Nvidia GeForce GTX 590 3 GB Review: Firing Back With 1024 CUDA Cores

[jprahman]

Wow, the noise difference is quite noticeable. At full speed the 6990 sounds like it's about to blow up. Although I do have to admit that the GTX 590 is pretty much tied with the 6990 in terms of performance.

 

[cryptz]

i wonder how it would match up if you water cool it and get the clock rates back up to the 580 lvl

 

[dragonsqrrl]

[citation][nom]cburke82[/nom]Thrown in the fact that in most cases your refresrate will be 60hz and both cards are over 60FPS. And you have a tie . If my screen wont display more than 60 FPS and the 6990 is at 70 and the 590 is at 78...Why do I care i wont see the extra 10-18 FPS anyhow.[/citation]
I'm sorry, but I can't entirely agree with that. So you're saying that any advantage or disadvantage over a 60fps average is negligible? You have to remember that these are averages, and that frame rates can often dip below that 60Hz limit you refer to during actual gameplay. If this is your primary concern (as opposed to absolute performance) then min frame rates should also be taken into account alongside averages. This is an area where the HD6000 series has improved significantly over the 5000 series, basically achieving parody with Fermi based cards.

 

[PudgyChicken]

Wow. While dual GPU cards are completely absurd, I don't think I could live with the 6990 after hearing that sound clip. And I'm coming from GTX 480 SLI, which at times, especially running F@H GPU, drives me batshit insane.

Then again, GTX 480 SLI ain't too shabby, so I wont find the need to upgrade for a while.

 

[srgess]

[citation][nom]cRACKmONKEY421[/nom]When will we see 590 in SLI? Not even sure you can do that ATM[/citation]

There already a lots of SLI 590 benchmark overnet. Seem a lots better then a CF 6990 at multiple screen and high res even if nvidia dont have dedicated quad SLI driver.

 

[cburke82]

I'm sorry, but I can't entirely agree with that. So you're saying that any advantage or disadvantage over a 60fps average is negligible? You have to remember that these are averages, and that frame rates can often dip below that 60Hz limit you refer to during actual gameplay. If this is your primary concern (as opposed to absolute performance) then min frame rates should also be taken into account alongside averages. This is an area where the HD6000 series has improved significantly over the 5000 series, basically achieving parody with Fermi based cards.

True I see your point. It can be hard to see minimum FPS like if i run the Metro 2033 benchmark it will say minimum is 6FPS but thats for 1 frame wile the graph shows much higher overall. I was basicly trying to say there is no REAL winner they are both good both win some loose some with the benchmarks. I guess if toms posted a graph of the framerates wile the benchmark was running we would have a better idea. If the range was say 30-70 VS 55-78 then yeah ill take the second card lol I do however stand by my statement that when dealing with this type of cards the setting should remain at max and never come down. And the resolution should start at 1080p no less.

 

[anavocado]

Some people care about noise level, some don't. Depends on if you use a headset, quiet or loud volume speakers, and the type of background noise you deal with all the time anyway.

I would've liked to see how the quad gpu setups handled a 1920x1080 resolution. They're sure to always be over 60fps, but I wonder if even $1400 of graphics is enough to play new games with high settings at the 120fps a 120Hz monitor allows. I'm betting that at least for metro and crysis, it's not.

Even with 4 gpus, running 3 monitors at 1920x1080 and 120fps isn't possible, judging by the fact that at the lower 2560x1600 resolution they hardly reach 120fps. This means that 1920x1080 is still an entirely valid choice for those that like the smoothness 120Hz provides.

Unfortunately, it also means that we're extremely unlikely to see a 2560x1600@120Hz (via displayport) monitor released anytime soon, as nearly nothing is powerful to play (new) games on it. Even 1680x1050 is still an important resolution to include on high-end card reviews, as the only 16:10 120Hz monitor uses this, and it takes a lot of power to drive new games at high settings and still get 120fps. (While 1920x1200 is still just within dual-link DVI specification, no one has made a 120Hz 1920x1200 monitor yet for some reason)

 

[Proximon]

I very much liked how this was not a cookie-cutter review and provided info I didn't get elsewhere.

