CrossfireX (Tri & Quad) @ Anandtech

BFG10K

Lifer
Aug 14, 2000
22,709
3,002
126
A good start for ATi as it seems to scale from 2 -> 3 GPUs better than Tri-SLI does.

Click.
 

nRollo

Banned
Jan 11, 2002
10,460
0
0
Originally posted by: BFG10K
A good start for ATi as it seems to scale from 2 -> 3 GPUs better than Tri-SLI does.

Click.

Well, sort of.

The problem with the "scales better" argument people like to make about ATi cards these days is they start out so much slower that "scaling better" still leaves them slower or, at best, equalling their competition*.

Let's look at the benches in common of Tri-fire.......Quad-fire.......3 way SLi

UT3....................................................112............114...................146
Bioshock..............................................86.............92......................103
COD4..................................................72..............93........................100
Crysis..................................................26..............27.........................43

1. Even with 4 GPUs, ATi's highest end product does not compete with the 3 highest NVIDIA GPUs.

2. Only one of those benches (COD4) has AA on, and we all know how the current ATi products shader resolve AA sinks performance.

So again we're left with an alternative, but probably not the best one. (although it is considerably cheaper than 3 way)


*by "competition" I mean "flagship product vs flagship product" here, as the 9800GTs and 9800GX2 are not out yet to compare on a price perspective
 

happy medium

Lifer
Jun 8, 2003
14,387
480
126
What cost more? 4 -3870's or 3- 8800gts's (g92). Seems to me 3- 8800gts's would still beat 4- 3870's.
 

Toonces

Golden Member
Feb 5, 2000
1,690
0
76
Originally posted by: nRollo
Originally posted by: BFG10K
A good start for ATi as it seems to scale from 2 -> 3 GPUs better than Tri-SLI does.

Click.

Well, sort of.

The problem with the "scales better" argument people like to make about ATi cards these days is they start out so much slower that "scaling better" still leaves them slower or, at best, equalling their competition.

Let's look at the benches in common of Tri-fire.......Quad-fire.......3 way SLi

UT3....................................................112............114...................146
Bioshock..............................................86.............92......................103
COD4..................................................72..............93........................100
Crysis..................................................26..............27.........................43

1. Even with 4 GPUs, ATi's highest end product does not compete with the 3 highest NVIDIA GPUs.

2. Only one of those benches (COD4) has AA on, and we all know how the current ATi products shader resolve AA sinks performance.

So again we're left with an alternative, but probably not the best one. (although it is considerably cheaper than 3 way)

Directly comparing FPS? Anand clearly wanted people to not take his CrossfireX and Tri-SLI results as comparable, as per his disclaimer on the first page of the review.

It was a closed AMD prepared platform running a 790FX motherboard and AMD Phenom Quad @ 2.6 v. an Intel Core 2 Extreme QX9650 @ 3.33GHz on an nForce 780i. Not to mention CrossfireX's beta driver status against nVidia's publicly available ones.

Sure, they're most likely running at GPU-bound resolutions, but again, direct comparisons are invalid. The whole point of the article was to show the difference in scaling between the two respective platforms - not the quantitative results of each.


 

nRollo

Banned
Jan 11, 2002
10,460
0
0
Originally posted by: toonces

Sure, they're most likely running at GPU-bound resolutions, but again, direct comparisons are invalid. The whole point of the article was to show the difference in scaling between the two respective platforms - not the quantitative results of each.

I somewhat agree with this as I didn't notice they actually used a Phenom for this test, but they should be GPU limited at these setting as you note.

I think future comparisons on comparable rigs will bear out my 3 way vs Quad conclusion, but the 3 way is more expensive, so that's as it should be.

From a gamers's perspective, to me "scaling" will never be the criteria I make buying choices on:

1. End minimum and average framerate are all that matter for gaming.

2. I'd much rather have a higher performing single GPU that scales worse (but ends up the same or better) because not all games scale. This is especially relevant given ATi's inability to disable CF with 3870X2s, force SFR, or create/edit profiles.

