 is not that impressive. If NV priced it like 680 at $499, that would be a lot more ground breaking. I mean using 780Ti's price as some justification for 880 low performance increase is just as erroneous as using 580 3GB's overinflated price to justify 7970 at $550.

So what is it? Was 7970 OC a failure bringing 45-80% over 580 OC but 880 is great being 20-30% faster for $550-600? It doesn't make any sense if we judge 880 with the same standards we judged all previous generation flagships, including 7970. People are just using 780Ti's ludicrous price to justify why the 880 is not underwhelming based in rumors. I feel like NV is not judged by the same standards on our forum as of late. Remember 480 and 580 had about $140-150 premiums over 5870/6970 for 18-20% more performance on average, but now 780Ti is $300-350 more expensive than after-market 290, which means its price is laughable at this point and not worth using as a reference point whatsoever.

Every single person who thought 7970 was disappointing should view 880 as an even bigger failure if it launches at $550-600 with only a 20-30% increase claiming to be a next gen "flagship". 7970 overclocked 25-30% too from 925 to 1150-1250mhz or a 24-35% headroom.

Yeah, like this one ^^

Your referring to the 7970 that a user had to overclock to the max and then justifying the price over a gtx580. At the same time smearing your hypothetical GTX880 for your guesstimate of a stock 20-30% performance increase.

Its wrong on every single imaginable way.

first, your willingness to forgive the HD7970's tiny 25% increase over the 580 because of its suggested overclocking head room. And because of that, (2nd) your in total support of the price premium AMD commanded over the gtx580.

So we have a double double standard. It is from you. And it really surprises me how much you attack others but are completely unaware of your own.

Double standard #1
You have insisted that people focus on the max overclocking performance of the original 7970 but you only care about the stock baseline performance of the GTX880.

Double standard #2
You use this to justify a price increase for a 7970 (a brand new series) while the rumors on the gtx880 suggest it will be cheaper than the 780ti

I really don understand how you can be so extremely lopsided

and disclaimer......
Please dont start turning this thread into "how to justify the original 7970 prices". I mean come on, overclocking results vary from person to person. I will always see it as a bonus and never will base it as my expected performance.
And lastly, the most important part -> I didnt care where AMD priced it.

I am never in the market for the highest end. I was more interested in the 7950 at the time but this is all old news that has nothing to do with the OP.

Just hope you can use some of those skills that let you see flaws in others on yourself. If you are gonna accuse others of double standards, make sure you dont have obvious ones flaring out in the exact same post.

 

[AtenRa]

Well if you ask me, i could see the GM204 at 780 or 780Ti performance with 770 power consumption. But 20%-30% higher performance than 780Ti ??? thats to much for a 400mm^2 at 28nm.

 

[Erenhardt]

Double standard #2
You use this to justify a price increase for a 7970 (a brand new series) while the rumors on the gtx880 suggest it will be cheaper than the 780ti

I really don understand how you can be so extremely lopsided

Here is a hint for you that you didn't catch earlier: Its requires quite an effort to release something even pricier than 780ti...

 

[Cloudfire777]

Well if you ask me, i could see the GM204 at 780 or 780Ti performance with 770 power consumption. But 20%-30% higher performance than 780Ti ??? thats to much for a 400mm^2 at 28nm.

GK107 118mm2
GM107 148mm2
Die increase: +25%
Performance: +80%
TDP: The same

GK104: 295mm2
GM204: 430mm2
Die increase: +45%
Performance: Depends on the TDP. But it should be a massive increase in performance over GK104 due to the die size increase vs the example above. Unless they aim for low TDP..
GTX 780 Ti: Roughly 50% faster than GTX 680. Do the math

 

[tviceman]

Well if you ask me, i could see the GM204 at 780 or 780Ti performance with 770 power consumption. But 20%-30% higher performance than 780Ti ??? thats to much for a 400mm^2 at 28nm.

