• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

XFire scaling on 6000 series ..

blackninja

blackninja

blackninja

Associate
Joined
13 Nov 2007
Posts
1,483
The Xfire scaling is very good on the new ATI cards, so much so that the new mid-range in Xfire is not that far off my overclocked 5970 with its 2x5870 GPUs on board ...

Is the Xfire connector the same as on the 5k series? Is the improved scaling down to hardware or software - are ATI holding back something from previous generation cards to make the 6K series look more attractive like with the MAA option .. just being curious ???
 
Is the xfire scaling that much better?

A couple of early reviews were almost orgasmic about it but subsequent ones have been shown more normal levels of scaling (better than the 5 series but not crazily so), added onto that the min fps when shown isn't always very impressive.
 
Last edited:
Multi-GPU scaling performance will always be a combination of hardware (GPU architectural design) and software (driver-level). Driver-level optimisations can help to achieve the best from the hardware, but the low-level hardware design will always be the driving force that determines what is possible in software. In this generation, AMD has introduced a little more modularity into the GPU design, which is likely to aid with the scaling of the architecture (either in the multi-GPU sense, or in the sense of increasing the number of shaders etc in the GPU and maintaining near-linear performance).

I highly doubt that there is anything "being held back" from the 5-series with regards to crossfire performance.
 
If AMD are concentrating on two card set-ups only for xfire then I would imagine the improved scaling should be at least more than minimal. By most accounts seemingly a reasonable improvement.
 
Last edited:
oweneades said:
HD68xx scaling is roughly on par with that shown by GTX460 in SLI.

Interestingly HD57XX scaling is also very good so I wonder if it's a cpu limitation (to a degree) holding back the HD58xx scaling.
Click to expand...

http://www.anandtech.com/show/3987/...enewing-competition-in-the-midrange-market/14

Well from that theres three things to see, a 5870 is a full 20% faster than a 6870 in single card, in xfire, the 6870 is now only 3% behind, so in BC2 at least, its got hugely better scaling.

Also keep in mind, 55.9 in single card, 100% increase would be 111.8fps, it gets 111.5fps, thats near enough 100% speed increase, you can not do better than that, its simply superb.

The 5870 goes up just under 80%, so essentially the 6870 scales 20% better than the 5870, thats incredibly good.

Similar to 460gtx sli scaling, single card 44.8 vs 76.6 in sli, which is basically bang on 71% better.

Sorry but 71% isn't in the same ballpark as give or take 100% scaling, likewise 5870 scaling beats the 460gtx scaling.


I'm sure I'll be accused of ignoring a review, I'm tired, I clicked the link, I checked the first one, a very quick look at Stalker shows 33.x vs 57.x for 460gtx sli scaling, the 6870 goes from 40.2 to 78.x. Okay I did the maths, 460gtx sli gives a 75% boost, the 6870xf is almost a 95% boost. A pretty large difference.


Comparitively the 5870 starts off 15% ahead of the 6870, and in xfire is behind it, its scaling under 70%.

The 6870 is offering some 20-30% better scaling in xf, than the 5870, and its a good 20-30% better than 460gtx scaling aswell. Sli scaling in general I'd guess as the 470gtx shows better scaling than the 5870 by a little, but worse than the 6870 aswell.

If Nvidia haven't bumped up Sli scaling somehow with the 580gtx, Xfire cards could end up gaining speed, which is frankly fantastic, especially as SLI drivers, and drivers in general is supposed to be "Nvidia's" thing.

EDIT:- ok I went further, but not much, Dirt 2 the 460/6870 both scale very well and both are very close to each other in single and dual card, both beat 470gtx scaling and 5870 scaling quite easily. but still not be a huge margin. Its slightly hard to tell as I'm not sure where the cpu limit would be, but theres a difference between all the sli/xf's that you'd expect given scaling of the 470gtx in previous reviews, IE not as good.

For that it seems pretty clear that best case scenario AMD scale noticeably better than SLI, and worst case they are about on par. I'm sure theres games where SLI has a clear lead, though if it has a clear lead in anything but Lost Planet 2, Metro 2033 I don't know.

6870 scaling is a dramatic step forward compared to the 5870, which itself wasn't doing half badly lately.
 
Last edited:
drunkenmaster said:
http://www.anandtech.com/show/3987/...enewing-competition-in-the-midrange-market/14

Well from that theres three things to see, a 5870 is a full 20% faster than a 6870 in single card, in xfire, the 6870 is now only 3% behind, so in BC2 at least, its got hugely better scaling.

