3 way SLI/CF scaling (Not English)

[Daedalus685]

http://www.hardware.info/nl-NL/articles/amdnampoZGCa/Clash_of_the_Titans_3way_SLI_GTX_480_test/13

Check out the scaling numbers. I am floored by how well CF scales, when did they fix that?

3 480s sure is a beast though.

The review is in German, numbers are still the same.. translate if you want .

To summarize what really stands out for me, 5970s scale 65.5% on average, 3 5870s gives 41% over two on average. SLI wins rather steadily in 2 card configurations but no where close in 3. Drivers perhaps? SLI is a lot more erratic across the tests.

 

[Apocalypse23]

Interesting, I can't believe how the reviewers missed out on BFBC2 performance, also it would have been even better to compare tri-crossfire with 3 2GB 5870 E6 editions.

 

[Madcatatlas]

So how about someone actually make some..."sense" out of those numbers, with some sort of resume. For those who abhor google translator. etc

Thank you in advance

 

[MarcVenice]

The review is in Dutch...

Americans and their geography... I know Holland / The Netherlands might seem small and insignificant, but it's not the same thing as Germany, where they speak German...

Decent enough review, but they don't mention minima. And that's where the GTX 480 usually does best...

 

[Attic]

How are minima being catogarized in todays benchmarks on review sites? If a hard drive thrashes wouldn't that cause a hiccup in framerate? Is a single frame really being counted as a minima? Is anyone averaging say the 5% of lowest frames?

When I see mimimums i'm always skeptical of there validity.

 

[yh125d]

How are minima being catogarized in todays benchmarks on review sites? If a hard drive thrashes wouldn't that cause a hiccup in framerate? Is a single frame really being counted as a minima? Is anyone averaging say the 5% of lowest frames?

When I see mimimums i'm always skeptical of there validity.

We need more sites to include FPS graphs

 

[thilanliyan]

The review is in Dutch...

Americans and their geography... I know Holland / The Netherlands might seem small and insignificant, but it's not the same thing as Germany, where they speak German...

 

[Daedalus685]

The review is in Dutch...

Americans and their geography... I know Holland / The Netherlands might seem small and insignificant, but it's not the same thing as Germany, where they speak German...

Decent enough review, but they don't mention minima. And that's where the GTX 480 usually does best...

Well, I'm not American... but yeah, my ignorance of language has nothing to do with geography I'm not sure how much of it relates to never studying language and thus not having a bloody clue and how much relates to good old dyslexia.. lol. I did not mean to offend anyone from the Netherlands, you folks will have to forgive me if I cant read dutch though.. I can barely communicate in English as it is. Perhaps I will just assign meaning to numbers and communicate in mathematics from now on.. Though I'm very glad someone who can understand how to make sense of language corrected that... Though I'll have to take your word on it.

I suppose I should have said "not in English" as opposed to blindly guessing something from central/western Europe.. In fact I changed it to that


Back to the review... The thing that surprises me the most is how well crossfire scales compared to SLI. The view around these forums seems to be that SLI is far superior in that regard.. though perhaps it may just be the odd fan who claims such things. I mean the scaling of two 5970s is not much worse than two 5870s. That is something right there, there was a time when quad crossfire was more or less considered a joke, it still seems that folks think the scaling is MUCH worse than two cards. By the numbers in that review we are looking at 70% scaling for the crossfire, and 65% on the quad fire.. That is very impressive.

 

[thilanliyan]

Back to the review... The thing that surprises me the most is how well crossfire scales compared to SLI. The view around these forums seems to be that SLI is far superior in that regard.. though perhaps it may just be the odd fan who claims such things.

I think that is the generally held belief from back when it was true. XFire has come a long way from that Master/slave card business they had when it first came out.

I'm still extremely hesitant to go near multi GPU. Is microstutter still a problem?

 

[Daedalus685]

I think that is the generally held belief from back when it was true. XFire has come a long way from that Master/slave card business they had when it first came out.

I'm still extremely hesitant to go near multi GPU. Is microstutter still a problem?

I suppose everything is historic isn't it. Just strange that in an industry where things move so fast such antique ideas can still hold even in the rational folks.

