HardOCP 6990+6970 CF vs 580 Tri Sli

[Mistwalker]

LOL, yeah thats just what Nvidia fanny's wanted. If that was the case, they would have tested Civ5, Crysis 2, Mafia II with PhysX enabled, and then turned on 3D, oh thats 2 things AMD cards can't do. Thats a very narrow set of games/settings picked to make the AMD cards 'win'.

Keep pushing that bar! If price/performance loses, insist on absolute performance. In the rare case that also loses, attack the feature set. I suppose if you don't mind spending large sums of money for slower cards, you can consider Physx (ironic if you want to talk about "narrow" sets of games) a $500 "feature."

To be honest, I didn't expect tri-SLI to lose even with CrossFires slightly better scaling, and I still think the 580 is a sexy card that is just way, horribly overpriced.

 

[Skurge]

LOL, yeah thats just what Nvidia fanny's wanted. If that was the case, they would have tested Civ5, Crysis 2, Mafia II with PhysX enabled, and then turned on 3D, oh thats 2 things AMD cards can't do. Thats a very narrow set of games/settings picked to make the AMD cards 'win'.

In a previous article.
Kyle has written a conclusion after choosing 2 of 4 games where AMD crossfire was broken. Thats a extreme example of his reviews , at times depending on the point he's trying to make. This time its AMD can provide a cost effective eyefinity experience, that would be much more expensive on Nvidia cards.
There is no getting around the price metric currently.
But games, resolutions, settings will return different endings. You can choose to accept any reality you want.

So you have a problem with them testing the 5 most demanding games around at the moment? 4 of them are DX11. At resolutions I think everyone here will agree are best suited to multiple cards. AMD is just faster in this scenario. Is it so hard to accept?

AT tested a similar configuration, but at lower settings with different games. The 6970s were just as fast as the 580s and had much better minimums. We all know that the 6970s scale better and handle higher resolutions better (even ignoring the vram advantage) So even with a different set of games, the conclusion would be similar. The only difference I can see is if you used lower settings, but if your argument is a $1500 setup does better than a $1000 setup at lower settings then you really aren't worth talking to.

 

[Skurge]

To be honest, I didn't expect tri-SLI to lose even with CrossFires slightly better scaling, and I still think the 580 is a sexy card that is just way, horribly overpriced.

Yep, as a single card ignoring price. It is an awesome piece of kit.

 

[3DVagabond]

I have a complaint about these [H] reviews though - they fail to indicate temperatures for the system, especially how other components like motherboard and CPU are affected by it.

Really hot inside a "regular" case (Antec Sonata III in this img) with just a single 6990. Makes HDD run at CPU temps. I wouldn't recommend anyone building a system like in the review without carefully considering the case's ability to keep everything cool.

Tri-fire faster than Tri-SLI isn't the whole story. They tested using triple monitors at an insanely high resolution. I'm wondering if that had anything to do with the performance results for two reasons:

1) On the NVIDIA side, each card gets one monitor while on the AMD side the primary card gets all three monitors. Perhaps Tri-SLI doesn't scale well in this exact circumstance. After all, supporting more than two monitors for NVIDIA cards is essentially a driver hack. How about a single monitor, maybe a nice 30" running 2560x1600 resolution? Besides, out of the five games tested, only one of those was able to run with average framerates above 60FPS (BFBC2 @65.0FPS) with the faster Tri-fire setup. Two of the five games averaged in the 30FPS range with Tri-fire.

2) The insane 5760x1200 resolution may have been too much for the lower amount of texture RAM in the GTX 580 to handle. How about a re-test using 3GB cards?

Well, they would have used this "insane" resolution to justify the GPU power. A single 2560 monitor doesn't require it. You'd spend 90% of your time CPU bound. The res and tri monitor set up is not the problem, it's the point of the exercise. Someone else has stated, they made sure the games weren't memory bound on the 580's.

As far as the low FPS goes, that's what [H] always shoots for. How much eye candy can they throw in and maintain minimum playable frame rates.

I have read reviews that state getting nVidia surround working is not a simple matter. So, I do understand what you are saying, "nVidia just might not do multi monitor as well". That hasn't shown to be the case though in general. It typically "scales" just fine. Just more work to get it going.