3. I'd think the Phenom itself would be the biggest problem with a rig like the tested one- why would anyone knowingly buy a comparably priced, slower processor that doesn't OC well?
 

Narse

Moderator<br>Computer Help
Moderator
Mar 14, 2000
3,826
1
81
Originally posted by: happy medium
What cost more? 4 -3870's or 3- 8800gts's (g92). Seems to me 3- 8800gts's would still beat 4- 3870's.

Only cards you can tri sli are the 8800GTX and Ultra. Sad you cant run 3 GTSs or GTs
 

dadach

Senior member
Nov 27, 2005
204
0
76
nTROLLO you are still a sad individual, and it is incomprehensible that they let you back to post the garbage arguments, motivated by bribe...although it does not surprise me considering keysplayer overstepping his boundaries as and impartial observer/moderator in ati/nvidia war

anyways are you saying that the price of more than a double for 3 GTX as oppose to 2+1 or 2+2 ati gpus, justifies the difference in performance?

and please all do not include nvidia game crysis in any of the ati benching...it is clearly just a tool to discredit atis performance (however fast or slow it may be)
 

happy medium

Lifer
Jun 8, 2003
14,387
480
126
Originally posted by: Narse
Originally posted by: happy medium
What cost more? 4 -3870's or 3- 8800gts's (g92). Seems to me 3- 8800gts's would still beat 4- 3870's.

Only cards you can tri sli are the 8800GTX and Ultra. Sad you cant run 3 GTSs or GTs

Thanks for the info. Yes your right it is sad that you can't tri 3 gts's.

Rollo do you know if the 9 series is gonna allow tri sli with mid range cards? 9600gt/9800gt and such?
Any reason they didn't let you tri sli 8800gts's? It would be real cost effective.
 

Blacklash

Member
Feb 22, 2007
181
0
0
I'd love to see someone do a review that shows which setups, Crossfire or SLi, actually function properly in the most games released over the last three years or so. Or that is do they provided some FPS boost on the order of 20% or greater. I imagine it would be a disappointment to invest large amounts of money in Crossfire|SLi or Multi-gpu and find yourself reduced to single card or worse performance in titles you enjoy, or some new title that just came out. That would be an unwelcome surprise.
 

apoppin

Lifer
Mar 9, 2000
34,890
1
0
alienbabeltech.com
Originally posted by: Blacklash
I'd love to see someone do a review that shows which setups, Crossfire or SLi, actually function properly in the most games released over the last three years or so. Or that is do they provided some FPS boost on the order of 20% or greater. I imagine it would be a disappointment to invest large amounts of money in Crossfire|SLi or Multi-gpu and find yourself reduced to single card or worse performance in titles you enjoy, or some new title that just came out. That would be an unwelcome surprise.

Well .. *we* could do that ;)

but i am not going to start a new *poll* or thread ... i am tired and not quite finished with my last benchmarking tests

and i no longer give a crap about any "just released" brand new game ... 9/10 are buggy as hell any way and usually take a couple of months to patch properly.

and i dunno ... i do not feel bad if i am reduced to a "single 2900xt performance" ... does anyone really feel bad playing with their 8800GTS-640? that 'extra' $150 i spent to Crossfire was never "wasted"

rose.gif
 

apoppin

Lifer
Mar 9, 2000
34,890
1
0
alienbabeltech.com
Originally posted by: Blacklash
I don't consider SupCom, TRL, NWN2 and Gothic 3 new games :p

The below cracks me up every time I see it-

http://www.techpowerup.com/rev...GT_Amp_Edition/15.html

Find the X2.

Well *two* of those games are still buggy and 1 will never be fixed.:p
[hint ... aurora engine ... genome "memory mismanagement engine" where nvidia GPUs still regularly crash]


NWN2 and G3 play decently on a 2900xt .. or almost as badly as on an 8800GTS ... depending on your PoV. :p
 

Peelback79

Senior member
Oct 26, 2007
452
0
0
Originally posted by: happy medium
Originally posted by: Narse
Originally posted by: happy medium
What cost more? 4 -3870's or 3- 8800gts's (g92). Seems to me 3- 8800gts's would still beat 4- 3870's.

