GeForce GTX 580 vs. Radeon HD 5970 2GB

[SlowSpyder]

I don't know what we are arguing about here.

The 5870 is/was a great card. But it's about 14 months old now. The 5970 is/was a great card, but it has all the issues that come with dual GPU, and AMD's CrossFire for the 5xxx didn't exactly gain a stellar reputation since the 5970 debuted.

The GTX580 isn't a perfect part, but it cleaned up a lot of the issues that the GTX480 had, mainly noise and heat/power. To me the GTX580 is what the 480 was supposed to be, it shows how flawed the 480 was. So on one hand, we have to hand it to Nvidia, the GTX580 is quiet and gives the 5970 a run for it's money, and is probably an overall better gaming experience the majority of the time. On the other hand it's kind of funny to me that they finally got out what they announced way back around the 5870 launch. But if AMD keeps delaying Cayman, that may actually play out in Nvidia's advantage, they'll just need this tweaked and fully functioning Fermi to have the best high end part.

 

[cusideabelincoln]

It's unimpressive because the numbers you are quoting are on the low end (i.e., you are likely ignoring the most intensive games like Metro 2033 and STALKER.)

TechSpot Review:

"In the 13 games we tested, the GeForce GTX 580 was on average a match to the Radeon 5970, 25% faster than the GeForce GTX 480 and 46% faster than the Radeon 5870. If you ignore the Tessellation-enhanced results from Metro 2033 and S.T.A.L.K.E.R.: Call of Pripyat, the GTX 580 was 7% slower than the Radeon 5970 and 33% faster than the Radeon HD 5870."

Computerbase.de Review - 2560x1600 4AA/16AF
GTX580 is 53% faster than HD5870

Bit-Tech.net Review
- In Dirt 2, the GTX 580 1.5GB was an incredible 63 per cent faster than ATI’s fastest single-GPU card, the Radeon HD 5870 1GB.
- In BF:BC2, at 1,920 x 1,200 with 4x AA the GTX 580 1.5GB managed a minimum of 48fps, a huge increase [60% faster] over the 30fps minimum of the HD 5870 1GB. While the HD 5970 2GB also put in some impressive numbers at 1,920 x 1,200, we encountered serious micro-stuttering at 2,560 x 1,200, meaning that the game was unplayable.

https://spreadsheets.google.com/ccc...lU2QXlQVEZoRnJlQ2ZUV3c&hl=en&authkey=CJKDzrcH

 

[RussianSensation]

https://spreadsheets.google.com/ccc...lU2QXlQVEZoRnJlQ2ZUV3c&hl=en&authkey=CJKDzrcH

Great work. :thumbsup:

DX11 average = 48.5%
DX11 average w/o Unigine and Hawx 2 = 39%

That's larger than the performance difference was between HD4870 $299 and GTX280 $499 (after $150 price adjustment).

 

[MrK6]

What? If you read the reviews around the internet, 30% is indeed the average amongst all settings and games... The examples you picked are so cherry picked its not funny

I could do the same and pick F1 2010 where the 5870 virtually ties the 580, and there goes your 50% out the window

You're going to see a lot of cheerleading no matter what happens. People will see what they want for whatever reason. If you throw out a statistic they don't like, they'll ignore it, insult you, or keep changing variables/constraints until they think they're right. There's little substance left on this forum.

 

[RussianSensation]

You're going to see a lot of cheerleading no matter what happens. People will see what they want for whatever reason. If you throw out a statistic they don't like, they'll ignore it, insult you, or keep changing variables/constraints until they think they're right. There's little substance left on this forum.

What? If you read the reviews around the internet, 30% is indeed the average amongst all settings and games... The examples you picked are so cherry picked its not funny

TechSpot averaged 13 modern games, where GTX580 was 46% faster.
Computerbase.de averaged 11 modern games.

How is this "cherry-picking" 1-2 specific games? I guess unless they include 30 games, the review is invalid?

Cool isn't it?

The performance difference has been shown ~ 6% at 4AA and 5% at 8AA on the HD6870 when using HQ. So that's 53% - 6% = still 47% faster for the 580.

 

[Lonbjerg]

Cool isn't it?

Oh snap...

http://forums.anandtech.com/showthread.php?t=2117420

Welcome to the present 😀

 

[cusideabelincoln]

Great work.

