The Inquirer-Home

Nvidia Quad SLI works - but not that well

First INQpression Remove one card and performance tumbles
Tue Aug 15 2006, 12:42
QUAD SLI finally works. It only took Nvidia eight months from its paper launch to get any kind of driver out. And even today the performance is really fishy. But as soon as the driver was out we decided to give it a try and we uncovered quite a few interesting facts about this current (91.37) driver.

Quad SLI is actually rather simple. You plug two cards into a machine, install a new driver, enable SLI - or should we say Quad SLI - and it works. You can enable or disable SLI if you don't want it for some reason.

alt='7950gx2'
1024MB of memory times two

An interesting thing that we found out, (Ok, Sanjin found out) is that if you disable Quad SLI and still have two cards on board you will get a significant performance decrease. We are talking about 50 frames loss - almost 50 percent of the overall performance. We did a chart that proves it and we hope that Nvidia is about to fix this issue with a new driver.

alt='s7950gx2'
Single 7950GX2 inserted

We got two Leadtek PX7950GX2 TDH clocked at 500MHz/1200MHz as we thought that we should not mix the cards. We had tried a Gainward/Leadtek mixture and blamed the different vendors for the issue but the same stuff occurs with two identical cards from the same manufacturer.

alt='d7950gx2'
Two 7950GX2 inserted

Benchmarketing
We used :
Abit Fatal1ty AN9 Nforce 590 SLI motherboard
Athlon FX 62 2800 MHz 90 nanometre Windsor core
2x1024 MB DDR2 Corsair CM2X1024-6400C3 memory
Leadtek PX7950GX2 TDH 500 MHz /1200 MHz
Seagate Barracuda 7200.8 400GB SATA NCQ hard drive
Akasa AK913 EVO 33 Athlon 64 cooler
Silverstone Strider ST60F, 600W ATX 2.0 SATA
Driver 84_21_forceware_winxp2k_whq

We tested three games and three 3Dmark routines to assess the performance and the funny little issue when you remove the second card. We heard that there will be one more performance driver but we haven't got that one yet, so this is the best you can get now - at least publicly.

