Possible GTX 880 benched in 3DMark Firestrike Extreme!

[bryanW1995]

I break 7000 on a 1300Mhz 780ti with a 5ghz 3930K, no tweaks beyond having texture filtering set to best performance, but none of the lucid virtu cheats or any of the outright cheats that you can use. Granted it is a 3D mark ranked score I offered, but it was the lowest one I could find using the same CPU.

Anyway I agree, everything about this image stinks to me. The performance is too low and it's likely a photoshop troll. Releasing a new flagship that amounts to the performance of an overclocked 780ti is just not reasonable.

All signs point to a 30-40% performance increase at best with this release, not surprising to me at all.

 

[Attic]

Crysis 3 IQ to hardware ratio is incredibly poor for very high. Drop some settings down to high, reduce SMAA to 2x and the 880 could easily play at 1600p.

You don't need to max things out, often doing so is incredibly pointless if there is virtually no IQ gain.

You are completely right. The curve of performance to quality goes berserk at a certain point in many games settings, crysis 3 included. I and a few others just can't help ourselves. Price will determine who the 880ti is for, I don't think it will be for folks who don't suffer from "ultra/max" or bust syndrome. I think it will be quite a bit better than 780ti all things considered (VRAM/performance/power efficiency).

I couldn't tell you why very high looks better than high, for the most part it just feels better. That initial level in the rain, wozers.

 

[toyota]

anyone thinking they are going to charge 600-700 bucks for a freaking mid range chip are crazy.

 

[Techhog]

anyone thinking they are going to charge 600-700 bucks for a freaking mid range chip are crazy.

I think it's possible unless they suspect AMD can put out something competitive by the end of the year.

 

[toyota]

I think it's possible unless they suspect AMD can put out something competitive by the end of the year.

personally I think there is zero chance of it being more than 550 and I would guess its going to be 500 bucks tops.

 

[Techhog]

personally I think there is zero chance of it being more than 550 and I would guess its going to be 500 bucks tops.

Why would they not milk it, though? I'll say $600. Nvidia is going to market it as an enthusiast card.

 

[VulgarDisplay]

How much is the 780ti selling for? Is this card faster? What do you think the price will be?

 

[toyota]

How much is the 780ti selling for? Is this card faster? What do you think the price will be?

so faster next gen cards always cost more? if that were true then we would be paying several thousands dollars for all gpus by now.

seriously what is up with all these ridiculous comments thinking this card is going to be more expensive than the 780ti?

 

[VulgarDisplay]

so faster next gen cards always cost more? if that were true then we would be paying several thousands dollars for all gpus by now.

seriously what is up with all these ridiculous comments thinking this card is going to be more expensive than the 780ti?

We are paying thousands for GPUs right now....

I don't think these will cost that much, but I'm guessing they will be closer to $750 because people will buy them at that price.

 

[Deders]

Don't they usually have price brackets that new cards occupy? The 780 and TI are different as they had to be priced higher. I expect 880/870 will come out at 680/670 prices.

 

[toyota]

We are paying thousands for GPUs right now....

I don't think these will cost that much, but I'm guessing they will be closer to $750 because people will buy them at that price.

I meant every card would cost thousands of dollars by now based on your logic.

and sorry but I think its flat out stupid to think it will be 750 bucks. this is the only forum I have seen even making such ridiculous comments. most people seem to have enough sense to know this is a mid range next gen chip that will used for the 880 and that it will be around 500 bucks.

 

[Keysplayr]

Guys just buy what you can afford. Prices go up. Just like everything else. maybe graphics cards were horrifically under priced for all these years and they are just now waking up.
Joking of course but come on. Buy what you want at the price you want. Who cares if there are thousands of dollar cards out there. Doesn't mean you have to buy one. Not even a little.
I'm going to expect an exact repeat of Kepler. The mid range GTX880 will be marketed as the flagship until Big Maxwell comes out possibly on 20nm.

 

[AtenRa]

I've said it before and i will say it again, unless GM204 is at 20nm there is no way it can be faster at high resolutions with only 256bit memory, 32ROPs and less Texture units(160 ??) than GTX780Ti 384bit, 48 ROPs and 240 Texture units. Especially at 4K.