 

[verbalizer]

Funny thing about assumptions, the card blew up in the techpowerup review while overvolting, looks like the safety failed.

http://www.techpowerup.com/reviews/ASUS/GeForce_GTX_590/26.html

more in particular:
http://www.techpowerup.com/reviews/ASUS/GeForce_GTX_590/27.html

Value and Conclusion


* According to NVIDIA, the target price for the GTX 590 is $699, but due to limited supply we expect higher pricing in the first weeks or months.

POSITIVE:
* High 3D performance
* Lower noise than HD 6990
* Supports more than two active displays
* Support for DirectX 11
* Support for CUDA/PhysX

NEGATIVE:
* Expensive & High power draw
* Performance is not higher than HD 6990
* In both 2D and 3D, the card emits a lot of fan noise
* Card blew up during testing, power limiting system does not work reliably
* Full maximum performance depends on SLI profile support
* Advanced overclocking could be complicated by NVIDIA's power limiting system
* It's harder than on other cards to get a stable overclock
* Limited relevance of DirectX 11

NVIDIA's new GeForce GTX 590 sets out to conquer the dual GPU performance throne. It battles AMD's Radeon HD 6990 that was released just two weeks ago. Designing this kind of graphics card means you are limited by power and heat instead of by the raw power of your GPU design. Basically you decide on the power input configuration, 2x 8-pin in this case, which means the card is specified to draw a maximum of 375 W (= 150+150+75) of power. Then you find a thermal solution that can move that kind of heat away from the card. In the next step you build the board and tune the clock speeds and voltage to fit into those limitations. A more power efficient GPU will let you clock it higher while staying within the power limit, which is what seems to be the deciding factor in the GTX 590 vs. HD 6990 duel. Based on our testing AMD's Cayman GPU is about 20% more efficient than NVIDIA's GF110, which means that AMD can run their GPU faster, letting it catch up on the single-GPU performance disadvantage it has. Now with two GPUs in place, single GPU performance doesn't matter anymore but performance per Watt.
As a result GTX 590 and HD 6990 are roughly the same performance when averaged. In lower resolutions GTX 590 wins, at 1920x1200 both are even, and at 2560x1600 AMD's HD 6990 wins by 3%. Since 2560x1600 is the primary resolution that both of those cards should be used for, my conclusion is that GTX 590 is slower than HD 6990, but not by much. It is still disappointing to see that NVIDIA could not turn their single-GPU winning GF110 into a dual-GPU design win too.
While I usually do not believe in large gains from new drivers, this time I am not so sure. If NVIDIA's power limiting system is active all the time, in all applications, then this could cause some applications to get capped down, hurting performance a lot. If NVIDIA manages to tune their algorithm some more, they could get some extra performance out of their design - without the typical driver/shader optimizations that we see from time to time in graphics drivers.

Both AMD's HD 6990 and NVIDIA's GTX 590 are hard to tame beasts during overclocking. On AMD you need to disable their ULPS ultra low power state to get anywhere. On NVIDIA I couldn't find a fix for the extremely unstable card during overclocking. When increasing clocks to find stable settings, the card often completely crashes the system in a test that it completed before, with Windows 7 the normal behaviour should be that the driver will stop rendering, show a "The driver has stopped responding" message, and you can just continue without reboot.
Once I got normal overclocking stable I went to work on voltage tuning which promises additional performance gains. Within a short time of testing, the card broke permanently by what seems to be an overcurrent situation. NVIDIA promises that their power capping will avoid such conditions by constantly monitoring the card's power draw and reducing clocks if necessary - in all applications. Apparently that system did not work and took the card with it. At $699 and possibly an RMA denial if that happens to you, my suggestion is that you should really not engage in overclocking with GeForce GTX 590 until this issue has been completely cleared up. For the record, the card worked perfectly fine and rock stable when it was just installed and left alone without any extra tweaking.
When directly comparing GTX 590 and HD 6990, it is difficult to find many clear wins for the GTX 590. It does emit a good deal less noise but is still noisy, all the other comparisons are either a draw or win for the HD 6990. Feature wise NVIDIA does bring a bit more to the table with 3D Vision Surround, CUDA and PhysX - if you use those. We wanted to bring you a Quad-SLI review of two of these cards in SLI, but NVIDIA did not have a driver ready. Quite surprising considering that this card is not much different to a GTX 480 Quad SLI setup. Functionally GF100 and GF110 are identical, the configuration of two GPUs on one card connected by NF200 are functionally the same as two cards connected by an NF200 chip located on the motherboard.