Doom 3 High Quality FSAA 4X + Aniso 8X
1024x768
1280x1024
1600x1200
2048x1536
Leadtek PX7950GX2 TDH Quad SLI
124.2
124.8
122.7
114.5
Leadtek PX7950GX2 TDH - with two card am board
107.7
77.7
57.5
37.1
Leadtek PX7950GX2 TDH - without other card am board
127.2
120.8
103.4
71.5
Abit Fatal1ty AN9 32X - Leadtek PX7950GX2 TDH
126.1
119.5
103.3
71.9
? ? ? ? ?
Far Cry High Quality FSAA 4X + Aniso 8X
1024x768
1280x1024
1600x1200
2048x1536
Leadtek PX7950GX2 TDH Quad SLI
92.16
92.26
92.79
91.35
Leadtek PX7950GX2 TDH - with two card am board
98.46
88.06
64.05
41.50
Leadtek PX7950GX2 TDH - without other card am board
95.35
95.26
94.02
76.40
Abit Fatal1ty AN9 32X - Leadtek PX7950GX2 TDH
102.79
101.35
100.27
76.80
? ? ? ? ?
F E A R
1024x768
1280x1024
1600x1200
2048x1536
Leadtek PX7950GX2 TDH Quad SLI
131
122
105
74
Leadtek PX7950GX2 TDH - with two card am board
73
51
36
24
Leadtek PX7950GX2 TDH - without other card am board
121
98
69
46
Abit Fatal1ty AN9 32X - Leadtek PX7950GX2 TDH
125
97
70
45
? ? ? ? ?
Composite Figures 3Dmark 03
3DMark 03
Game2
Game3
Nature
Leadtek PX7950GX2 TDH Quad SLI
36499
351.3
267.5
204.3
Leadtek PX7950GX2 TDH - with two card am board
18555
145.0
119.4
106.7
Leadtek PX7950GX2 TDH - without other card am board
30114
262.2
213.3
181.7
Abit Fatal1ty AN9 32X - Leadtek PX7950GX2 TDH
30023
261.6
213.0
181.9
? ? ? ? ?
Composite Figures 3Dmark 03
Single Texturinng
Multi Textur.
Vertex Shader
Pixel Shader 2.0
Leadtek PX7950GX2 TDH Quad SLI
15500.1
40372.0
90.4
588.6
Leadtek PX7950GX2 TDH - with two card am board
4179.2
11522.5
49.2
229.3
Leadtek PX7950GX2 TDH - without other card am board
8247.7
22438.4
88.7
410.4
Abit Fatal1ty AN9 32X - Leadtek PX7950GX2 TDH
8246.1
22445.7
88.7
411.1
? ? ? ? ?
Composite Figures 3Dmark 05
3DMark 05
Game1
Game2
Game3
Leadtek PX7950GX2 TDH Quad SLI
13576
52.1
42.3
72.8
Leadtek PX7950GX2 TDH - with two card am board
8565
37.9
25.8
41.4
Leadtek PX7950GX2 TDH - without other card am board
13797
53.3
42.8
73.7
Abit Fatal1ty AN9 32X - Leadtek PX7950GX2 TDH
13664
52.3
42.1
73.9
? ? ? ? ?
Composite Figures 3Dmark 05
Single Texturinng
Multi Textur.
Pixel Shader
VS /VS complex
Leadtek PX7950GX2 TDH Quad SLI
17999.4
41268.1
1024.6
126.6//95.9
Leadtek PX7950GX2 TDH - with two card am board
4805.9
11761.7
333.0
65.8/49.2
Leadtek PX7950GX2 TDH 2 - without other card am board
9456.7
22763.2
638.0
129.9/95.2
Abit Fatal1ty AN9 32X - Leadtek PX7950GX2 TDH
9468.9
22765.0
638.9
130.0/95.0
? ? ? ? ?
Composite Figures 3Dmark 06
3DMark 06
? ?
Leadtek PX7950GX2 TDH Quad SLI
8539
? ?
Leadtek PX7950GX2 TDH - with two card am board
4857
? ?
Leadtek PX7950GX2 TDH - without other card am board
8083
? ?
Abit Fatal1ty AN9 32X - Leadtek PX7950GX2 TDH
8080
? ? ?
? ? ? ? ?
SM2.0 Test
Score
GT1
GT2


Leadtek PX7950GX2 TDH Quad SLI
3974
32.005
34.234
30.8
Leadtek PX7950GX2 TDH - with two card am board
1937
15.607
20.774
64.8
Leadtek PX7950GX2 TDH - without other card am board
3626
28.955
31.477
Abit Fatal1ty AN9 32X - Leadtek PX7950GX2 TDH
3628
28.971
31.503
?
? ? ? ? ?
HDR/SM3.0 Test
Score
HDR1
HDR2


Leadtek PX7950GX2 TDH Quad SLI
3657
33.751
39.382
30.8
Leadtek PX7950GX2 TDH - with two card am board
2116
17.099
20.774
64.8
Leadtek PX7950GX2 TDH - without other card am board
3484
31.044
38.632
Abit Fatal1ty AN9 32X - Leadtek PX7950GX2 TDH
3475
30.841
38.661
?
? ? ? ? ?
CPU Test
Score
CPU1
CPU2


Leadtek PX7950GX2 TDH Quad SLI
2143
0.681
1.079
30.8
Leadtek PX7950GX2 TDH - with two card am board
2116
0.682
1.050
64.8
Leadtek PX7950GX2 TDH - without other card am board
2138
0.670
1.091
?
Abit Fatal1ty AN9 32X - Leadtek PX7950GX2 TDH
2140
0.681
1.075
?

The first score refers to the Quad SLI - two 7950 GX2s rendering together. The second scores are from two 7950 GX2 cards with SLI disabled in software. The third case is with the second 7950 GX2 card physically removed and the last one is a 7950 GX2 card tested as a single card.