 

[blackened23]

edit: dp

 

[blackened23]

LOL. Just like the 680 wasn't faster than the 580 (better paper specs), right?

Paper specifications do not mean everything. With the 680..... Paper specs didn't mean much with the 7970 nor the 580 which both had superior paper specs to the 680. But the 680 launched with every review declaring it the new single GPU king despite the 7970 and 580 having better paper specs. Fixating on a single number such as the bus width without regard for architecture is meaningless. It's almost as stupid as fixating on something ridiculous such as more cores on CPUs. We know that doesn't help the FX CPUs from not being mediocre.

Oh yeah. Anyone remember the R600? ATI HD 2900 series anyone? 512 bit memory bus? I seem to recall that having superior paper specifications with a 512 bit memory bus width. What happened there? Squat, because it was not even remotely close to NV's competing solution at the time, which, oddly enough, had worse paper specifications.

It's also pretty hilarious that anyone thinks a next gen flagship will be slower than prior gen. I'm pretty sure that aint what's really happening. I mean....there's not even a point to a next gen GPU series if it were slower. Give me a break. But, if you want to think that paper specs are the end all be all, keep thinking that.

Personally, I cannot see the 880 not being faster than the 780ti. But we just don't know what the speed difference will be. 10%? 15%? I don't know, but I cannot see it being slower. When has NV ever released a flagship slower than prior gen? That would eliminate the purpose of next gen.

 

[Keysplayr]

I've said it before and i will say it again, unless GM204 is at 20nm there is no way it can be faster at high resolutions with only 256bit memory, 32ROPs and less Texture units(160 ??) than GTX780Ti 384bit, 48 ROPs and 240 Texture units. Especially at 4K.

I'm not going to say you're wrong, because anything is possible, but I'd say examine what was done with GM107. More with a lot less everything. I'd also say that GM107 was a learning experience for Nvidia. Seeing how far they can tweak the architecture on 28nm. That was a long while ago now and I'll also assume that it is possible that further improvements are being accomplished on 28nm with the rest of the Maxwell family GM2xx.

EDIT: And Blackened makes a good case with GTX680 vs GTX580.

 

[Grooveriding]

I meant every card would cost thousands of dollars by now based on your logic.

and sorry but I think its flat out stupid to think it will be 750 bucks. this is the only forum I have seen even making such ridiculous comments. most people seem to have enough sense to know this is a mid range next gen chip that will used for the 880 and that it will be around 500 bucks.

It's really not though because of a few reasons. We've seen the board shot and it's estimated at about 430mm2, that is a fairly large die, about the size of Hawaii - about 50% larger than the GK104 in gtx 680 & 770. I think it's fair to say that it would be mid range if it was coming out as intended on 20nm/16nm and would have a smaller die size packing in the same transistors. But, it looks like nvidia can no longer afford to jump to the latest nodes immediately as they become available so we're getting 4-5 years of 28nm.

Performance wise it won't be that impressive coming from the past generation's flagship, just like the 680 wasn't, but it will have to be a bit faster - even 20%. Considering the 780ti is $700, what reason is there to change anything about that price point as there will be no competition for a card that is 780ti +15-20% more from AMD ? Again another difference from the 680 launch where the 7970 was just as fast as the 680 with both cards trading blows depending on the game.

Wouldn't surprise me to see it arrive as a $1000 Titan 2, at the least it will be a $700-$750 card. Probably will be able to get the 870 for $450-$500 and get 780ti performance out of that card.

 

[dacostafilipe]

EDIT: And Blackened makes a good case with GTX680 vs GTX580.

The 680 had the same bandwidth as the 580 (192 Gb/s).

The 780ti with this memory at 1.7Ghz get's 326Gb/s and to get the same bandwidth on 256bit you would require 2.5ghz ... that's a lot.

 

[Keysplayr]

It's really not though because of a few reasons. We've seen the board shot and it's estimated at about 430mm2, that is a fairly large die, about the size of Hawaii - about 50% larger than the GK104 in gtx 680 & 770. I think it's fair to say that it would be mid range if it was coming out as intended on 20nm/16nm and would have a smaller die size packing in the same transistors. But, it looks like nvidia can no longer afford to jump to the latest nodes immediately as they become available so we're getting 4-5 years of 28nm.

