27 & 28 & 29. We go back to PCMARK, a general system benchmark. The first version is the 2002 one. We are interested in the individual scores. PC mark 2002 has tests for processor , memory and hard drive. On the KU8 system you can clearly see in Everest that a Raptor is installed in the system specifically for the hard drive tests:

Performance for all 3 systems is very close, practically within 2% across all tests. So CPU and RAM performance is equal for all AGP platforms.

nd22 wrote on 2025-11-06, 13:24:

27 & 28 & 29. We go back to PCMARK, a general system benchmark. The first version is the 2002 one. We are interested in the individual scores. PC mark 2002 has tests for processor , memory and hard drive. On the KU8 system you can clearly see in Everest that a Raptor is installed in the system specifically for the hard drive tests:

You're doing great job.
But something seems odd to me.
Older interfaces like PCI or AGP should show similar results, but yours deviates by 5-10% at 3Dmarck.
Newer (at the time) interfaces like SATA 1 should have issues, and all implementations should be significantly different.
AMD and Intel chipsets, even with SATA 2, showed significant differences.
But your results are completely identical.
Perhaps you could use a more specialized test for storage devices?

Thank you sir. SATA implementation is excellent, brilliant on KU8 - when we reach the game tests i will explain. I got crystal disk mark 5.1.2 and the scores remain very close for all 3 platforms. I think using a raptor has some benefits - my HDD scores are higher than those in period correct reviews.
Any idea on what to use?

Here a significant difference of 20+% is indicated in SATA.
https://www.aceshardware.com/pdfs/AcesHardware-65000298.pdf

PСmark shows the difference between the SSD and your Raptor at 1.5 times.
That's impossible, no matter how fast the Raptor is.

PS. Sorry for screenshot size, its "time appropriate" Fujitsu-Siemens 15" 1024@768

Raptor was chosen as it's the next fastest drive to SSD that can be used on s754 given that using SSDs with these early SATA controllers is problematic. It is tougher for the controller to deal with than any period correct drives. If period correct reviews show higher variance then driver issues could be the cause. To be sure about the 3d mark discrepancy on nForce3 nd22 would have to test on non ABIT nForce3 board to confirm the finding. It could be BIOS, driver or chipset related.

It was the implementation of SATA that revealed the nuances of the chipsets of those years.
After all, this was the first generation of SATA.
A difference of a couple of points with a total score of almost 2000 is impossible.
Maybe this would be useful
https://www.guru3d.com/download/as-ssd-benchmark/

SSD benchmarks are useful once we get to nForce 4. On nForce 3 using SSD is not reliable so there isn't much to benchmark. On ABIT nd22 reported he couldn't even get to post properly. On my Gigabyte board I could boot Windows XP but the first boot always froze. That is for SSD attached to SATA port.

AlexZ wrote on 2025-11-06, 16:38:

SSD benchmarks are useful once we get to nForce 4. On nForce 3 using SSD is not reliable so there isn't much to benchmark. On ABIT nd22 reported he couldn't even get to post properly. On my Gigabyte board I could boot Windows XP but the first boot always froze. That is for SSD attached to SATA port.

THIS!!! Alex is 101% right! Wait until nforce4 comes and no more problems with SSD! Wait for part 3 with PCI-express!

shevalier wrote on 2025-11-06, 15:35:

It was the implementation of SATA that revealed the nuances of the chipsets of those years. After all, this was the first genera […]
Show full quote

It was the implementation of SATA that revealed the nuances of the chipsets of those years.
After all, this was the first generation of SATA.
A difference of a couple of points with a total score of almost 2000 is impossible.
Maybe this would be useful
https://www.guru3d.com/download/as-ssd-benchmark/

The nuances you are talking about - I encountered them on socket A! On socket 754 SATA is very good across all systems and excellent on ULI and nforce4! I got 3 different sets of results for each system and SATA raptor is performing practically the same every single time on every system. Yes, there are problems with nforce3 and K8T800 and SSD's but with classic hard drives there are no such problems. Also, wait for the rest of the PCMARK.
PS: I am not using RAID, single drive or SSD with raptor in case of ULI.

shevalier wrote on 2025-11-06, 15:19:

Here a significant difference of 20+% is indicated in SATA. https://www.aceshardware.com/pdfs/AcesHardware-65000298.pdf […]
Show full quote

Here a significant difference of 20+% is indicated in SATA.
https://www.aceshardware.com/pdfs/AcesHardware-65000298.pdf

PСmark shows the difference between the SSD and your Raptor at 1.5 times.
That's impossible, no matter how fast the Raptor is.

