SLI benchmarking issue, looking for help

[Knytestorme]

Hi all, first time posting here and hoping I can get some help on a perculiar problem.

I brought a new system last week comprised of:

1. Gigabyte GA-K8N Pro-SLI motherboard
2. 2GB GEIL PC3200 Ram
3. 2 x Pixelview 7800GTX Cards
4. Athlon AMD64 3800+ X2
5. 2 X 250GB WD SATA II drives (configured as 2 seperate jbod volumes)
6. Antec Super Lanboy + 550W Truepower 2.0 PSU

After doing a fresh install of Xp, the AMD cpu driver, coolbits reg hack, nForce 4 6.66 drivers and Forceware 78.01 drivers I have been running benchmarks with Aquamark 3 and 3D Mark 03/05 with strange results.

Aquamark gives me results of around 28k in SLI Mode (GPU loading enables and both bars are rising and falling) while 3d Mark 03 gives me a result of 16k and an fps for game 1 of around 106fps with all options in forceware and 3D Mark at default.

Comparing this to the results shown on benchmarks here as well as at Xbitlabs shows that I shoudl be looking at around 300fps in Game 1 and 24k+ in 3D Mark, which is what concerns me.

Of further concern is that I just did a test of removing the card from PCI_1, setting the mb back to normal mode with the flipper chip and putting the card from PCI_2 into PCI_1 and running the benchmarks in single card mode. Here the results were an Aquamark of around 28k (with the same fps's as in SLI mode, vsync is off), a 3D Mark fo around 14k and game 1 fps of 103fps.

As can be seen, there is somethign weird happening here, and any advice on what could be causing it or things to test would be most appreciated....as would any benchmarks from someone with a similar setup just to check if there actually might be an issue.

One other thing that I'm not sure about, but in system details form 3D Mark it only shows 1/1 GPU, not sure if it should show 2 GPU's in there in SLI mode.

 

[biostud]

unless you've overclocked your CPU their CPU run 20% faster and have 1Mb cache.
I'm not sure but isn't there an option in the Forceware drivers to enable/disable SLI?

If you have two identical Nvidia graphics cards (identical brand, model and clock speed) and installed in an SLI-capable motherboard, this section will be available to you to use. To enable SLI, tick the 'Enable SLI multi-GPU' checkbox. If you want to see how the graphics cards are sharing the load (e.g. if you want to test the various SLI modes on the previous page), tick the 'Show GPU load balancing' option which will display, though this is not normally necessary and should be left unticked.

 

[Fern]

Hi Knytestorme and Welcome to the forums!

I'll send a PM to a member named Rollo who seems to have done lots of benchies with SLI systems.

Might wanna use the search function, use his name and do it by author for more info in the meantime.

Fern

 

[Knytestorme]

Thanks Fem, will do the search now

 

[nRollo]

Your benchmarks seem to be more in line with my single card benches for 3d03:
http://endeavorquest.net:8880/SLI.htm

What are the gaming benchmarks like? (i.e. similar lack of performance?)

I had that motherboard as my first SLI motherboard and had nothing but bad luck with it, and have a friend who had the same miserable luck with it, so I'm not a huge fan of that board.

(although I assume some must work)

 

[Knytestorme]

Rollo,

The gaming benchmarks in 3d Mark are also fairly low but haven't checked in HL2 or Doom 3 yet (due to my ignoracne of how to as I've never bothered before)

I did some further testing because I had the same thought, that it was only getting single card performance, so I ran aquamark 3 with one card in the system and then two, and it gave the same score in either config but 3d Mark 03 only gives around 12k in single card mode but around 16k in dual mode. The other thing I noticed is that while in either mode the gaming benchmarks come out about the same, the fill rate and elephant tests all show a double up in framerate when going from one card to two. For example, the elephant test is around 300fps constant in single card mode but 600 in dual, fill-rate 2 test is around 190fps in single mode and 380fps in dual mode.

I'm just stumped, think the Giga will be getting replaced with an Asus on saturday though and see if that helps.

Oh yeah, one other thing I just thought of. When I first installed the cards they were both running at around 53c temp and stable, after I swapped them around to make sure that they were both actually working the one in pci_2 now shows a temp of only around 43c while the main card is still around 53c even after gaming useage.

 

[Knytestorme]

Welp, went and swapped the gigabyte for an A8N-SLI Premium today, got the machine back up and running and benched with same drivers and configuration as the Gigabyte, only thing that has changed is the motherboard and the results are freaking astounding imho.

All benchmarks are at their default settings except X2 (auto quality, bumpmapping, shadows, antialiasing) and CS:S (10x7, all advanced options maxed out)

Gigabyte GA-K8N-SLI Pro Motherboard
3D Mark 03 - 15,552
3D Mark 05 - 5,495
Aquamark 3 - 32,085
X2 Rolling Demo - 37fps
CS: Source - 33.7fps

Asus A8N-SLI Premium
3d Mark 03 - 25,511
3d Mark 05 - 9269
Aquamark - 72,825
X2 Rolling Demo - 127fps
CS:S - 113.7fps

I honestly don't know what to say about those differences, I even tried testing each of the card in each of the PCI-e slots to make sure they were both working and in single gpu mode they both gave the same benchmarkings so there was/is nothing wrong with the cards and in SLI mode on the Gigabyte the drivers were telling me the cards were in SLI.

The most interesting thing is that in all of the benchmarks now the GPU scaling graph shows the cards working in places they weren't under the gigabyte, and by not working under the gigabyte I mean the graphs were stuck at the midline and not showing any growing/shrinking during the benchmarks as though it was just working in cpu mode the entire time.

Oh well, I'm happy now, marks are close to Rollo's but I'm using slightly slower RAM at 3.0-4-4-8-2T Geil but thought I'd post a follow-up with my findings in case anyone else has the same issues.