Only cards you can tri sli are the 8800GTX and Ultra. Sad you cant run 3 GTSs or GTs

Thanks for the info. Yes your right it is sad that you can't tri 3 gts's.

Rollo do you know if the 9 series is gonna allow tri sli with mid range cards? 9600gt/9800gt and such?
Any reason they didn't let you tri sli 8800gts's? It would be real cost effective.

They probably did it out of neccesity. If you could dual or tri the gt(s)'s there'd be no reason really to buy the Ultra's or GTX's.
 

nitromullet

Diamond Member
Jan 7, 2004
9,031
36
91
Originally posted by: nRollo

I think future comparisons on comparable rigs will bear out my 3 way vs Quad conclusion, but the 3 way is more expensive, so that's as it should be.

From a gamers's perspective, to me "scaling" will never be the criteria I make buying choices on:

1. End minimum and average framerate are all that matter for gaming.

2. I'd much rather have a higher performing single GPU that scales worse (but ends up the same or better) because not all games scale. This is especially relevant given ATi's inability to disable CF with 3870X2s, force SFR, or create/edit profiles.

3. I'd think the Phenom itself would be the biggest problem with a rig like the tested one- why would anyone knowingly buy a comparably priced, slower processor that doesn't OC well?

I'd actually really like to see a comparison of two systems that compare more closely price wise. 8800GTX SLI vs. 3870X2 Crossfire, with an 8800GT SLI setup thrown in the mix to would be a good comparison. Given that 3x 8800GTX is at minimum ~$1200 (unless you're Rollo ;) ) and the cost of a two 3870X2s is ~$900, a comparison between 3-way SLI and 3870X2 CrossfireX isn't really fair.
 

bryanW1995

Lifer
May 22, 2007
11,144
32
91
a single ultra cost, what, $650+? 2 3870 x2's are about $900. If you go with current prices, it's probably $1100 for 3 8800gtx's that run very hot and have old tech vs $760 for 4 x 3870 that are a LOT cooler running than 3 gtx's. I think that it's very tough to justify tri-sli over quad fire, um, I mean, it will be when amd finally releases drivers for it. Of course, the real issue is going to be how long after that 9800gtx is released: if it comes out a month later there'll be a lot of quad-fired systems with gpu-envy, if it comes out in nov-dec there will be a lot of "enzyte" smiles on amd gpu-owners faces.
 

Blacklash

Member
Feb 22, 2007
181
0
0
Crossfire causes a performance hit in some titles that makes the GPUs perform less than a single one would. You could increase the X2's numbers in the SupCom by turning off CAT AI which would disable on board Crossfire. Then you'd have a little less than the performance of a single HD 3870 for 450usd. You'd also lose all driver fixes for games while CAT AI remained disabled.

Other titles I am aware of that Crossfire does this in would be; Gothic 3, NWN2, Need for Speed: Pro Street, WiC on the DX10 path with AA, Lost Planet: Extreme Condition, Tomb Raider: Legend, Hellgate: London, and Jericho with edge smoothing on.

Those are only the ones I am personally aware of, there are likely others.
 

apoppin

Lifer
Mar 9, 2000
34,890
1
0
alienbabeltech.com
Originally posted by: Blacklash
Crossfire causes a performance hit in some titles that makes the GPUs perform less than a single one would. You could increase the X2's numbers in the SupCom by turning off CAT AI which would disable on board Crossfire. Then you'd have a little less than the performance of a single HD 3870 for 450usd. You'd also lose all driver fixes for games while CAT AI remained disabled.