Also keep in mind, 55.9 in single card, 100% increase would be 111.8fps, it gets 111.5fps, thats near enough 100% speed increase, you can not do better than that, its simply superb.

The 5870 goes up just under 80%, so essentially the 6870 scales 20% better than the 5870, thats incredibly good.

Similar to 460gtx sli scaling, single card 44.8 vs 76.6 in sli, which is basically bang on 71% better.

Sorry but 71% isn't in the same ballpark as give or take 100% scaling, likewise 5870 scaling beats the 460gtx scaling.


I'm sure I'll be accused of ignoring a review, I'm tired, I clicked the link, I checked the first one, a very quick look at Stalker shows 33.x vs 57.x for 460gtx sli scaling, the 6870 goes from 40.2 to 78.x. Okay I did the maths, 460gtx sli gives a 75% boost, the 6870xf is almost a 95% boost. A pretty large difference.


Comparitively the 5870 starts off 15% ahead of the 6870, and in xfire is behind it, its scaling under 70%.

The 6870 is offering some 20-30% better scaling in xf, than the 5870, and its a good 20-30% better than 460gtx scaling aswell. Sli scaling in general I'd guess as the 470gtx shows better scaling than the 5870 by a little, but worse than the 6870 aswell.

If Nvidia haven't bumped up Sli scaling somehow with the 580gtx, Xfire cards could end up gaining speed, which is frankly fantastic, especially as SLI drivers, and drivers in general is supposed to be "Nvidia's" thing.

EDIT:- ok I went further, but not much, Dirt 2 the 460/6870 both scale very well and both are very close to each other in single and dual card, both beat 470gtx scaling and 5870 scaling quite easily. but still not be a huge margin. Its slightly hard to tell as I'm not sure where the cpu limit would be, but theres a difference between all the sli/xf's that you'd expect given scaling of the 470gtx in previous reviews, IE not as good.

For that it seems pretty clear that best case scenario AMD scale noticeably better than SLI, and worst case they are about on par. I'm sure theres games where SLI has a clear lead, though if it has a clear lead in anything but Lost Planet 2, Metro 2033 I don't know.

6870 scaling is a dramatic step forward compared to the 5870, which itself wasn't doing half badly lately.
Click to expand...

Meanwhile the site we can't link to properly from here paints a different picture:

url=http://www.h e x u s.net/content/item.php?item=27167&page=8

Based on my own testing I have seen between 80 and 90% scaling on my 460's in SLI.

EDIT: I got bored, from the Anandtech review posted by DM the following scaling at 1920x1200 4xAA is noted:


Untitled-1.jpg


I have Removed the two anomalies which showed poor scaling on both setups (CIV 5 and Wolfenstein (does anyone actually play this?!)).

Add in the H e x u s results for games not included in the above review (note they are scaled at 2560x1600) you get an overall average result of:

Overall 84% 87% (460 - HD6870)


Not a great deal in it and very game dependant. So in other words it's as I formerly suggested, scaling on the GTX460 and HD68xx is roughly on par.

EDIT: Results including HD5870 CF scaling (as per above)

Untitled2.jpg


Make of that what you will.
 
Last edited:
drunkenmaster said:
http://www.anandtech.com/show/3987/...enewing-competition-in-the-midrange-market/14

Well from that theres three things to see, a 5870 is a full 20% faster than a 6870 in single card, in xfire, the 6870 is now only 3% behind, so in BC2 at least, its got hugely better scaling.

.....

6870 scaling is a dramatic step forward compared to the 5870, which itself wasn't doing half badly lately.
Click to expand...

It's figures like those that DM has taken the time to convey that are the basis of my original question .. relative to a 5870 (where all clocks match and I have actually used 1, 2 and 3 of these GPUs in anger) I reckoned on x1.6 for the 5970 and approx x2 with the 5870 thrown in, depending on game etc .. I will go on price but I really like using 3 screens now so if Nvidia can match price/performance then no reason not to try them again! Had 2x260GTX last time from the 8800GTX before that so my wallet helps me make the main decisions :)

I was thinking of this as I might well get 2xCayman cards to play with as by my reckoning (all guesswork of course, with guestimates taken from various sources!) with Cayman @ 1.3x5870 and with say 20% better scaling we have approx 2.25 x 5870 in reality and that is with 2GB per GPU which will do my 3x1080p monitors just fine I reckon .. and that will be way better than running 5970+5870 which did not really work in tri-fire for Eyefinity so I sold the Vapor-X 5870 a month ago (for £18 less that what I paid last year as it happens!!). This means that with those ratios : 1.3 X 5870 for Cayman & 1.75 for 2 Cayman gpu Xfire scaling. Of course you can adjust those figures as necessary but I feel that in reality the numbers will be a bit higher tbh, maybe even to deliver 2.5 x 5870 performance..

