Nah you should run barsWF x64 CUDA client to bench your system better than that 3D mark nonsense really, it's far more to the point and it will show what your setup is caapable in real time CUDA based apps.
3D Mark is unprecise and packed with cheats and bugs you name it, I always avoid it even if my life would depend on it, just due to it's bad scoring system, since every run I do with it the score can have more than 2000 points difference, with BarsWF it's spot on, you can grab it from here:
http://3.14.by/en/md5 Download the CUDA client test it by unpacking it running the x64 exe file and to stop the benchmark press space.
it's a MD5 Hash benchmark, this will utilize your system to it's 100% performance potential.
here my results
Blue Leader SMP 2009 with 2x Six-Core OpteronDP 2435 D0 & 2x EVGA GTX 295 Red Editions, 4x G200-400-B3;
12 CPU Cores @ 2600Mhz + 960 G200 CUDA Cores @ 1242Mhz:
Blue Leader SMP 2009 with 2x Quad-Core OpteronDP 2380 C2 & 2x EVGA GTX 580 SCE, 2x GF110-375-A1;
8 CPU Cores @ 2500Mhz + 1024 Fermi CUDA Cores @ 1594 Mhz:
Firefox SMP 2006 with 2x Dual-Core OpteronDP 285 HE E6 & 1x EVGA GTX 480, 1x GF100-375-A3;
4 CPU Cores @ 2600Mhz + 480 Fermi Cuda Cores @ 1401 Mhz:
All in all this proved me that the G200 CUDA Cores are byfar the most powerful per Mhz and per Core.