Micro stutter is a "problem" for some. I have never been able to notice it, and can see a clear performance advantage (with my eye and with the fps counter) with CF on and off on my 4890s... that being said, I can't say whether 90fps looks like 90fps or not.. it might look like 60fps, but it is still better than the 40fps I might see otherwise.

I'm sure folks more sensitive to frequency changes might notice it. I honestly have no idea if there are additional features now a days that tries to synchronize the AFR such that they are evenly spaced. The only issue I find with CF is the odd bug where it won't work, and there are times when it can be a touch unstable with new drivers.

 

[mhouck]

I think that is the generally held belief from back when it was true. XFire has come a long way from that Master/slave card business they had when it first came out.

I'm still extremely hesitant to go near multi GPU. Is microstutter still a problem?

I haven't noticed it nearly as much w/ xfire 5870 as I did when I had 9800 GX2, but that may not be a fair comparison to previous xfire tech. Different companies, 2 generations apart...

 

[Daedalus685]

It is just such a funny thing to compare. I have never seen a benchmarking suit that can actually quantify it, so we only have the opinions of us folks with CF/SLI... and we all see it differently.

What I thought was also interesting was the Hardware Canucks review of the 470 SLI. They have scaling included at the end (only for two GPUs) but also include minimums. CF was scaling it's minimums as well as teh average, which years ago was not the case.. CF would give you about the same minimums at times, but much higher max. That being said, even without the same scaling on minimum from the 470s the inherent advantage of Fermi in this regard kept it firmly in the lead.

I bring that up for the interest factor.. but also because many folks historically misunderstood micro stutter as stutter, which was much more pronounced on CF given the minuscule improvement on minimums. I wonder if micro stutter is not reported as much any more as such a vast majority of people just can't see it, and those who thought they could were reporting stuttering.

 

[MarcVenice]

Well, I'm not American... but yeah, my ignorance of language has nothing to do with geography I'm not sure how much of it relates to never studying language and thus not having a bloody clue and how much relates to good old dyslexia.. lol. I did not mean to offend anyone from the Netherlands, you folks will have to forgive me if I cant read dutch though.. I can barely communicate in English as it is. Perhaps I will just assign meaning to numbers and communicate in mathematics from now on.. Though I'm very glad someone who can understand how to make sense of language corrected that... Though I'll have to take your word on it.

I suppose I should have said "not in English" as opposed to blindly guessing something from central/western Europe.. In fact I changed it to that


Back to the review... The thing that surprises me the most is how well crossfire scales compared to SLI. The view around these forums seems to be that SLI is far superior in that regard.. though perhaps it may just be the odd fan who claims such things. I mean the scaling of two 5970s is not much worse than two 5870s. That is something right there, there was a time when quad crossfire was more or less considered a joke, it still seems that folks think the scaling is MUCH worse than two cards. By the numbers in that review we are looking at 70% scaling for the crossfire, and 65% on the quad fire.. That is very impressive.

I wasn't trying to be offensive, so hopefully you didn't take it that way

I'm Dutch though, so I'm pretty sure it's Dutch, and not German. Americans often mistake Dutch for German, which isn't exactly offensive, but it is a bit annoying

As to how minima are recorded, obviously there are hiccups once in a while. If you run a benchmark often enough, you can filter those out. If they are on the other hand persistent they should be mentioned. In a shooter, there's nothing worse then a huge framedrop during a firefight...

 

[Daedalus685]

I wasn't trying to be offensive, so hopefully you didn't take it that way

I'm Dutch though, so I'm pretty sure it's Dutch, and not German. Americans often mistake Dutch for German, which isn't exactly offensive, but it is a bit annoying

As to how minima are recorded, obviously there are hiccups once in a while. If you run a benchmark often enough, you can filter those out. If they are on the other hand persistent they should be mentioned. In a shooter, there's nothing worse then a huge framedrop during a firefight...

Heh, no offence at all sir.

As for the minimums, Hardware canucks http://www.hardwarecanucks.com/foru.../30999-nvidia-geforce-gtx-470-sli-review.html showed the advantage of scaling holds for minimums as well.. However, the huge lead for fermi in this was not cut into much.

Minimums are certainly important though, mind you I really wish folks that quoted them would include a time graph as context.