DX11 average = 48.5%
DX11 average w/o Unigine and Hawx 2 = 39%

That's larger than the performance difference was between HD4870 $299 and GTX280 $499 (after $150 price adjustment).

Yes it is bigger; performance is more analogous to the 4870 (512?) to the GTX 285.

 

[GaiaHunter]

The performance difference has been shown ~ 6% at 4AA and 5% at 8AA on the HD6870 when using HQ. So that's 53% - 6% = still 47% faster for the 580.

Plus HD5870 has always been tested with HQ and you didn't have a problem with that in the last 12 months. Suddenly it's a problem for you?

That isn't the problem.

Older review http://www.computerbase.de/artikel/...vidia-geforce-gtx-460/4/#abschnitt_testsystem

Treibereinstellungen: ATi-Grafikkarten (RV7x0, RV8x0)

Catalyst A.I.: Standard
Mipmap Detail Level: High Quality

See - default driver settings with Catalyst A.I.: Standard vs OFF.

Those figures presented are only valid for 6800 series.

 

[GaiaHunter]

Oh snap...

http://forums.anandtech.com/showthread.php?t=2117420

Welcome to the present 😀

Cool.

Since you read that thread, you probably have seen my reserves to the fact there was no newer games examples.

Actually there wasn't even a single game that was actually part of their benchmark.

Can you provide examples or just link to stuff I'm obviously aware?

 

[RussianSensation]

See - default driver settings with Catalyst A.I.: Standard vs OFF.

Those figures presented are only valid for 6800 series.

GTX580 is 15-20% (low-high) faster than a GTX480
GTX480 is 15-20% (low-high) faster than an HD5870

Take the mid-point of each = 17.5% average

Therefore,

HD5870 (85%) --> GTX480 (100%) --> GTX580 (117.5%)
----------------> 17.5% increase -----> 17.5% increase

117.5 / 85 = 38% (only 1% off average from the Reviews summary chart provided by cusideabelincoln). No matter how we get there, we still get 38-39%; which supports the notion that the 30% estimate in this thread was on the low-end.

 

[GaiaHunter]

gtx580 is 15-20% (low-high) faster than a gtx480
gtx480 is 15-20% (low-high) faster than an hd5870

take the mid-point of each = 17.5% average

therefore,

hd5870 (85%) --> gtx480 (100%) --> gtx580 (117.5%)
----------------> 17.5% increase -----> 17.5% increase

117.5 / 85 = 38% (only 1% off average from the reviews summary chart provided by cusideabelincoln). No matter how we get there, we still get 38-39%; which supports the notion that the 30% estimate in this thread was on the low-end.

computerbase.de review - 2560x1600 4aa/16af
gtx580 is 53% faster than hd5870

38%, 53%.

 

[RussianSensation]

38%, 53%.

That's because Computerbase.de used 8AA in some of the games where the 5870 ran into 1GB VRAM limitation.

5870 vs. 580

Splinter Cell Conviction - 8AA/16AF - 15 vs. 44.5
Crysis Warhead - 8AA/16AF - 4.9 vs. 19
Battleforge - 8AA/16AF - 18.9 vs. 41.5
Metro 2033 - 4AA/16AF - 0.6 :\ vs. 27.7
STALKER:Cop - 4Aa/16AF - 21.6 vs. 35.6

 

[tannat]

GTX580 is 15-20% (low-high) faster than a GTX480
GTX480 is 15-20% (low-high) faster than an HD5870

Take the mid-point of each = 17.5% average

Therefore,

HD5870 (85%) --> GTX480 (100%) --> GTX580 (117.5%)
----------------> 17.5% increase -----> 17.5% increase

117.5 / 85 = 38% (only 1% off average from the Reviews summary chart provided by cusideabelincoln). No matter how we get there, we still get 38-39%; which supports the notion that the 30% estimate in this thread was on the low-end.

I'm sorry but the bold part looks like plain wishful thinking and blindfolded use of basic multiplication. I really feel sorry for multiplication to be used in such a degrading manner🙂

It's just pushing noise around.

Check the covariance between results of same games from different sites. That covariance is a measure of the reliability of the data.

BTW the spreadsheet prove once again why [H] is so extremely valuable for decent real life perspective to contrast the the canned bench sites. - Even if people may find their testing confusing.

