Under Windows 98 SE, you can see it scales rather nicely.  With the GF2 Ultra at stock speed being the bottom benchmark for all the other tests. At 4201 that's no shabby score.

I am very interested in any advantages or disadvantages the different driver versions made in the performance.  From Win98 SE you can see there is little if any difference in speed between them on the GF3 or the Ultra.

This was a test that many people will see as invalid as it doesn't use the GF3's core to its peak potential.  I see it as very valid as most people do NOT use games that can use the GF3 core yet anyway. 

  In here you can see some breaking from the nice scaling we had on the 2001 test.  In this one its pretty clear that Dx7 API apps can turn the tables on the big bad GF3. At this point raw MHZ seem to mean more then the technology in the core.  But still Clocked almost 70Mhz slower at the core, and 20Mhz slower at the ram it still comes within 60Marks of the GF2 Ultra.

The GLMark was a great test to run as many games and apps use Open GL more then D3D currently.   This also is an area the GF3 just KILLED the competition.  And also exposed some nice speed differences. In the tests you can see the GF3 using V11.01 at stock speed is second at the top. But using the v12.00 driver it fourth from the bottom. In this test you can see that the v12 drivers suck all the way around on OpenGL performance for the GF2 and GF3.   You can see I did OC the GF3 a bit more as it seemed to remain stable at 520Mhz for the Ram in this test.

The GF3 Enhanced tests using the Pixel shaders and such, did take a bit of a performance impact on that was not bad, but at the same time made the benchmark look just amazing.

The W2K Benchmarks were the first I ran and forgot about the Pixel Shader Nature Demo issue when comparing the cards. So I ran an extra one to show the drop in 3DMark scores with the Pixel Shader Tests Disabled.

Once again in W2K you can see the nice scaling between the cards with the drivers playing almost no roll in performance differences.

What was really amazing is that in W2K 3DMark was giving higher scores then Windows 98 SE!!! THATS A SHOCK.   As its always been known that W2K's HAL layer was a major slowdown for many 3D apps. Seems Nv's drivers are overcoming those issues.

I was just in amazement in the performance difference in this test. You can see clearly that the v12.00 Dets still suck, but wait.. now a GF3 with a v12.00 scored the highest on the entire benchmark!!! All I have to say is this is one wacky driver under W2K, in one test is bites and in another it rules.  But then again we are talking a difference of 20 - 30 Marks. 

I didn't use GLMark much in the W2K test as I was more interested in seeing the Win98 Tests but I did run them all at 11.01 as it seemed to be the fastest driver for Open GL that Nv has out.  You can see there is a totally difference of 4Fps in this chart, a MUCH different story from the Win 98 test where we saw the differences of 30 - 40Fps.

This is a nice representation of what kind of performance degradation you can expect from using AA in games. You will be able to see the difference in the next part of this review of what the different AA levels look like.

The main thing is how close the Quincunx AA is to the 2x AA. Quincunx is a new method for doing AA that gives 4x AA at 2x Performance hit. You can see its a bit more of a hit, but really not much.

I wanted to see how fast I can run this card.  So I ran some strictly GF3 speed tests. I got the card running stable at 240/535. From the marks I got of 7028 and 6234 I have beat by over 400 the highest marks recorded on MadOnions site for those tests as of 4/17/01.

Its clear that at the 640x480 res the CPU is the limiting factor as the card is being very under worked.   At 1280x1024 I was just amazed that it still performed so well!! I ran the base speed Vs the OC speed for the 640x480 and the 1280x1024.