I recently installed 2 9600GT to replace my old 8800GTS 640mb and have been very disappointed with the results. It seems to get better frame rates in Crysis but not much and only scores 12,365 on 3dmark 06. That seems really low for an SLI setup. My old single 8800GTS was 10,325. Maybe I am expecting too much for main stream cards, but I thought I had seen numbers around 17k for 96 in SLI. Does anyone know if this is about normal?  
 

According to this link two 9600GTs should get much better scores across the board in 3dmark06 at 19x12 resolution (although it should be noted the scores in the article are not overall scores, just individual tests). If you look at the game framerates, the difference is similarly large. And keep in mind that the 8800GTS in the article is the newer, faster 512mb g92 version (aka better than your old card). Which means your new setup should kill your old setup. If it doesn't, you have a problem of some sort.

I figure it should be scoring higher than it is. It's hard to find benchmarks that are not altered by overclocked CPUs. I see some as high as 19k but they are overclocking. I know something is off, it has been a royal pain ever since I installed them. First time around everything was freezing up and BSODing, then I reinstalled windows and still have issues with freezing up from time to time with Crysis.

Hmmm. I assume your drivers are up to date? Have you tried pulling out one card and benchmarking them individually? A hardware issue in unlikely but always possible....

Yes, the drivers are up to date in fact I just installed 175.12 beta to see if it would help, no difference. I guess it's possible that one card is not working properly. I'll try them individually and see what happens. The only other thing I think it would be is a motherboard issue. I'll try the cards and if that the same I might see if I can get EVGA to replace the motherboard under warranty.

This is the first computer problem that has had me stumped.

I have a single 9600GT in one of my PC's.  3DMark06=10,618
Some sing wong with 2x9600GT=12,365 IMO

SLi doesn't scale well at default 3dmark settings. as an example, look how little Fudzilla's default 3dmarks went up with sli: (12K with one 8800GT, 13.7K with dual 8800GT) http://www.fudzilla.com/index.php? [...] &task=view
 
You'll get higher 3dmarks OC'in your CPU than adding a second GPU.  Not a very useful test.  Try COD4, Oblivion, or Crysis with fsaa.  
 

Indeed, 3dmark isn't a good test if the cpu is limiting the vidcards.

I don't know, why does the 9800GX2 get such high scores then? and the 3870 X2

It doesn't do that good compared to in actual games.   If you see incredible scores, it's due to massive clock speeds, especially the CPU.  
 
Take a look at Fudzilla, and you'll see paired with a core2 extreme 6800, the GX2 scores 13,714  while the G92 8800GTS scores 12,115.  Compare that to Fear at 19x12 4x/16x and it's 101 fps vs 55 fps.  http://www.fudzilla.com/index.php? [...] &Itemid=40
 
Default 3dmark score mean squat IMO.  Want a higher score, OC your CPU:
http://www.extremetech.com/article [...] 307,00.asp
 

 
That is an excellent point. I agree that 3dmark 06 really responds more to overclocking than additional GPUs. It is difficult to find a benchmark for comparison since almost everyone on there is running their processors overclocked. After really looking at things I realize that number is about right anyway. When you look at a 9800GX2 is 13,260 or so, my 12,500 sounds about right at stock speeds. Even though the performance of the 9600GTs together was not what I initially expected, I think I will keep them since they are silent and they run very cool for passive cooling. I plan to replace them after the next gen cards are released anyway. I plan to put one in my wife's computer and the other in a HTPC since they have HDMI. I'm still tweaking them so I should be able to squeeze a little more out of them. We will see...