Why? It won't have all the compute transistors, so that is saved space on die size, and GM107 outperforms GK107 by a conservative 70% with only a 25% increase in die size. If GM204 outperforms GK104 similarly in perf/mm^2, a 410mm^2 GM204 would = 1.9x the performance of GK104 (gtx 770).

That is exactly 30% faster than a gtx 780 TI. Looks very plausible to me.

EDIT: Cloudfire beat me to it. I was too quick to respond before reading through the thread.

 

[Erenhardt]

I expect (as always) GTX8** to offer better performance/$. Sadly, Nvidia have a track record of not delivering...

 

[toyota]

I expect (as always) GTX8** to offer better performance/$. Sadly, Nvidia have a track record of not delivering...

what does that even mean? Nvidia has always given better performance per dollar with a next gen card. AMD is the only company that gave LESS when they came out with the 7970 for 550 bucks which was 50% more than the 6970 launched at while giving only 40% more performance.

 

[Erenhardt]

what does that even mean? Nvidia has always given better performance per dollar with a next gen card. AMD is the only company that gave LESS when they came out with the 7970 for 550 bucks which was 50% more than the 6970 launched at while giving only 40% more performance.

reality check

 

[toyota]

reality check
http://tpucdn.com/reviews/NVIDIA/GeForce_GTX_580/images/perfdollar.gif

lol the 580 was not a next gen card and you know that. and it still gave the same performance per dollar at least.

EDIT: actually 580 was the SAME price as the 480 so in fact it DID offer more performance per dollar. they were looking at the current 480 selling price not the launch price.

 

[Keysplayr]

fuggeddaboudit

 

[el etro]

100% perf/watt divided by 65% perf/watt = 1.54, aka 54% more efficient. The gtx 650 TI is considered to be as efficient as the most efficient GCN or Kepler based products.

In techpowerup.

 

[AtenRa]

GK107 118mm2
GM107 148mm2
Die increase: +25%
Performance: +80%
TDP: The same

GK104: 295mm2
GM204: 430mm2
Die increase: +45%
Performance: Depends on the TDP. But it should be a massive increase in performance over GK104 due to the die size increase vs the example above. Unless they aim for low TDP..
GTX 780 Ti: Roughly 50% faster than GTX 680. Do the math

Why? It won't have all the compute transistors, so that is saved space on die size, and GM107 outperforms GK107 by a conservative 70% with only a 25% increase in die size. If GM204 outperforms GK104 similarly in perf/mm^2, a 410mm^2 GM204 would = 1.9x the performance of GK104 (gtx 770).

That is exactly 30% faster than a gtx 780 TI. Looks very plausible to me.

EDIT: Cloudfire beat me to it. I was too quick to respond before reading through the thread.

Dont measure die size to calculate performance. GM107(1.87B) has ~43% more transistors than GK107(1.3B) and average performance close to 40%-50% higher.

GM204 at ~400mm2 may have close to 5B to 5.5B transistors, it should be close to 780Ti at 1080p(perhaps a little faster depending on the clocks) but slower at higher resolutions if the rumors for 256bit memory is true.

 

[boxleitnerb]

Why? Memory bandwidth itself says nothing. I just don't understand why people are so stupid to focus on one number. Just like the GHz race in the P4 days.
It's how efficiently bandwidth is used that matters. Hint: Caching system.

 

[Enigmoid]

Dont measure die size to calculate performance. GM107(1.87B) has ~43% more transistors than GK107(1.3B) and average performance close to 40%-50% higher.

GM204 at ~400mm2 may have close to 5B to 5.5B transistors, it should be close to 780Ti at 1080p(perhaps a little faster depending on the clocks) but slower at higher resolutions if the rumors for 256bit memory is true.

GM107 is about 70-100% faster than GK107.

http://www.anandtech.com/bench/product/1130?vs=1125

 

[AtenRa]

Why? Memory bandwidth itself says nothing. I just don't understand why people are so stupid to focus on one number. Just like the GHz race in the P4 days.
It's how efficiently bandwidth is used that matters. Hint: Caching system.