Price-wise both HD 6990 and GTX 590 are tied around $700 which is a lot of money to spend on a graphics card. My recommendation would be to go with a single GPU GTX 580 and wait what the future brings in terms of games - most games are console ports, Crysis 2 is DX9. Developers! The PC needs more love from you.

 

[cangelini]

[citation][nom]srgess[/nom]There already a lots of SLI 590 benchmark overnet. Seem a lots better then a CF 6990 at multiple screen and high res even if nvidia dont have dedicated quad SLI driver.[/citation]

It's in this story, even!

 

[Bigmac80]

Wow the 590 is actually a bit cheaper on newegg and alot quieter and smaller then the huge 6990. 590 is 11inches and 6990 is about 12 inches. 590 has physX it's quiter and smaller...Nvidia wins imo..

 

[MasterMace]

I'M SORRY AMD FANBOYS, I COULDN'T HERE YOUR COMPLAINTS OVER THE SOUND OF YOUR TORNADO ALARM... I MEAN 6990!

 

[cburke82]

Some people care about noise level, some don't. Depends on if you use a headset, quiet or loud volume speakers, and the type of background noise you deal with all the time anyway.

I would've liked to see how the quad gpu setups handled a 1920x1080 resolution. They're sure to always be over 60fps, but I wonder if even $1400 of graphics is enough to play new games with high settings at the 120fps a 120Hz monitor allows. I'm betting that at least for metro and crysis, it's not.

Even with 4 gpus, running 3 monitors at 1920x1080 and 120fps isn't possible, judging by the fact that at the lower 2560x1600 resolution they hardly reach 120fps. This means that 1920x1080 is still an entirely valid choice for those that like the smoothness 120Hz provides.

Unfortunately, it also means that we're extremely unlikely to see a 2560x1600@120Hz (via displayport) monitor released anytime soon, as nearly nothing is powerful to play (new) games on it. Even 1680x1050 is still an important resolution to include on high-end card reviews, as the only 16:10 120Hz monitor uses this, and it takes a lot of power to drive new games at high settings and still get 120fps. (While 1920x1200 is still just within dual-link DVI specification, no one has made a 120Hz 1920x1200 monitor yet for some reason)

I guess call me crazy but if a solid 60FPS with no major drops in FPS looks very very smooth I dont see a big improvement going to 120hz and definatly not if you have to lower your resolution to do so. after 60 FPS RESOLUTION > FPS

 

[ionut19]

Thank you for the article. I enjoyed reading it, was pleasantly surprised with the multiple monitor tests. Didn't expect that.
As a small request please post the Exhaust Air
138.5 degrees F
138.8 degrees F
Recirculated Air
160.5 degrees F
148.9 degrees F
in C (Celsius) values also.

 

[wuggi]

Quick question, the gtx 590 is essentially two gtx 570's in a single card right ?

 

[luc2k]

Quick question, the gtx 590 is essentially two gtx 570's in a single card right ?

Nope, two 580s downclocked.

 

[iam2thecrowe]

im shocked that nvidia outright lied about the power requirement for the card. They need to change that spec quick fast!

 

[verbalizer]

just like the summary says, I'll stay with the GTX580 (if I had one...) and then play the waiting game...

