Fermi's lead over Cypress shrinks w/ new drivers

[v8envy]

Well, it's gamespot, but here you go:

http://www.gamespot.com/features/6264201/p-2.html

A few games where 3x CF beats out 2x 480 SLI, and is once again beaten by 3x 480 SLI.

But in general you are correct, SLI scaling is much better than CF scaling. There are plenty of cases where CF does nothing if you're lucky and actually costs performance if you're not.

 

[ZimZum]

I dont understand why reviewers use time demos to begin with. Rather than actual gameplay performance. At the very least include the realworld gaming numbers along side the time demo.

 

[luv2increase]

I dont understand why reviewers use time demos to begin with. Rather than actual gameplay performance. At the very least include the realworld gaming numbers along side the time demo.

The experiment is not valid otherwise. It is impossible to reproduce a real-world gaming benchmark to be held 100% constant each and every time.

The methodology that the reviewers used to test the 10.5 drivers vs the 257 drivers would never hold up as being scietifically sound due to the experiment not "truly" being "100%" reproducible every time.

 

[Lonyo]

I dont understand why reviewers use time demos to begin with. Rather than actual gameplay performance. At the very least include the realworld gaming numbers along side the time demo.

Because it's easier to set a computer to run a timedemo?

 

[RussianSensation]

The experiment is not valid otherwise. It is impossible to reproduce a real-world gaming benchmark to be held 100% constant each and every time.

Run the real-time game section 3x and average the scores. Both companies have been known to optimize performance for canned demo runs. In reality, most demos are not representative of actual gameplay. While this would make it harder to compare the results across various websites which test the hardware, at least the actual reviews would be a more realistic representation of what the gaming experience will be like.

Also, please go check your benchmarks again. Outside of Far Cry 2, GTX480 is nowhere near 25% faster than a 5870: http://ixbtlabs.com/articles3/video/gf100-3-p3.html At 2560x1600, on average, it's about 10-15% faster: http://www.techpowerup.com/reviews/Zotac/GeForce_GTX_480_Amp_Edition/29.html

 

[Daedalus685]

The experiment is not valid otherwise. It is impossible to reproduce a real-world gaming benchmark to be held 100% constant each and every time.

The methodology that the reviewers used to test the 10.5 drivers vs the 257 drivers would never hold up as being scietifically sound due to the experiment not "truly" being "100%" reproducible every time.

Repeatability is not measured by exactness of each point.

In a review like the one in question the fps/time graphs are shown which gives a very good idea of the performance comparison (they follow the same trend).

Obviously more runs are nice, and as much like each other as possible is best.

It is likely best to record your own time demo, but the mere fact of not being able to do the 'exact' same thing each time does not at all make the tests unscientific, it just makes it better at a different thing.

Mind you, some of those graphs needed some running averaging..

 

[Keysplayr]

Glad to see this thread getting back on topic, but just a forewarning. Any further comments like "You sound just like Wreckage", or "You sound just like an ATI billboard" will get you a few days off. I'm getting really tired of the childish subtle snubs and rubs. SUBJECT MATTER. Nothing more than that please.
Thanks in advance,
Anandtech Moderator - Keysplayr

 

[luv2increase]

Also, please go check your benchmarks again. Outside of Far Cry 2, GTX480 is nowhere near 25% faster than a 5870: http://ixbtlabs.com/articles3/video/gf100-3-p3.html At 2560x1600, on average, it's about 10-15% faster: http://www.techpowerup.com/reviews/Zotac/GeForce_GTX_480_Amp_Edition/29.html

Why are you telling me to check the benchmarks again?

Nowhere did I ever state the 480 was 25% faster than the 5870. I said when both the 480 and 5870 are in dual-GPU configs, the 480 will on average be more than 25% faster than the 5870 due to the superior scaling of the 480 in SLI. There wasn't any benchmarks for CF or SLI in the links you provided either. I don't know what I was supposed to be looking at. lol

 

[SolMiester]

I don't agree. What it shows is: if you put the competing cards into two otherwise identical machines with water cooling and have a typical person play a wide selection of games at playable settings they will not be able to reliably determine which hardware is in which machine.

If that's the case then why put up with one of the solutions broiling your toes while it deafens you and costs more to both buy and operate?

If using WC as you stated, your 2nd comment is irrelevant!

 

[Puffnstuff]

My fermi's can't hear you they're too busy swimming right now.

 

[HurleyBird]

I said when both the 480 and 5870 are in dual-GPU configs, the 480 will on average be more than 25% faster than the 5870 due to the superior scaling of the 480 in SLI.

Again, that's true... But such a system would be like having a space heater under your desk. Not really practical. Same thing goes for Trifire. I pity the fool who runs Tri-480's for anything other then benchmarking.

Nvidia's superior scaling will be practical once they get power consumption down, but not really until then.

 

[evolucion8]

Why are you telling me to check the benchmarks again?