PS. Sorry for screenshot size, its "time appropriate" Fujitsu-Siemens 15" 1024@768

The difference of 20% is in RAID! in single drive they even mention:
page 9: Single drive numbers were so similar on both setups that I did not even bother to discuss them. Basically VIA is about 3% faster (transfer rate) while NVIDIA's IDE implementation has around 5% lower CPU overhead. The RAID 0 results were much more interesting, however.
So in single drive SATA performance is similar!

AlexZ wrote on 2025-11-06, 16:38:

SSD benchmarks are useful once we get to nForce 4. O

Nope.
AS SSD tests the performance of storage devices, but it has become relevant with the advent of SSDs.
Don't focus on its name.
This utility tests HDDs very well, but it takes a long time.

nd22 wrote on 2025-11-06, 17:29:

So in single drive SATA performance is similar!

I saw tests comparing the nF3-150 with the K8T800, even with different HyperThreading speeds.
In terms of graphics card performance, they were identical.
If even the SATA tests show the same result, then the chipsets are, alas, completely identical.

Based on my experience:
- The nF-3 was ready for overclocking immediately out a box.
- K8T800 worked slightly better with peripheral devices.
In terms of processor and video subsystem performance, they were identical (1-3% depend of BIOS).
I don't know anything about ULI/Aladdin, as I've never encountered them.
PS. And then there are the eternal problems with VIA drivers.
That's why nVidia chipsets were considered more prestigious.

I did some testing of my GeForce 7600 GT AGP and found out it to be very slightly faster than the tested 7800 GS. 3998 marks in 3d mark 2003, 1600x1200 (anti-aliasing 4 sample, texture filtering anisotropic 16) on Athlon XP 2800+. It did better in Battle of Proxycon, Troll's Lair and Mother Nature. It seems to have better Fill Rate but worse Pixel Shader 2.0. 7600 GT with DDR2 did much worse than GDDR3 version.

30 & 31 & 32 & 33. Next version is PCMARK 2004 a far more strenuous test than the old 2002 version that also adds a few video tests for which you will get a graphic score. Let's see how things are standing:
A. NF8
B. KU8
C. KV8

CPU and RAM scores are pretty much the same for all 3 boards as they should be which – in my opinion – validates the methods used but of course I could be wrong!
For the first time we see something that we shouldn’t: nforce3 graphic score is lower than those of VIA and ULI. VIA also is slower than ULI, K8T800 is 2% behind M1689. Against all odds ULI wins a victory here. Maybe it’s just a fluke lets’ move on…

34 & 35 & 36 & 37. The final version is PCMARK 2005. This one features the same basic tests as the previous version only much tougher on the system. It has the same basic scores: CPU, RAM, CPU and HDD.

Well it was not a fluke; nforce3 AGP performance is lower than the other 2! This is repeatable; no matter how many times I ran the tests on 2 identical systems the scores stays the same (well, not identical but extremely close). This will be reflected in the scores for all 3D tests: aquamark, half life, Farcry and so on! Again ULI gets a win, even if it’s a symbolic one this time around as VIA makes a comeback being practically tied to ULI.

That’s why VIA and ULI outscore NVIDIA in the synthetic 3D tests (in the games too but let’s not anticipate) and we get the same results in purely CPU tests such as Super PI – AGP implementation on NF8 leaves much to be desired. VIA also got its problems that’s why ULI will get ahead in some of the games. ULI problem is the sound implementation which is really poor but, again, let’s not anticipates.

Does nForce3 run in AGP 8x mode and 66Mhz? See computer->overclock tab in Everest if there is anything unusual. It would be great to run Everest/Aida64 memory/cache test to see memory latency. This would rule out poor configuration by BIOS.

Yes, as someone here said it, integrated memory controller negates any chipset related enhancements and should make all processors perform equal on all the boards and indeed CPU and memory results are exceptionally close for all 4 boards (nforce4 will join later the party) however AGP performance is still motherboard related!
Also note the excellent hard drive performance for ULI!