 

[silverblue]

I'm wondering if they're overpricing the 580s or severely cutting into their profits with the 590.

 

[dragonsqrrl]

[citation][nom]iam2thecrowe[/nom]im shocked that nvidia outright lied about the power requirement for the card. They need to change that spec quick fast![/citation]
This is not a lie... typical TDP for a card is not the same as load power consumption in W. The HD6990 is in the same situation with a 375W TDP, while load power consumption is around 400W. Depending on the game, actual load power consumption is very similar between these two cards.

http://www.anandtech.com/show/4239/nvidias-geforce-gtx-590-duking-it-out-for-the-single-card-king/16
http://www.guru3d.com/article/geforce-gtx-590-review/7

 

[dragonsqrrl]

[citation][nom]silverblue[/nom]I'm wondering if they're overpricing the 580s or severely cutting into their profits with the 590.[/citation]
lol... probably both.

 

[woshitudou]

Nice work nvidia. It's been a few years to see a top card from you. Shame on Tom's for intentionally leaving the 5970 out of benchmarks for the last year. It's nice to see it get beat but it would have been nice seeing it on top you crooked **** ****.

 

[cburke82]

Nice work nvidia. It's been a few years to see a top card from you. Shame on Tom's for intentionally leaving the 5970 out of benchmarks for the last year. It's nice to see it get beat but it would have been nice seeing it on top you crooked **** ****.

Claiming the tops spot based on noise is sort of weak if you ask me. Now if the 590 beatdown the 6990 like the 580 will the 6970 thats one thing. However I dont see anyway to call this test anything but a tie.

 

[cburke82]

Ok so here is what i just did. I went through the first 6 games benchmarked here ( didnt cont WoW because all cards did very well there also my single 6950 never goes below 50 fps for more than a half sec) and only looked at 1080p and up and only looked at the results where full AA setting where used(come on who spends $700 to turn settings down). I took the FPS difference between both the 590 and 6990 for each game a wrote it down. then I came up with the avg difference between the two over all six games. here is what I found.

With full settings and high res:

1080p: The avg shows that at this setting the 590 is -3.8 FPS slower. If you take out F1 with its total AMD bias the avg is -0.94 FPS less that the 6990. So at 1080p with full setting these 2 cards go back and forth with AMD winning just barley when you AVG the results out.

1600p: First of all if you look at 1600p with full AA the 590 does not win any of the first 6 games benchmarked. But here is what I got as far as averages go. at this setting level the 590 was 6.98 FPS slower than the 6990, if you again take out F1 you get 3.68 FPS slower avg across all 6 games. Keeping in mind that at this level and with full settings the 590 was behind in all 6.

So I dont see how toms would say the 590 is the better card. At the higher settings and resolutions the 6990 wins hands down and as toms even says thoes are the resolutions and settings these cards are meant for, If you game with no AA or at 720p just pick what ever single gpu card you like (580 would be the winner there I guess right?). So we know the 590 is not as loud, but strictly looking at the performance of both cards at high settings the 590 would be a close second to the 6990.

If anybody sees any glaring errors in how I came up with this conclusion please point them out. I am not trying to favor one card or the other obviously i would use either one if they were given to me lol. I didnt crunch any numbers one the multi screen benches. If you look though the 6990 wins most of those as well with full settings. Flips on B2BC with a win in no AA and loss with full AA. And loses outright in Just Cause 2.

By the way I used the numbers from the stock 6990 not the one at 880mhz. I felt that was fair being that the 590 was not OC so these numbers would be if you purchased one of both out of the box and installed it with no OC.

 

[foogoo]

Wow. Is anybody else sick of seeing these dual card/gpu solutions being the fastest offerings from these 2 manufacturers? Also whats with the crappy cooling? For me the high end started dying when the x800 (i think that was it anyway) series came out and they started this crossfire/sli competition again. Didn't these dual card/gpu solutions die out once before? I think it was 3dfx then nvidia or something. Ya know things go extinct for a reason.