Nowhere did I ever state the 480 was 25% faster than the 5870. I said when both the 480 and 5870 are in dual-GPU configs, the 480 will on average be more than 25% faster than the 5870 due to the superior scaling of the 480 in SLI. There wasn't any benchmarks for CF or SLI in the links you provided either. I don't know what I was supposed to be looking at. lol

Which proves that the GTX 480 doesn't scale "much better" than Crossfire. SLI scaling is just more consistent and slightly faster where the HD 5x00 scaling doesn't always gain that much. For some reason I think that HD 4x00/ HD 3x00 series had better scaling.

I can say that in the best case scenario, SLI can scale up to 10%-15% in average better than Crossfire. Why? Because if the GTX 480 in average is between 10%-15% faster than a single HD 5870, 10x2 GTX480=20% higher performance than a HD 5870 CF or 15%x2 GTX480=30% higher performance than an HD 5870CF.

PS: Metro 2033 anti aliasing bug also happens with my setup. Activating regular 4x MSAA blurries all the screen washing away all the detail. I think that's more game engine related than driver related. AAA shows much sharper details.

 

[SolMiester]

Which proves that the GTX 480 doesn't scale "much better" than Crossfire. SLI scaling is just more consistent and slightly faster where the HD 5x00 scaling doesn't always gain that much. For some reason I think that HD 4x00/ HD 3x00 series had better scaling.

I can say that in the best case scenario, SLI can scale up to 10%-15% in average better than Crossfire. Why? Because if the GTX 480 in average is between 10%-15% faster than a single HD 5870, 10x2 GTX480=20% higher performance than a HD 5870 CF or 15%x2 GTX480=30% higher performance than an HD 5870CF.

PS: Metro 2033 anti aliasing bug also happens with my setup. Activating regular 4x MSAA blurries all the screen washing away all the detail. I think that's more game engine related than driver related. AAA shows much sharper details.

What?...Multi-GPU scaling SLI v CF has nothing to do with what % 1 GPU is better against another!

 

[luv2increase]

What?...Multi-GPU scaling SLI v CF has nothing to do with what % 1 GPU is better against another!

I don't think evolucion8 really knows what he is talking about here. The first thing I found odd was that he contradicted his first sentence by his second sentence. I actually had a chuckle when I read it.

I'd put a rough guestimate on it and say that @ stock speeds, 2 x 480s in SLI will be on average around 40% faster than 2 x 5870s in CF. That is just on the FPS front. You'd also obviously be able to crank up the quality settings as well.

@HurleyBird
I honestly can't tell the difference in room temperature from when I had 2 x 5870s installed compared to these 2 x 480s. I'm talking about when just browsing the web and heavy gaming sessions. If you had them, you'd know what I mean. Until then...

 

[MoMeanMugs]

If using WC as you stated, your 2nd comment is irrelevant!

How is that irrelevant? If you're using watercooling over air, you're dumping more heat out of the case. Watercooling doesn't magically make heat disappear.

 

[SolMiester]

How is that irrelevant? If you're using watercooling over air, you're dumping more heat out of the case. Watercooling doesn't magically make heat disappear.

Well, that's actually correct regarding heat dissipation, however the process or dissipation into the air isn't as...concentrated?....ie:, radiator has bigger surface area, isn't getting blasted out of small grid like a card

 

[tincart]

Strange how everyone is now avoiding the elephant in the room. This thread is (nominally) about the Techspot findings regarding real-world performance of 4 cards. I can't fathom how CF vs. SLI scaling is taking priority in discussion.

 

[evolucion8]

I don't think evolucion8 really knows what he is talking about here. The first thing I found odd was that he contradicted his first sentence by his second sentence. I actually had a chuckle when I read it.

I'd put a rough guestimate on it and say that @ stock speeds, 2 x 480s in SLI will be on average around 40% faster than 2 x 5870s in CF. That is just on the FPS front. You'd also obviously be able to crank up the quality settings as well.

I think that my post couldn't be clear, but that's ok. Is off topic nevertheless

What?...Multi-GPU scaling SLI v CF has nothing to do with what % 1 GPU is better against another!

In english please :awe:

Back in topic, you are right tincart, I can say for sure that this GTX 480 vs HD 5870 is pretty much the same fight like the GTX 280 had vs the HD 4870, except that nVidia's solution didn't consume much more power at time like's doing now. GTX = Guzzling The Xtreme 480 of Watts in power consumption :awe: J/K.

 

[luv2increase]

Strange how everyone is now avoiding the elephant in the room. This thread is (nominally) about the Techspot findings regarding real-world performance of 4 cards. I can't fathom how CF vs. SLI scaling is taking priority in discussion.

It is actually more relevant when you think of it. I've already stated the reasons why in previous posts, but I'll be happy to go over it again.

The 5870 and GTX480 are "very" expensive graphics cards. To buy an expensive graphics card you have to:

1. Have the money
2. Have an unsatiable appetite for increases in gaming performance

Due to facts 1 & 2, MANY people buying the 5870 and GTX480 will buy more than just one for multi-GPU gaming goodness. Due to this, seeing benchmarks of single-GPU performance differences between the 5870 and GTX480 simply don't matter if you have 2 x 480s or 2 x 5870s or you want or plan to buy 2 x 480s or 2 x 5870s. Due to this, multi-GPU performances differences between the 5870 and GTX480 are more important than their single-GPU config results.