 

[Adul]

I was sure that nvidia would win that battle but lose out in price. Turns it was even worse for nvidia there.

 

[n0x1ous]

I can't believe the power consumption, over 1000 watts for Tri sli. Thats nutty.

So the conclusion I guess is that ATI has definitely overthrown nvidia in scaling and performance when it comes to the ultimate setup...

Its more due to the limiting of 1.5gb of Vram at the rez that H was testing at.

nvidia really needed to go 3gb on 580 to let them breath at the resolutions that they would get used at like this.

check out the Vega 5803gb quad sli thread on H......nvidia scaling is phenomenal when the gpus aren't bottlenecked and have enough vram.

the guy had 4 6970's as well and the 4 580's slaughter them

 

[badb0y]

Its more due to the limiting of 1.5gb of Vram at the rez that H was testing at.

nvidia really needed to go 3gb on 580 to let them breath at the resolutions that they would get used at like this.

check out the Vega 5803gb quad sli thread on H......nvidia scaling is phenomenal when the gpus aren't bottlenecked and have enough vram.

the guy had 4 6970's as well and the 4 580's slaughter them

Wasn't that because of some Crossfire bridge problem? I have to go check that again.

 

[n0x1ous]

Wasn't that because of some Crossfire bridge problem? I have to go check that again.

It was one of his factors.....he tested at lower rez as well where that wasn't a problem.

He has like a 5.5ghz 990x i believe and he plays on 3 x 30 inch panels and the gpus are getting near 100% utilization and the scaling was close to perfect on some games.

Used over 2gb of vram on some of them. So the scaling advantage that AMD had in the H test i would say isnt due to drivers, its due to the nvidia cards running out of vram or being cpu bottlenecked.

 

[SlowSpyder]

Its more due to the limiting of 1.5gb of Vram at the rez that H was testing at.

nvidia really needed to go 3gb on 580 to let them breath at the resolutions that they would get used at like this.

check out the Vega 5803gb quad sli thread on H......nvidia scaling is phenomenal when the gpus aren't bottlenecked and have enough vram.

the guy had 4 6970's as well and the 4 580's slaughter them

So $2300 worth of non-reference GTX580 (and probably a second power supply?) beat $1360 worth of reference 6970's?



AMD really did a great job on these cards when it comes to multi-GPU scaling. The value three 6970's provide compared to the GTX580 is amazing. And if you consider 6950's that can be unlocked... Nice job on AMD's part.

 

[Jionix]

It was one of his factors.....he tested at lower rez as well where that wasn't a problem.

He has like a 5.5ghz 990x i believe and he plays on 3 x 30 inch panels and the gpus are getting near 100% utilization and the scaling was close to perfect on some games.

Used over 2gb of vram on some of them. So the scaling advantage that AMD had in the H test i would say isnt due to drivers, its due to the nvidia cards running out of vram or being cpu bottlenecked.

But Kyle clearly pointed out that fact that they used 'Apple to Apple' settings that weren't hitting the VRAM limit in these tests.

 

[n0x1ous]

So $2300 worth of non-reference GTX580 (and probably a second power supply?) beat $1360 worth of reference 6970's?



AMD really did a great job on these cards when it comes to multi-GPU scaling. The value three 6970's provide compared to the GTX580 is amazing. And if you consider 6950's that can be unlocked... Nice job on AMD's part.

No, im not saying the 6970's arent a better value for the money, im just saying its not indicative of nvidia scaling problems relating to drivers or anything. Im saying if money is no object, Nvidia still has the fastest solution money can buy like they have for quite a few generations now. AMD has better value for the money at the top no doubt about it, but if someone has over $1000 to spend just on their GPU's then they probably have 2000 and 3000.....performance per dollar isn't nearly as important up in this range as it is below 300 where most of us buy.

And yes Nvidia made a mistake with only 1.5gb of vram on the market the 580 is intended for.

 

[n0x1ous]

But Kyle clearly pointed out that fact that they used 'Apple to Apple' settings that weren't hitting the VRAM limit in these tests.

Does he mention how much vram was used?