Performance wise it won't be that impressive coming from the past generation's flagship, just like the 680 wasn't, but it will have to be a bit faster - even 20%. Considering the 780ti is $700, what reason is there to change anything about that price point as there will be no competition for a card that is 780ti +15-20% more from AMD ? Again another difference from the 680 launch where the 7970 was just as fast as the 680 with both cards trading blows depending on the game.

Wouldn't surprise me to see it arrive as a $1000 Titan 2, at the least it will be a $700-$750 card. Probably will be able to get the 870 for $450-$500 and get 780ti performance out of that card.

IMHO, the mid range GTX880 will be marketed as the flagship until Big Maxwell comes out possibly on 20nm. Big Maxwell (GM210 or whichever number it's assigned) will show up on 20nm later on and that will be the GTX980/Ti and Titan 2/ Black GPUs.

Exactly like we have now with the entire Kepler line. Only difference is all of that was on one (28nm) node. A mid range chip on 28nm at 430mm2 isn't that out of the question when we are literally at the tail end of a very very mature 28nm node process. A GTX 980 Die might be the same size on 20nm as a GTX 880 is on 28nm. I'm speculating of course but does my speculation sound at all unreasonable?

 

[blackened23]

The 680 had the same bandwidth as the 580 (192 Gb/s).

The 780ti with this memory at 1.7Ghz get's 326Gb/s and to get the same bandwidth on 256bit you would require 2.5ghz ... that's a lot.

The 780ti also has roughly the same memory bandwidth as the 290X. More in fact. But others will continue to play games and say that paper specifications mean everything. Bus width means EVERYTHING. Everything. Even though I can think of 10 examples off the top of my head of that being NOT the case, including the 2900XT as I mentioned..

Anyway, i'm pretty certain the 880 will be just as the name implies...a flagship. I cannot see a next gen flagship being slower than prior gen. I mean, just why? there's no point to 880 if it doesn't perform as an x80.

 

[Keysplayr]

The 680 had the same bandwidth as the 580 (192 Gb/s).

The 780ti with this memory at 1.7Ghz get's 326Gb/s and to get the same bandwidth on 256bit you would require 2.5ghz ... that's a lot.

Maxwell may need significantly less if the onboard cache is anything like GM107. What is the memory bandwith of GM107 compared to say GK107 or even 106?

The GTX 650 Ti (GK106) has a 128-bit memory bus clocked at 1350MHz, for a total of 86.4GBps of memory bandwidth.

The GTX 650 TiB has a 192-bit memory bus clocked at 1500MHz, for a total of 144.2GBps of RAM bandwidth.

GK107 (GTX650) with the 128-bit memory interface and 1GB of GDDR5 memory clocked at 1.25GHz (5GHz effective), the GTX 650 has a total memory bandwidth of 80GB/s

GM107 (GTX750Ti) has just 86.4GB/s 2GB of GDDR5 RAM, which is clocked at 1350MHZ (5400MHZ effective).
And it hangs with a GTX 650TiBoost or just shy with literally 60% of the bandwidth and less shaders.

http://www.xbitlabs.com/articles/graphics/display/geforce-gtx-750-ti_8.html

 

[AtenRa]

LOL. Just like the 680 wasn't faster than the 580 (better paper specs), right?

Paper specifications do not mean everything. With the 680..... Paper specs didn't mean much with the 7970 nor the 580 which both had superior paper specs to the 680. But the 680 launched with every review declaring it the new single GPU king despite the 7970 and 580 having better paper specs. Fixating on a single number such as the bus width without regard for architecture is meaningless. It's almost as stupid as fixating on something ridiculous such as more cores on CPUs. We know that doesn't help the FX CPUs from not being mediocre.

Oh yeah. Anyone remember the R600? ATI HD 2900 series anyone? 512 bit memory bus? I seem to recall that having superior paper specifications with a 512 bit memory bus width. What happened there? Squat, because it was not even remotely close to NV's competing solution at the time, which, oddly enough, had worse paper specifications.

