Most likely is that vsync is turned on and putting a cap on the fps. This happens a lot in 3dmark99, I think because it defaults to triple buffering. You have to use your video driver settings or a utility to force vsync off.

I just tested 3dmark99, 3dmark2000 and 3dmark01 in Athlon64 2.6Ghz win98se and Geforce 7800gs agp with the unofficial drivers and I got agains some weird results. Which do you think is the bottleneck in this system gpu or the cpu when benchmarking with those 3dmark?

Maybe the newer cards are just better optimized for later games, using more texture passes. Also a lot of emphasis was put on the pixel shader which old games don't use.

Here is some benchmarks with a Core2 Duo 3GHz, faster than what you are using:

Geforce 7950GT

3DMark99 Max 39289
3DMark2000 33857
3DMark2001SE 41466

Main difference with that is that I got higher score in 3dmark2000 than in 3dmark99.

3DMark99 doesn't use hardware T&L so it might be more dependent on the CPU.

My general feeling is that 3DMark2000 runs bad on newer cards, I think it it badly optimized (as are many games).

An even more recent graphics card, Radeon HD 3850:

3DMark99 Max 51579
3DMark2000 22368
3DMark2001SE 44852

I think using software T&L might speed up 3DMark2000 in this case

The point I was trying to make is that newer hardware doesnt always play well with older stuff, there is nothing odd about your results.

I tend to use games like Quake 3 for testing, it scales much better (up to a point)

1600x1200x32

Geforce 4 4600: 133
Geforce FX 5950 Ultra: 253
Geforce 6800 Ultra: 530
Geforce 7950GT: 705

Every Readme from every version of 3DMark says something like this:
"3DMark<your version> Benchmark Results are incompatible with results from the previous versions of 3DMark. You should not compare the results obtained from different versions."

My 3DMark03 score is way higher than my '01 score. My '05 score is far lower than my '03 and '01 scores. Different scoring systems produce different numbers.

I know they are different. I was more like trying to figure out if my cpu or gpu is the bottleneck with each version and if my card handles games better that use specific directX version compared to others with assumption that they would each year make the benchmark harder to have high scores considering the new hardware that has come out.

AGP only cards (expept GTS 450 and HD 5870 😉).
Overall score for 3DMark 99 Max likes CPU speed, and memory speed.
Rasteriser part should give you best GPU based score (when you pass certain CPU speed).
3DMark 2000 is basicly a CPU/RAM test on anything faster than 4600 Ti
3DMark 2001 SE is cache/memory benchmark (asuming you have CPU/GPU power to back it up).
3DMark 03 is best optimised overall, can scale with clock speed and GPU power WAY beyond fastest AGP cards.

Baoran wrote:

I know they are different. I was more like trying to figure out if my cpu or gpu is the bottleneck with each version and if my card handles games better that use specific directX version compared to others with assumption that they would each year make the benchmark harder to have high scores considering the new hardware that has come out.

-3DMark99 and 2000 are pretty much all on the CPU these days. You could strap an ancient GeForce 6800 Ultra to a Core i7 system and still be limited by the CPU.

-3DMark01 is also mainly CPU limited. On a Core i7 machine, it typically stops scaling with cards faster than the GTX 280. However, on modern systems, the motherboard's UEFI implementation can heavily influence your 3DMark01 score. I'm not exactly sure how it works, but I've seen differences as high as ~20k points between different motherboards running the same CPU, GPU, RAM, and freshly installed OS. But just because a motherboard sucks at 3DMark01 doesn't mean that it's a slow system...far from it! It just means that the system's UEFI is not optimized for whatever the hell it is that 3DMark01 likes. Because of this, 3DMark01 is no longer a good choice for benchmarking single-threaded CPU performance on modern UEFI based systems. There are just too many unknown variables.

For example: my main system seems to have a 3DMark01-friendly UEFI and scores 119,886 / 95,428 / 90,311 under XP/Win7/Win10. My HTPC, with similar CPU and GPU performance, only scores 82,520 / 78,490 under Win7/Win10.

-3DMark03 is easily the most GPU-limited version of the "classic" 3DMark range (1999-2006). On a modern Core i7 machine, you'll see scaling right up to the GTX 980Ti, especially if you increase the resolution.

-3DMark05 and 06 are much more CPU-limited than '03. With these versions, I believe Futuremark's goal was to more accurately represent the performance of a typical game engine, whereas 3DMark03 was more of a synthetic GPU benchmark.