 

[Lepton87]

GTX 580SLI 3-way should be compared to 2x6990 or they should compare 3x 570 to 3-way CF. This comparison isn't fair.

 

[Skurge]

GTX 580SLI 3-way should be compared to 2x6990 or they should compare 3x 570 to 3-way CF. This comparison isn't fair.

Well its 3 nVs top GPUs vs 3 of AMD's top gpu. Same comparison as anands. If you think its unfair cause of price, I agree.

 

[Lepton87]

Well its 3 nVs top GPUs vs 3 of AMD's top gpu. Same comparison as anands. If you think its unfair cause of price, I agree.

It's not even AMD's 3 top GPUs. It's not the fastest triple GPU setup from AMD. 3 cards vs 3 Cards would have been slightly more fair.

 

[Madcatatlas]

Read the first page and then read the rest of the review if you dont belive Kyle and Brent when they outright state how the test was conducted.
Ill motivate you by saying this: there are two occurances of hitting the vram on the 580s, and in both cases they turn down the AA and AF to levels where its not the vram limiting the fps. But, dont take my word for it, find out yourself.


Its really not that much to ask.

 

[Madcatatlas]

And let me add, because i dont belive its been stated clearly enough:

This was with Standard position on the 6990 card and therefor a DOWNCLOCK on the 6970.

With the "Awesome" switch position, the outclassing would be even more pronounced.

So how about a title change to "downclocked TRIFIRE faster than TRISLI"?

Incredible value (although id need to be rich before i spent that much on graphic cards.)

 

[Lonyo]

Read the first page and then read the rest of the review if you dont belive Kyle and Brent when they outright state how the test was conducted.
Ill motivate you by saying this: there are two occurances of hitting the vram on the 580s, and in both cases they turn down the AA and AF to levels where its not the vram limiting the fps. But, dont take my word for it, find out yourself.


Its really not that much to ask.

Actually it is, asking people to read is too much to ask, and even when they read things they don't believe it.

Also, Obama isn't American.

 

[Madcatatlas]

hehe

 

[dac7nco]

Just sold my EVGA 580s @ $700 for the pair... I'll get a GTX550 for TmpGENC.

Daimon

 

[wahdangun]

wow what an amazing performance in there, amd have truely come a long road from completely garbage to amazing scaling,

and yes nvdia decision to use odd vram configuration have taking its victim, but its can't be avoided because big gpu strategy that nvdia use.

 

[badb0y]

Just sold my EVGA 580s @ $700 for the pair... I'll get a GTX550 for TmpGENC.

Daimon

I hope it wasn't because of this review lol.

I am quite impressed by the Crossfire scaling as well.

I still want a GTX 580 though :X

 

[Mopetar]

And yes Nvidia made a mistake with only 1.5gb of vram on the market the 580 is intended for.

I think that they needed to skimp on the VRAM in order to keep the prices down. Their chips are bigger than the Radeons and more expensive as a result, therefore they include less memory to make up the difference.

Besides, for most scenarios, 1.5 GB is enough. Anyone who needs more can buy the non-reference version. Even the [H] article pointed out that memory wasn't a limiting factor in the tests that they ran.

 

[dac7nco]

I hope it wasn't because of this review lol.

I am quite impressed by the Crossfire scaling as well.

I still want a GTX 580 though :X

No, The GTX580s run just fine on surround; I've been eyeballing the 6970 Lightnings, and I need VRAM with my resolution. Getting a pair of the Palig 580 3GB cards would cost me another $500+ even after my sale; the MSI cards will cost me $60. ATI has sold me this generation, and as a rule I don't tolerate Radeons. Color me red.

Daimon

Edit: I also like the idea of adding two additional monitors at a later date.

 

[wahdangun]

I think that they needed to skimp on the VRAM in order to keep the prices down. Their chips are bigger than the Radeons and more expensive as a result, therefore they include less memory to make up the difference.

Besides, for most scenarios, 1.5 GB is enough. Anyone who needs more can buy the non-reference version. Even the [H] article pointed out that memory wasn't a limiting factor in the tests that they ran.

no, that because nvdia use suck gddr5 memory controler, so they was forced to use larger buss. And in the end they ended up in odd vram configuration