"1+1=?": ATI CrossFireX and Nvidia SLI on Intel Core i7 Platform

[Final8y]

[02/03/2009]
Today we are going to check out the scalability of ATI Radeon HD 4870 and NVIDIA GeForce GTX 260(216SP) graphics cards in CrossFireX and SLI configurations built on an Intel X58 Express based platform.
Intel’s recently-released platform with Core i7 processors is universal when it comes to today’s multi-GPU technologies, ATI CrossFireX and Nvidia SLI. Mainboards based on the Intel X58 Express chipset and certified by ATI and Nvidia allow the user to build a graphics subsystem out of two cards from either GPU developer. As you surely know, it had not been possible for one mainboard to support both CrossFireX and SLI before.

You may be wondering how efficient these multi-GPU technologies are on the new platform. In what case will you have the highest performance benefits from installing a second identical graphics card into your system and what problems there can arise? This article will answer these and some other important questions.

# Graphics card driver for ATI GPU: Catalyst 8.12 (8.561.3 Hotfix, 12/16/08)
# Graphics card driver for Nvidia GPU: GeForce 181.22 (WHQL, 01/22/09)
# Physics acceleration driver: PhysX 9.09.0010 (WHQL, 01/13/09)

xbitlabs

 

[Axil]

interesting read - if anything it's srt of put me off using either -

 

[Happy]

Nice read, thanks.

 

[gurusan]

Interesting.

Dual GPU setups aint for me :/

 

[Rroff]

I guess they were just going for out the box compatibility and performance... but if they'd taken 2 minutes to play around with custom SLI profiles they would have seen around +60% performance in the Unigine Tropics Demo on nVidia with a custom profile and possibly X3, left 4 dead has some SLI scaling issues but other benchmarkers have managed to get some small gains.

 

[jigger]

Great read, proper reviews seem thin on the ground nowadays.

 

[J1nxy]

In left 4 dead I go from 65-70 fps with a single 280GTX @1920 x1200 with max af and aa with sli I get 145+ fps with the same settings...pinch of salt required me thinks.

 

[rjk]

In left 4 dead I go from 65-70 fps with a single 280GTX @1920 x1200 with max af and aa with sli I get 145+ fps with the same settings...pinch of salt required me thinks.

same here

an interesting read - but proved wrong by my own experiences
a lot of games there dont scale well with sli either

all my games add loads of performance with my cards in sli - everything on max

heres my fps averages @ 1920*1200 everyhting on max settings with rough percentage increases

GRID - Single = 65+
GRID - SLI = 110+
SLI PERFORMANCE GAIN = 40%+


COD4 - Single = 80+
COD4 - SLI = 160+
SLI PERFORMANCE GAIN = 50%+

L4D - Single = 65+
L4D - SLI = 120+
SLI PERFORMANCE GAIN = 50%

Crysis DX10 - Single = 35+
Crysis DX10 - SLI = 55+
SLI PERFORMANCE GAIN = 45%


3dmark vantage - Single = P14966
3dmark vantage - SLI = P26359
SLI PERFORMANCE GAIN = 45%

but thats just what i found
using the 180.20 drivers here

and for the record - an 8800gt could probably play L4D at max so sli performance is pointless - source engine is cpu reliant anyway

 

[Final8y]

same here

an interesting read - but proved wrong by my own experiences
a lot of games there dont scale well with sli either

all my games add loads of performance with my cards in sli - everything on max

heres my fps averages @ 1920*1200 everyhting on max settings with rough percentage increases

GRID - Single = 65+
GRID - SLI = 110+
SLI PERFORMANCE GAIN = 40%+


COD4 - Single = 80+
COD4 - SLI = 160+
SLI PERFORMANCE GAIN = 50%+

L4D - Single = 65+
L4D - SLI = 120+
SLI PERFORMANCE GAIN = 50%

Crysis DX10 - Single = 35+
Crysis DX10 - SLI = 55+
SLI PERFORMANCE GAIN = 45%


3dmark vantage - Single = P14966
3dmark vantage - SLI = P26359
SLI PERFORMANCE GAIN = 45%

but thats just what i found
using the 180.20 drivers here

and for the record - an 8800gt could probably play L4D at max so sli performance is pointless - source engine is cpu reliant anyway

Your percentages are wrong.

Example COD4 - Single = 80+
COD4 - SLI = 160+
SLI PERFORMANCE GAIN = 100%+

-50% of 80=40 40+80=120 not 160.

Example COD4 - Single = 80+
COD4 - SLI = 120+
SLI PERFORMANCE GAIN = 50%+

 

[muon]

Good review to compare the HD4870 and GTX260 (factory overclocked). The HD4870 really does do very well. I had always thought it was worse than thew 216 core gtx 260. Its even matching it in crysis warhead, which has normally been nvidia territory.

edit: I see its the 1 gig hd4870 though which helps. But they lowered both cards to stock clocks rather than factory clocks which is good for the comparison.

edit: 8.12 drivers from 16th december, while nvidia's are from 22nd january. A shame 9.1 was released so late.

 

[Rroff]

The HD4870 really does do very well. I had always thought it was worse than thew 216 core gtx 260. Its even matching it in crysis warhead, which has normally been nvidia territory.

When the first proper "big bang" nvidia drivers came out - the 260 cards jumped from slightly behind the 4870 to a decent amount infront... ATI then release a couple of driver sets that pushed them back slightly infront of the 260, nvidia then released another driver set thats made them pretty much equal overall. Tho 182.xx and 185.xx betas are showing gains again over ATI to around 7%.

Unfortunatly this review is pretty much invalid by now tho - I'm seeing people who are managing to get almost 100% scaling in left 4 dead with SLI now, and comparing stock clocks is getting a bit silly as pretty much no one is selling or running a 260 anything close to stock clocks, in many cases as much as 100-200Mhz higher.