It's hard to draw any other conclusion than the difference between GTX580 and HD5870 being 20 % when gaming but 40% when benching...

 

[RussianSensation]

It's hard to draw any other conclusion than the difference between GTX580 and HD5870 being 20 % when gaming but 40% when benching...

{H} tested 5 games in their review (F1 2010, BF:BC2, Metro 2033, Civ5 and Medal of Honor). So, you are drawing your "average" 20% conclusion on just 4-5 games? :hmm:

Also, if you actually read other reviews, most of them already record real-world performance with FRAPs (Bit-tech.net, Hardware Canucks, PCGamesHardware, gamegpu.ru just off the top of my head). To imply all websites outside of {H} only do "canned" benchmarks is misleading when it's clearly not true. And just because a benchmark is built into the game, doesn't mean it's not representative of actual gameplay. Some are and some aren't. There is no clear cut rule. Dirt 2 benchmark is 100% represenative of real world gameplay for example.

You are now arguying that only {H} does proper videocard reviews, discounting all other reviews.

I recommend you spent 30 min or so on gamegpu.ru. They ONLY test games with manual benches (i.e., 10 min manual walk-through sections of games). Their results show GTX480 consistently beating the 5870 by 15-25%.

So before you jump to conclusions, consider that there are other websites out there that don't do any canned benchmarks that still show 5870 behind the 480 (especially in non-mainstream games), nevermind the 580, in actual gameplay.

Every single benchmark below is a Manually recorded run at 1920x1080 VHQ - 4AA or 8AA

480 vs. 5870

1. Test Drive Unlimited 2 = + 39%
2. Two Worlds 2 = + 48%
3. COD : Black Ops = + 25%
4. ArcaniA: Gothic 4 = + 9%
5. Medal of Honor = equal at 62 fps (but GTX480 has a 24% advantage in Minimum framerates)
6. Front Mission Evolved = + 12%
7. Lara Croft and the Guardian of Light = + 21%
8. Darksiders = + 40%
10. F1 2010 = - 8%
11. Final Fantasy 14 = + 28%
12. Kane and Linch 2 = - 10%
13. Lost Planet 2 = + 95%
14. Mafia 2 (without PhysX) = + 45%
15. Starcraft 2 = + 14%
16. ArmA 2 = + 19%
17. Ghost Sniper Warrior = + 8%

Interesting how HD5870 performs even worse with more and more games being included, each of which tested was tested manually. I guess it makes sense why HardOCP included BF:BC2 and F1 2010 into their review - the few games remaining where 5870 has any chance at all against the 580. But since HD5870 costs $270 on Newegg, it's obviously completely unfair to expect the 5870 to be as fast as a $499 card. So a fair competitor to the 580 is the 6970.

 

[Jdawg84]

{H} tested 5 games in their review (F1 2010, BF:BC2, Metro 2033, Civ5 and Medal of Honor). So, you are drawing your "average" 20% conclusion on just 4-5 games? :hmm:

Also, if you actually read other reviews, most of them already record real-world performance with FRAPs (Bit-tech.net, Hardware Canucks, PCGamesHardware, gamegpu.ru just off the top of my head). To imply all websites outside of {H} only do "canned" benchmarks is misleading when it's clearly not true. And just because a benchmark is built into the game, doesn't mean it's not representative of actual gameplay. Some are and some aren't. There is no clear cut rule. Dirt 2 benchmark is 100% represenative of real world gameplay for example.

You are now arguying that only {H} does proper videocard reviews, discounting all other reviews.

I recommend you spent 30 min or so on gamegpu.ru. They ONLY test games with manual benches (i.e., 10 min manual walk-through sections of games). Their results show GTX480 consistently beating the 5870 by 15-25%.

So before you jump to conclusions, consider that there are other websites out there that don't do any canned benchmarks that still show 5870 behind the 480 (especially in non-mainstream games), nevermind the 580, in actual gameplay.