In Doom 3, with high quality 4X FSAA and 8X Aniso you need to get to higher resolutions to show the power of Quad SLI. There is a difference of five frames at 12x10, almost 20 frames difference at 16x12 and respectable 42.5 frames difference in 20x15. Quad SLI is massive 62.8 percent faster than a single 7950 GX2 card, and that is a lot.

You can see that the third score with the second 7950 GX 2 removed after the driver was installed will score virtually the same as the single card but software disabled SLI setting with two cards will result in at least 20 frames drop and, in the worst case scenario, it will be almost half the speed of a single card. Something is certainly wrong here but we are sure it will be fixed sooner or later.

Far Cry at high quality FSAA 4X + Aniso 8X will show ten to fifteen frames advantage, not that much but will get you to 91 frames in 20x50 while a single card cannot perform at more than 76 FPS. The Quad SLI system with one card removed will perform almost identically as the single Leadtek 7950 GX2 card but, yet again, if you disable SLI and keep both cards in the PCIe slots you will see some dramatic performance decrease and the card will run up to 30 frames slower than a single card.

FEAR loves Quad SLI. It starts with six frames advantage at 10x7, continues with a massive 25 FPS advantage at 12x10, manages to get the 35 frames advantage and finally settles at 29 frames advantage at 20x15. It runs up to thirty percent faster than a single card. You can again see the issue with two 7950 GX2 cards and SLI disabled as the performance drops tremendously.

Quad SLI is good for 3Dmark 03 show off. A single 7950 GX2 card scores just above 30000 marks while the Quad SLI 7950 GX2 scores 36500 some six and a half thousand faster. The game 2 will run a massive 90 frames faster that with a single card. Jet again the SLI disabled with two cards on board will drop the score to 18000 only. The card can theoretically do twice as much single texturing and almost twice as much multi texturing, will do exactly the same amount of vertex Shader information but will run pixel operations 50 percent faster.

In 3Dmark05 the Quad and single 7950 GX2 scores are almost identical and a single card is actually a few points faster. In theory, single and multi texturing is almost twice as fast on Quad SLI but this isn't shown in our results on games. Quad SLI can do significantly more pixel operations but not the vertex ones, as the single card is a few frames faster.

3Dmark 06 scores are less than impressive. Quad SLI runs some 500 points faster. We are talking about five and a half percent differences. Shader model 2.0 score is 300 points faster almost nine percent where you can get three to four frames difference in in-game game tests.

HDR and Shader model 3.0 tests is less than 200 points faster on Quad SLI. It runs one to three frames faster in games. We have to admit it is less than impressive. CPU scores are almost identical.

Both 3Dmark05 and 06 will drop the performance significantly, almost cutting in it half if you disable SLI and leave the two 7950 GX2 cards inside.

In Short

Except in a single case in Doom 3 we can easily state that Quad SLI needs some more work.

We expected to see more performance and we are still waiting for a new performance driver. Nvidia had to release the driver prior to its conference call with financial analysts and it did. But the software is clearly rushed. It is stable but does not perform as expected. It runs up to thirty per cent faster if you're lucky but you have to pay hundred per cent more to get thirty percent back.

We will compare this with the current X1900 XTX Crossfire and the upcoming X1950 XTX Crossfire when the time comes but this is what you are getting with two 7950 GX2 at the moment. As for disabling SLI with two cards in their slots and the related severe performance drop, we are sure it is a bug that Nvidia can and will fix soon. We will be happy to retest with a new performance driver once Nvidia releases one, at least we hope it will. ?

Reviewed and tested by Sanjin Rados and Fuad Abazovic

 

Share this:

blog comments powered by Disqus
Advertisement
Subscribe to INQ newsletters

Sign up for INQbot – a weekly roundup of the best from the INQ

Advertisement
INQ Poll

Dead electronic devices to be banned on US-bound flights

Will the new rules banning uncharged devices be effective?