Now, if the review would have tested multi-GPU differences between the 5870 and GTX480 using real-world in real-time benchmarking using FRAPS, we'd really have ourself a superbly outstanding review!

 

[ronnn]

Hard[OCP] showed at launch that there was very little difference in the cards when it came to real world gaming, not canned bechmarks. This just reinforces that conclusion.

What he said.

 

[luv2increase]

The Toms series of reviews about building a balanced gaming PC (no GTX400 in there though) also show that the GTX280 and GTX295 seems to depends on more CPU power to reach its full potential.

http://www.tomshardware.com/reviews/balanced-gaming-pc-overclock,2625.html and previous.

I think that DX11 performance is what matters most nowadays. Both the cards can easily play all DX9 and DX10 titles fairly well. I just wonder how each are going to do over the next couple of years in upcoming heavily tessellated titles. That will be the real kicker. :awe:

 

[busydude]

Due to facts 1 & 2, MANY people buying the 5870 and GTX480 will buy more than just one for multi-GPU gaming goodness. Due to this, seeing benchmarks of single-GPU performance differences between the 5870 and GTX480 simply don't matter if you have 2 x 480s or 2 x 5870s or you want or plan to buy 2 x 480s or 2 x 5870s. Due to this, multi-GPU performances differences between the 5870 and GTX480 are more important than their single-GPU config results.

These so called "facts" may be true in your case, but what you are inferring to as facts are actually assumptions. You cannot simply say that people who already have 5970/480 will buy a second one. Many people are in fact waiting for the Septembers refresh from ATI.

Multi-GPU is a niche market. I haven't come across a single review which mentions micro stutter when dealing with Multi-GPU's. Multi-GPU sure has its advantages but one should not discount its drawbacks too.

 

[akotlar]

I was very skeptical about Fermi, like most of you, but benchies that I'm seeing show it to be consistently above Cypress. It doesn't really matter that it is only 2 months old, both are the current top solutions. Users switching between the two say that 1 480 is perceptively equal to 1 5870, but that SLI provides a more consistent, noticeably faster framerate than Crossfire. Of course, there is a $100 premium per card, but essentially some users are saying that Crossfire isn't really worth it due to lack of significant gameplay benefits in the most demanding titles, when compared to 2 Fermis or a single 5870.

 

[HurleyBird]

It is actually more relevant when you think of it. I've already stated the reasons why in previous posts, but I'll be happy to go over it again.

The 5870 and GTX480 are "very" expensive graphics cards. To buy an expensive graphics card you have to:

1. Have the money
2. Have an unsatiable appetite for increases in gaming performance

Due to facts 1 & 2, MANY people buying the 5870 and GTX480 will buy more than just one for multi-GPU gaming goodness.

Not at all. Most people only want a single GPU as to avoid the issues that multi-GPU solutions have. You hear this all the time in the enthusiast community. If anything, multi-GPU is more popular in the midrange since that is where you have the most value (eg. two 5770's offer similair performance to a single 5870 when you have games with good scaling, offer twice the tesselation perfmance, and save you ~100 dollars.)

Running two 480's is a very expensive and scaldingly hot proposition. You don't really see SLIed 480's outisde of competitive benchmarking. In terms of what is practical, Nvidia's SLI scaling advantage really doesen't help it because you won't notice the extra performance in 99% of cases, although uncomfort due to excess heat is going to be noticed whenever you're gaming.

 

[luv2increase]

These so called "facts" may be true in your case, but what you are inferring to as facts are actually assumptions. You cannot simply say that people who already have 5970/480 will buy a second one. Many people are in fact waiting for the Septembers refresh from ATI.

Multi-GPU is a niche market. I haven't come across a single review which mentions micro stutter when dealing with Multi-GPU's. Multi-GPU sure has its advantages but one should not discount its drawbacks too.

I see you misunderstood the point of my post. This isn't just about the 5870 and 480. The is about the cream of the crop of every generation release from ATI and Nvidia. MANY people who buy the absolute cream of the crop cards will also buy a second or third one for SLI/CF. That isn't my opinion. That is a fact. I would almost say that more people buying the cream of the crop will buy more than 1 rather than just buying 1. I'm not talking about mid-range or low-end cards here. I'm not even talking about the second in-line cards like the 470 and 5850.

It is interesting you brought up micro-stuttering. Micro-stuttering is an issue that plagues multi-GPU setups. I really can't say I had any problems with that on the 5870s or 480s to be honest.

Your thinking is back in 2006 or 2007. I say this because you talk about multi-GPU rigs being a niche market. Well, that really isn't the case anymore when dealing with enthusiasts and the cream of the crop cards. In case you haven't noticed, the review in this thread is comparing the cream of the crop cards. I'm not referencing mid-range or low-end cards. I also say this because micro-stuttering isn't nearly as bad as it used to be anymore with ATI's and Nvidia's latest offerings.

Lastly, I must sincerely ask you the question if you have or have had any 5xxx or 4xx series cards?