Every single benchmark below is a Manually recorded run at 1920x1080 VHQ - 4AA or 8AA

480 vs. 5870

1. Test Drive Unlimited 2 = + 39%
2. Two Worlds 2 = + 48%
3. COD : Black Ops = + 25%
4. ArcaniA: Gothic 4 = + 9%
5. Medal of Honor = equal at 62 fps (but GTX480 has a 24% advantage in Minimum framerates)
6. Front Mission Evolved = + 12%
7. Lara Croft and the Guardian of Light = + 21%
8. Darksiders = + 40%
10. F1 2010 = - 8%
11. Final Fantasy 14 = + 28%
12. Kane and Linch 2 = - 10%
13. Lost Planet 2 = + 95%
14. Mafia 2 (without PhysX) = + 45%
15. Starcraft 2 = + 14%
16. ArmA 2 = + 19%
17. Ghost Sniper Warrior = + 8%

Interesting how HD5870 performs even worse with more and more games being included, each of which tested was tested manually. I guess it makes sense why HardOCP included BF:BC2 and F1 2010 into their review - the few games remaining where 5870 has any chance at all against the 580. But since HD5870 costs $270 on Newegg, it's obviously completely unfair to expect the 5870 to be as fast as a $499 card. So a fair competitor to the 580 is the 6970.

With the 5870 THAT low. How does a Pair in Xfire compare to the 580?

 

[RussianSensation]

With the 5870 THAT low. How does a Pair in Xfire compare to the 580?

AnandTech has tested 580 SLI vs. 5870 in CF.

At 2560x1600, 580 SLI is faster than 5870 CF by 47% in Crysis, 48% in BattleForge, 42% in HAWX, tied in Civ5 (CPU limited), 25-50% in BF:BC2 (chase - waterfall benches), 29% in STALKER:CoP, 41% in Dirt 2, 41% in Mass Effect 2, but only 12% faster in Wolfenstein and loses in Metro 2033.

I don't think it's fair to compare HD5870 CF which you can get for $540 vs. $1000 GTX580 in SLI. HD5870 CF or even 6870 CF still provide superior value since GTX580 SLI is nowhere near 2x faster despite 2x the cost. But diminishing returns are generally expected for the fastest setups.

 

[Jdawg84]

I don't think it's fair to compare HD5870 CF which you can get for $540 vs. $1000 GTX580 in SLI.

What i meant was 5870 in X fire vs a single 580 since the prices are right around each other.

Edit - I can't think today. 5970 vs 580 is pretty much the results i was looking for

 

[tannat]

{H} tested 5 games in their review (F1 2010, BF:BC2, Metro 2033, Civ5 and Medal of Honor). So, you are drawing your "average" 20% conclusion on just 4-5 games? :hmm:

Also, if you actually read other reviews, most of them already record real-world performance with FRAPs (Bit-tech.net, Hardware Canucks, PCGamesHardware, gamegpu.ru just off the top of my head). To imply all websites outside of {H} only do "canned" benchmarks is misleading when it's clearly not true. And just because a benchmark is built into the game, doesn't mean it's not representative of actual gameplay. Some are and some aren't. There is no clear cut rule. Dirt 2 benchmark is 100% represenative of real world gameplay for example.

You are now arguying that only {H} does proper videocard reviews, discounting all other reviews.

No, I said the inconsistencies in the data in the spreadsheet showed that you were working with noise. The number of games does not help if you're spitting out noise. Look at it as taking a picture of a a wall in a dark room. Now take 100 pictures. Did they help you learn the color of the wall?

Why do you bring up testsites that are not in the spreadsheet? Did you look at the data critically or did you just use it? The data comes from canned benchmark sites.

I would think that one possible explanation the data is so inconsistent between games and testsites is that it's only at 2560x1600/highest quality canned bench. It's begging for trouble to superpose that to general performance.

BTW, you happened to name bit-tech. They are actually in the spreadsheet. agreeing well with [H] results. Done deal.

Hardware canucks are also there but they use a little of everything so one have to take a look at the source input...

I don't have a horse in the race. I just said that the data on it own shows that there is too much noise.

And that the fact the data being noisy makes a good case for real game reviews.

 

[RussianSensation]

And that the fact the data being noisy makes a good case for real game reviews.

I agree that it would be much better if websites stopped using canned benchmarks. Sometimes they may not have enough time to create their manual runs. I think ABT only got their cards 1 week before release of their GTX580 review. That doesnt' give them enough time to test the cards for 10 min/game through a manual run x 5-6 cards x 15-20 games.