It's also pretty hilarious that anyone thinks a next gen flagship will be slower than prior gen. I'm pretty sure that aint what's really happening. I mean....there's not even a point to a next gen GPU series if it were slower. Give me a break. But, if you want to think that paper specs are the end all be all, keep thinking that.

Personally, I cannot see the 880 not being faster than the 780ti. But we just don't know what the speed difference will be. 10%? 15%? I don't know, but I cannot see it being slower. When has NV ever released a flagship slower than prior gen? That would eliminate the purpose of next gen.

unless GM204 is at 20nm

Comparing 40nm GTX580 to 28nm GTX680 is apples to oranges.

If GM204 is 28nm same as GK110 with only 256bit memory controller that most probable will have 32ROPs and perhaps 160 Tex units, it cannot be faster than GTX780Ti at 4K.

You cant have faster ROPs and Texture Units without significantly increasing die space at the same 28nm process. Either you keep the same ROPs and you increase Stream Processors while decreasing Texture Units to save die area(what NVIDIA did with GM107) or you increase ROP and Texture performance at a die size penalty.

GTX680 may have less ROPs than GTX580 but you don’t know how much faster each of those 28nm ROPs are compared to 40nm ROPs on the GTX580.
Yes the architecture plays a role in all this but you cannot magically increase performance that much more with less transistors and less compute units at the same process.

NVIDIA could very well position the GTX880 as a 1080/1440p part to counter R9 290X and leave GTX780Ti as the Flagship for 4K and 3x Monitor High resolution Surround until they will release the GM200.

 

[AtenRa]

Maxwell may need significantly less if the onboard cache is anything like GM107. What is the memory bandwith of GM107 compared to say GK107 or even 106?

The GTX 650 Ti (GK106) has a 128-bit memory bus clocked at 1350MHz, for a total of 86.4GBps of memory bandwidth.

The GTX 650 TiB has a 192-bit memory bus clocked at 1500MHz, for a total of 144.2GBps of RAM bandwidth.

GK107 (GTX650) with the 128-bit memory interface and 1GB of GDDR5 memory clocked at 1.25GHz (5GHz effective), the GTX 650 has a total memory bandwidth of 80GB/s

GM107 (GTX750Ti) has just 86.4GB/s 2GB of GDDR5 RAM, which is clocked at 1350MHZ (5400MHZ effective).
And it hangs with a GTX 650TiBoost or just shy with literally 60% of the bandwidth and less shaders.

http://www.xbitlabs.com/articles/graphics/display/geforce-gtx-750-ti_8.html

Have a look from the xbitlabs.com link at GTX650Ti Boost and GTX750 Ti when AA filters are used.

The GTX750Ti at default clocks (1020/5400) is always slower than GTX650Ti Boost because of lower Memory bandwidth and less ROP performance.

Thats what is going to happen with a 256bit 32ROPs GTX880 vs 384bit 48ROPs GTX780Ti. When AA filters and higher resolutions will be measured the GM204 will strangle to keepup.

 

[Keysplayr]

Have a look from the xbitlabs.com link at GTX650Ti Boost and GTX750 Ti when AA filters are used.

The GTX750Ti at default clocks (1020/5400) is always slower than GTX650Ti Boost because of lower Memory bandwidth and less ROP performance.

Thats what is going to happen with a 256bit 32ROPs GTX880 vs 384bit 48ROPs GTX780Ti. When AA filters and higher resolutions will be measured the GM204 will strangle to keepup.

I don't think you're seeing this clearly. The GTX750Ti has a deficit at every aspect and still manages to stay withing spitting distance of GTX650TiBoost. with or without AA. OF COURSE a bandwidth deficit is going to impact performance when higher levels of AA are used especially at higher resolutions. Point is, Nvidia did a whole lot more with a whole lot less with 750Ti. You don't agree?

 

[Keysplayr]

Comparing 40nm GTX580 to 28nm GTX680 is apples to oranges.

Which is, again, why you should look more closely at the 28nm GM107 compared to GK107/GK106.

Same process node. Much better performance at much lower power consumption. Apples to Apples.

Extrapolate to higher end Maxwell parts.