Other titles I am aware of that Crossfire does this in would be; Gothic 3, NWN2, Need for Speed: Pro Street, WiC on the DX10 path with AA, Lost Planet: Extreme Condition, Tomb Raider: Legend, Hellgate: London, and Jericho with edge smoothing on.

Those are only the ones I am personally aware of, there are likely others.
personally aware of
??
wtF? You sold your X2 after 2 days :p
:confused:

G3 scales ... NWN2 also ... Hellgate certainly does ... Jericho too ... Lost Planet scales excellently .. no issues ... i haven't had to disable Crossfire ... and i play with max settings
-but then this is for my system

do you HAVE a Crossfire rig ... or is this *all* from 24 hours with 3870x2? or are you quoting old and 2nd hand news?
 

dadach

Senior member
Nov 27, 2005
204
0
76
i can run/test 2x x2 or 1x x2 + 1x normal 3870 on e6600@3.3ghz, ix38 quadGT abit, vista 64-bit...as soon as anand sends me that beta cX driver :D
 

apoppin

Lifer
Mar 9, 2000
34,890
1
0
alienbabeltech.com
Originally posted by: apoppin
Originally posted by: Blacklash
Crossfire causes a performance hit in some titles that makes the GPUs perform less than a single one would. You could increase the X2's numbers in the SupCom by turning off CAT AI which would disable on board Crossfire. Then you'd have a little less than the performance of a single HD 3870 for 450usd. You'd also lose all driver fixes for games while CAT AI remained disabled.

Other titles I am aware of that Crossfire does this in would be; Gothic 3, NWN2, Need for Speed: Pro Street, WiC on the DX10 path with AA, Lost Planet: Extreme Condition, Tomb Raider: Legend, Hellgate: London, and Jericho with edge smoothing on.

Those are only the ones I am personally aware of, there are likely others.
personally aware of
??
wtF? You sold your X2 after 2 days :p
:confused:

G3 scales ... NWN2 also ... Hellgate certainly does ... Jericho too ... Lost Planet scales excellently .. no issues ... i haven't had to disable Crossfire ... and i play with max settings
-but then this is for my system

do you HAVE a Crossfire rig ... or is this *all* from 24 hours with 3870x2? or are you quoting old and 2nd hand news?

Just to be *sure* ... i loaded up Jericho ... it is a real Hogg at maxed out settings .. i get low 20s with Crossfire *disabled* ... and low 30s with it enabled [averages]

Crossfire gives about a 50% + improvement in Jericho ... Cat 8.2
 

BFG10K

Lifer
Aug 14, 2000
22,709
3,002
126
The problem with the "scales better" argument people like to make about ATi cards these days is they start out so much slower that "scaling better" still leaves them slower or, at best, equalling their competition*.
The point is ATi's immature CrossfireX drivers are off to a good start.

Crysis? lack of scaling is simply fascinating and I suspect if someone were to use SFR it would actually run better.

I think ATi should've forced super-tiling into all games on the 3870 X2 instead of AFR. This mode actually operates at the hardware level with no driver intervention required so performance gains would have been more consistent and also likely more compatible too.

While it probably would?ve been a bit slower than best-case AFR scaling, it?s still much better to have a more robust product IMO.
 

nRollo

Banned
Jan 11, 2002
10,460
0
0
Originally posted by: BFG10K
The problem with the "scales better" argument people like to make about ATi cards these days is they start out so much slower that "scaling better" still leaves them slower or, at best, equalling their competition*.
The point is ATi's immature CrossfireX drivers are off to a good start.

Crysis? lack of scaling is simply fascinating and I suspect if someone were to use SFR it would actually run better.

I think ATi should've forced super-tiling into all games on the 3870 X2 instead of AFR. This mode actually operates at the hardware level with no driver intervention required so performance gains would have been more consistent and also likely more compatible too.

While it probably would?ve been a bit slower than best-case AFR scaling, it?s still much better to have a more robust product IMO.

I agree that these results look good so far.

I'm surprised they don't let the user pick which method to use, having AFR, SFR, and Tiling at your disposal would be a selling point.