Its not only the memory bandwidth, you will have less ROPs as well. Also, if they will use the same Architecture as the small Maxwell GM107, then GM204 will also have less Texture Units than GK110.

 

[tviceman]

Dont measure die size to calculate performance. GM107(1.87B) has ~43% more transistors than GK107(1.3B) and average performance close to 40%-50% higher.

GM204 at ~400mm2 may have close to 5B to 5.5B transistors, it should be close to 780Ti at 1080p(perhaps a little faster depending on the clocks) but slower at higher resolutions if the rumors for 256bit memory is true.

You are demonstrating your lack of knowledge regarding Nvidia products. GM107 is 70-80% faster than GK107. Every. Single. Review. Shows. That.

No one really knows how much of an effect Maxwell's new caching system will have on higher performing chips. However, putting your therory of ROP's and bandwidth to the test shows....misunderstanding. Despite only having a 32 ROP's and 17% more bandwidth, GTX 770 easily beats the gtx 580 by over 35% at 1440p. And interestingly, gtx770's lead decreases with resolution downsizing, despite having less ROP's. Throws half your hypothesis regarding ROP's and higher resolution out the window with respect to next-gen midrange vs. last gen flagship. Similarly, the GTX 750 TI is only 9% slower at both 1080p and 1440p than the gtx 650 TI boost, despite having 50% less ROP's and 67% less bandwidth. This mysterious performance drop off due to limited bandwidth and ROPs at higher resolutions doesn't exhibit itself here, either. On top of that, an overclocked GTX 750 TI easily surpasses a gtx 650 TI boost in performance, even while still having 45-50% less bandwidth.

It's every bit a guess with using transistors to estimate upcoming performance as it is die size. GM107 is 70-80% more efficient than GK107. If GM104 comes with the same efficiency leap and TDP as GK104, then it'll only be 10-20% faster. But if Nvidia engineered the chip to use higher leakage transistors, then it'll have a higher TDP.

 

[boxleitnerb]

Its not only the memory bandwidth, you will have less ROPs as well. Also, if they will use the same Architecture as the small Maxwell GM107, then GM204 will also have less Texture Units than GK110.

So you concede the point about memory bandwidth? Good.
Then you don't know how many ROPs per memory controller will be used in GM2xx. Also you don't know about the capabilities/throughput of the ROPs and TMUs themselves.
Finally you don't know if and how much the ROPs/TMUs are a limiting factor at all. Especially since for instance the TMUs are directly coupled with the Cuda Cores. Less TMUs -> less Cuda Cores. There is no way to tell if the Cuda Cores or the TMUs are the limiting factor if there are performance differences.

To sum up:
You/we simply know too little to make assumptions like that.

 

[Deders]

Why? Memory bandwidth itself says nothing. I just don't understand why people are so stupid to focus on one number. Just like the GHz race in the P4 days.
It's how efficiently bandwidth is used that matters. Hint: Caching system.

I would have thought that the memory interface would have gotten better and better over the years, despite this the 670 was crippled at only 6GHz and 256 bit, overclocking it to 6.5-7GHz gave massive gains, almost 20% in some cases.

This was the difference between playable and unplayable at maximum detail.

 

[boxleitnerb]

I would have thought that the memory interface would have gotten better and better over the years, despite this the 670 was crippled at only 6GHz and 256 bit, overclocking it to 6.5-7GHz gave massive gains, almost 20% in some cases.

This was the difference between playable and unplayable at maximum detail.

At 10-20% difference? Hardly. And what you write about the improved interfaces is true but it doesn't touch on the matter of caching systems and bandwidth utilization. 8GHz is the most I see for GDDR5 for power reasons.

 

[Deders]

At 10-20% difference? Hardly.

The difference was huge when I overclocked my old 670, it was close to 20% increase in most cases. Much bigger impact than core frequencies.

 

[boxleitnerb]

What now, some cases or most cases? 7 vs 6 GHz is 16.6% increase. I would say that is hardly noticeable in terms of fps. Placebo effect at play

 

[tviceman]

I would have thought that the memory interface would have gotten better and better over the years, despite this the 670 was crippled at only 6GHz and 256 bit, overclocking it to 6.5-7GHz gave massive gains, almost 20% in some cases.

This was the difference between playable and unplayable at maximum detail.

20% is impossible since going from 6 to 7ghz is only 16.7%. And for a game to scale 100% with memory bandwidth, that means it was ENTIRELY bandwidth constrained even up to 7ghz. I owned a gtx 670 and did not ever see 100% scaling with memory overclocks in any game. Not once. Not even when I overclocked the core first to make the card have even more texture/pixel fillrate per gb of bandwidth. The best I got was about 15% in game when overclocking both the core and memory to their max. But it was never 100% scaling with the memory no matter what my core clock was at.

 

[Deders]

Maybe it was 20% with the core clock included, but I was astounded by the increase in Crysis 3 and other games I was limited in from just overclocking the memory alone, it pretty much was 1:1 increase in performance. I remember Furmark also showing the 20% increase when switching from my stock profile to the overclocked one.

I would say that any game I was already getting 60fps in I obviously wouldn't notice any difference as I was using Vsync, but games that could take advantage of the additional memory bandwidth really did and like I said before it was the difference between unplayable and comfortably playable at max settings.

 

[RussianSensation]

what does that even mean? Nvidia has always given better performance per dollar with a next gen card. AMD is the only company that gave LESS when they came out with the 7970 for 550 bucks which was 50% more than the 6970 launched at while giving only 40% more performance.

Hmm...not exactly true. GTX780Ti at $699 provided inferior performance/$ over the 680/770 or 780, whichever of these you view as a previous gen flagship vs. 780Ti. Not to mention when 580 replaced 480, the price/performance was horrendous since after-market 480s cost $300-350 when 580 came out at $499.

Yeah, like this one ^^

Your referring to the 7970 that a user had to overclock to the max and then justifying the price over a gtx580.

No, I am not. 7970 at stock provided 25% out of the box 1080P but it was 30-33% at 1600p and it gave 3GB of VRAM. I also criticized this card btw at the time of launch but many don't remember. My point is the same people who criticized 7970 with it delivering only 25-30% more stock vs. stock performance for $550 should be ripping the 880 just the same if it only beats 780Ti by 25-30% and launches at $550 but I am not seeing the same critique at all. If anything, people are saying it would be a great deal! For me if 880 is 25-30% faster at $550, it would be good but not particularly special. It's about what's expected given the delays at 20nm.

At the same time smearing your hypothetical GTX880 for your guesstimate of a stock 20-30% performance increase.
Its wrong on every single imaginable way.

first, your willingness to forgive the HD7970's tiny 25% increase over the 580 because of its suggested overclocking head room.

Not only because of overclocking, even stock vs. stock as explained above.

Secondly, PC enthusiasts do overclock and it's a huge factor on forums such as this. This isn't BestBuy or VGChartz. 7970 is among the most legendary GPUs ever made despite being 2.5 years old:

1) An overclocked 7970 trades blows with an overclocked 770 and even 2.5 years since it launched, you have to spend A LOT to get 30-40% more performance over the overclocked 7970.

2) To outperform 7970 OC CF today, you need to spend at least $700 on dual 290s and to get > 35%-40% faster, you have to spend $1200 on 780Tis OC on NV side:
http://www.sweclockers.com/image/diagram/7051?k=7486a0c3a2fb5b024a9eb1702d2ade0f
http://www.sweclockers.com/image/diagram/7052?k=42bba2ee27c49cedcf595be5b4a96e8e

3) 7970 paid for itself many times over with bitcoin and script mining. Even though I complained about 7970's price for months -- true -- but then I realized even if 7970 cost me $1,000, I would have bought them anyway since over time they became free and made $$$ for a many generations of free GPU upgrades. R9 390/490/590/690, etc. GTX800/900/1000 will never be able to claim such a feat. You realize when all is said and done getting dual 7970s cost $0? Those who didn't' participate won't understand how epic 7970 was since it was a gaming GPU that happened to make $ on the side.

Double standard #1
You have insisted that people focus on the max overclocking performance of the original 7970 but you only care about the stock baseline performance of the GTX880.

Incorrect. I considered both situations already. I didn't say we must talk about some max overclocked 7970 @ 1250-1270mhz which is what the best overclocking 7970's reached.

Case 1 (stock vs. stock): 7970 stock 25-30% for nearly the same price as a 580 3GB. People complained. If 880 stock beats 780Ti by only 25-30% for $550, it's not any better than 7970 as a generational jump. Since people complained that 7970 at stock provided such a "tiny" increase at $550 as you put it, the same people should be just as pissed that 880 may provide only a 25-30% increase at $550.

Case 2 (OC vs. OC): 7970 OC outperformed 580 OC by 40-80%. You think 880 OC will outperform a 780Ti OC by 40-80%? Even if it does, this level of performance only matches what 7970 brought to the market over the 580.

Difference is, 880 can't make $1 mining so it'll never be as legendary as the 7970 even if it provides 25-30% stock and 40-80% OC performance for $550. Sorry, but there will never be a card as good as 7970. 7970 will forever remain the ONLY card ever made which paid for itself 2-10x over while providing 70% of flagship card's performance for 2.5 years of its existence.

Double standard #2
You use this to justify a price increase for a 7970 (a brand new series) while the rumors on the gtx880 suggest it will be cheaper than the 780ti

880 is also a brand new series so what's your point?

Did you see me criticize 880 the same way I did 7970 if 880 were to launch at $499-549? Nope. I said that's where I would be happy for the cards to land. If 880 costs more, it's another increase over 680, clearly. And I just pointed out that people here are not making a huge deal if 880 is $600-700 but there was a lot of backlash when 7970 launched for $550, while suggesting that 880 at $550 would be an amazing deal, which it isn't. That's double standard unless 880 beats 780Ti by > 30% stock vs. stock and by 40-80% OC vs. OC.

I also don't think you realize how awesome 7970 became over time. The chance of 880 beating 780Ti by anywhere near in % terms compared to how 7970/7970GE demolishes 580 in modern games is very small. We don't even need to talk about Max OC 7970. Let's take a 1.05Ghz 7970 GE clocks -- it mops the floor with a 580. For all intends and purposes an after-marker 7970 = 7970GE = 280X. And by June 2012, 7970 already dropped to just $400. What are the chances that 880 hits $400 in 5-6 months from launch and still ends up laying waste to 780Ti by 50-90% like this? I say almost 0%.

2.5 years after 7970 launched, with a mild OC to R9 280X/7970GE, it's beating 580 by 53% and in a pair as HD7990 they not far behind the 780 SLI in 4K leave 690 far behind.

http://gamegpu.ru/images/remote/htt...Battlefield_4_Dragons_Teeth-test-bf4_3840.jpg

So I don't even bother comparing GM204 880 to 7970 price/performance or on value because 880 will lose automatically since there is no more script/bitmining that will make 880 free. If we talk about 7970 vs. 880 purely from a gaming perspective, I would be more than satisfied if 880 is 25-30% faster at $499-549 because I don't think NV will even deliver that given how well their $650 780 and $700 780Ti sold.

 

[Kippa]

Out of the upcoming 3 new NVidia cards coming out soon is one of them going to be the Titan II or is that coming out next year? I got a Titan last year and love it for 3d rendering, and gaming.