But, if the scaling doesn't scale then we'll see something different!
 
Last edited:
oweneades said:
Meanwhile the site we can't link to properly from here paints a different picture:

url=http://www.h e x u s.net/content/item.php?item=27167&page=8

Based on my own testing I have seen between 80 and 90% scaling on my 460's in SLI.

EDIT: I got bored, from the Anandtech review posted by DM the following scaling at 1920x1200 4xAA is noted:


GTX460 SLI HD6870 CF
Game Scaling Scaling

Crysis Warhead 81.0% 81.6%
Battleforge 91.1% 95.1%
Metro 2033 66.7% 63.8%
HAWX 81.7% 84.2%
CIV 5 12.5% 45.3%
BF:BC2 71.0% 99.5%
Stalker:COP 74.6% 94.5%
Dirt 2 83.3% 82.3%
Mass Effect 2 94.5% 87.5%
Wolfenstein 57.8% 22.0%

Overall Average 71.4% 75.6%

Removing the two anomalies (CIV 5 and Wolfenstein (does anyone actually play this?!)) the average scaling is:

Overall Average 80.5% 86.1%

Not a great deal in it and very game dependant. So in other words it's as I formerly suggested, scaling on the GTX460 and HD68xx is roughly on par.
Click to expand...

The point was, in general the 6870 has great scaling, NOT that its vastly better than Nvidia's, with a certain troll claiming it was utter rubbish.

The point being, AMD have better scaling, they didn't before, thats a HUGE turn around. Its also worth keeping in mind though that the two "best wins" ignoring the two daft ones where crossfire/sli is basically not working, are both Physx and heavily backed by Nvidia games, Metro 2033 in the Anand review, Mafia 2 in the other one. LIkewise, the 460gtx didn't come out, honestly my memorys already gone, 2 weeks ago, 1 week, 3 weeks, I have no concept of time anymore, so I'd suggest a not completely new, but a new balancing architecture has more to gain from drivers than the GF104, which probably has more to gain driver wise over a 480/470 due to its marginal superscalar setup, but its probably got most of its stuff sorted by now.

But if you added in 5870 scaling, its not that far behind, no where near as far behind as "some" people would have you suggest.

Also worth pointing out either ***** got VERY lucky with their game choice, or, well, honestly for both AMD/Nvidia the numbers are a little too good across the board, but apart from the Mafia 2 result its ahead in every other benchmark with them suggesting 90% for the 460gtx and 95% they think with a "fixed" Mafia 2 result.

This is in general why lately I've been having a go at almost all review sites, everyone uses different settings, everyone almost automatically has a list of things they do when they install an Nvidia or AMD card these days, I know exactly what settings I like in the CCC and on my laptop trying to find and sort out multiscreen to hook it up the TV is a damned nightmare more because I'm simply not used to it than Nvidia drivers suck, thats the huge thing people miss out, familiarity with a set of drivers will fix 98% of problems for most people. Thats why people who just switched "think the drivers are awful and can't work them out", etc, etc.

Both sites have used different games, so different scaling is expected, no one likes running the same set of games so different sites show some comparable numbers, and it would be almost boring/pointless if they did. LIkewise one place prefers mipmap to ultra quality and another to performance, they all make marginal difference.

The thing I suggest, as always is to look at the actual results across several reviews, generally ignore all conclusions, and decide for yourself which results are outliers/pointless/clearly going to get fixed by a driver, which are CPU limited and drag a faster cards performance down comparitively. Basically make up your own mind, I've met the owners/reviewers of some sites, heard about many others, theres a lot of complete idiots who do reviews to be honest. Even the better sites [H] make glaring mistakes which are still unfixed.

Right now my general view across multiple Sli/xfire recent reviews is, 6870/50 scale better, but not massively, but it signals a pretty big turn around as its often 15-25% better than previous AMD cards scaling, so they've not just closed the performance gap in multi card, but smashed through and got a little ahead aswell.

The most important thing could be Antilles, crap 40nm process, crap availability and the RRP of the 5970 turned into a travesty in retail.

RRP's, 5870 £300, 5970, £440, reality, £550.

If due to better supply Antilles offers something in the realms of 2 cores for only 50% extra, and scaling that BEATS Nvidia, then Nvidia are in trouble because if Antilles only costs 50% more for a second core, when the first one already costs significantly less, ouch. You could be looking at Antilles for £500 matching £800-900 580GTX SLI.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom