Until I retest Unreal tournament on KV8 PRO we continue with some CPU benchmarks!
46. Super pi is a single thread application that care about IPC and clock frequency above all else! Let's see if dual channel makes a difference:
A. KU8
B. KV8 PRO
C. UL8
D. AV8

And it does! 10% just like the PR difference between the socket 939 Athlon 64 3500 and socket 754 Athlon 64 3200!! For the first time we get what AMD told everyone: dual channel leads to ~10% better performance!

47. CINEBENCH 2003 is an old version that is perfect for period correct single core systems and dual cores/hyper threaded ones. We are interested in the CPU score:
A. KU8
B. KV8 PRO
C. UL8
D. AV8

And just where you thought dual channel justifies the PR rating we got this! Absolutely no scaling at all! Cinebench cares about the frequency and only frequency!

48. Tom2D is the only GUI test in the entire suite; it is highly controversial with nforce2 wiping the floor with VIA on socket A and K8T800 wiping the floor with nforce3 on socket 754.
A. KU8
B. KV8 PRO
C. UL8
D. AV8

All 4 platforms have serious problems with GUI acceleration. And socket 939 perform worse than their socket 754 counterparts! Unless a miracle happens the PR of socket 939 Athlon's is vastly overestimated!

And we reach the last test: Unreal tournament 2004. One of the last DirectX8 games, UT 2k4 is a really good arena shooter. Umark benchmark utility refuses to run with UT 2k4 GOG edition that I have so I used fraps 3.4.7 to test the Death match Antalus map in spectate mode. After loading the map I waited 1 min before pressing F11 for benchmark. After 1 minute of testing I changed the resolution and waited again 1 minute before benchmarking and so on. I repeated 3 times and kept the final 3 runs at 1024*768, 1280*1024 and 1600*1200. Please note that I used maximum details and also sound!
37. UT 2K4 - 1024*768:
Abit KU8 - socket 754
2026-02-15 05:33:38 - UT2004
Frames: 7644 - Time: 60000ms - Avg: 127.400 - Min: 85 - Max: 183

Abit KV8 PRO - socket 754
2026-04-05 07:36:57 - UT2004
Frames: 7739 - Time: 60000ms - Avg: 128.983 - Min: 81 - Max: 202

Abit UL8 - socket 939
2026-02-21 06:05:17 - UT2004
Frames: 7686 - Time: 60000ms - Avg: 128.100 - Min: 73 - Max: 179

Abit AV8 - socket 939
2026-03-03 03:49:51 - UT2004
Frames: 7653 - Time: 60000ms - Avg: 127.550 - Min: 67 - Max: 190

There is nothing to separate the Athlon 64 3500 from the Athlon 64 3200 here! All 4 systems perform so close that we can not see any advantage from the dual channel platform!
Both sockets perform the same! Please note that I used the exact same stepping on both CPU's: E6 - socket 754 has only the part number ADA3200AIO4BX available which is E6 while on socket 939 you have 2 options: E3 with part number ADA3500DAA4BP and E6 with part number ADA3500DAA4BW. I used the latter.

38. Next resolution is 1280*1024:
Abit KU8 - socket 754
2026-02-15 05:37:21 - UT2004
Frames: 7560 - Time: 60000ms - Avg: 124.000 - Min: 71 - Max: 180

Abit KV8 PRO
2026-04-05 07:39:50 - UT2004
Frames: 7234 - Time: 60000ms - Avg: 122.567 - Min: 82 - Max: 195

Abit UL8 - socket 939
2026-02-21 06:08:16 - UT2004
Frames: 7401 - Time: 60000ms - Avg: 123.350 - Min: 60 - Max: 179

Abit AV8
2026-03-03 03:46:08 - UT2004
Frames: 7476 - Time: 60000ms - Avg: 124.600 - Min: 68 - Max: 175

Again socket 939 and dual channel have no advantage over the older brother!

39. UT 2K4 at the last resolution - 1600*1200:
Abit KU8 - socket 754
2026-02-15 05:35:22 - UT2004
Frames: 6493 - Time: 60000ms - Avg: 108.217 - Min: 68 - Max: 171

Abit KV8 PRO - socket 754
2026-04-05 07:35:26 - UT2004
Frames: 6902 - Time: 60000ms - Avg: 115.033 - Min: 81 - Max: 169

Abit UL8 - socket 939
2026-02-21 06:06:57 - UT2004
Frames: 6739 - Time: 60000ms - Avg: 112.317 - Min: 67 - Max: 161

Abit AV8 - socket 939
2026-03-03 03:47:52 - UT2004
Frames: 7249 - Time: 60000ms - Avg: 120.817 - Min: 83 - Max: 195

Even at the most demanding settings there is less than 5% difference between the dual channel platforms and single channel ones! Please note again the weakness of both ULI systems at 1600*1200 resolution!

5-6% boost is nothing significant for AGP (as all games until 2004 will run fine) and the extra memory capacity isn't useful for dual boot Win98/XP. Games suitable for AGP don't need 3GB RAM. It's more about local availability.

s939 will matter more with PCIe and games from 2005 which can benefit from 3GB RAM sometimes and the slight single core boost.

I have looked at local s754 offerings and it is much more numerous, there are even 3 full size PCIe boards on sale (Asus K8N4-E Deluxe being one of them). There are very few s939 boards on sale, only very expensive ones from people who know what they are selling. There are a few cheap complete PCs with Athlon 64 3700+ (s939, 2.2Ghz), but with MSI boards (average BIOS options, they were cheap boards).

I've got the ASRock Abomination (939A 790GMH) and thought it was pretty good. It's probably 5 years newer than a normal 939 board (age related reliability), solid state caps, SATA 2, PCI-e 2.0 (not that it is needed for GPUs that you'd be pairing with an Athlon 64/X2) PCI-e x1 for my Creative X-Fi. I had it with an Athlon 64 x2 until I bent a load of pins. It was a great middle step between a Pentium 4 system and Core 2 system. I could never get it to work on Windows 7 though (chipset issues). Having HDMI output right on the motherboard with a socket from 2004 is pretty cursed though.

Almoststew1990 wrote on 2026-04-08, 20:22:

II could never get it to work on Windows 7 though (chipset issues). Having HDMI output right on the motherboard with a socket from 2004 is pretty cursed though.

Hi! Could elaborate on these two?
I have a round of PSUs on the bench for repairs and next will be this one.

Windows 2K / XP / Vista? Any issues?
I tried XP and it was ok.
W7 64-bit had drivers issues but in the end it was ok but didn't test thoroughly.

Thanks for any input!

AlexZ wrote on 2026-04-07, 13:32:

s939 will matter more with PCIe and games from 2005 which can benefit from 3GB RAM sometimes and the slight single core boost.

I have looked at local s754 offerings and it is much more numerous, there are even 3 full size PCIe boards on sale (Asus K8N4-E Deluxe being one of them). There are very few s939 boards on sale, only very expensive ones from people who know what they are selling. There are a few cheap complete PCs with Athlon 64 3700+ (s939, 2.2Ghz), but with MSI boards (average BIOS options, they were cheap boards).

This is my feeling too! Socket 754 is more than enough for early Windows XP period; it already solves both the performance deficit that socket A has and the 5V rail problem! You do not actually need strong pci-express cards for that! I did not tested yet the pci-express platforms yet but I consider socket 939 more suited for late Windows XP era - 2005 and 2006 - when you really need more RAM and high performance video cards.

My best friend has that Asus! I never seen a socket 754 board with so many extra controllers: 8 SATA ports!, firewire, optical IN/OUT, game port... Performance wise is right below my NV8, no more than 1-2%. The main problem is that he tried installing 98SE on it and failed just like me!
754 is the bridge between the 9X era and the XP era.

Almoststew1990 wrote on 2026-04-08, 20:22:

I've got the ASRock Abomination (939A 790GMH) and thought it was pretty good. It's probably 5 years newer than a normal 939 board (age related reliability), solid state caps, SATA 2, PCI-e 2.0 (not that it is needed for GPUs that you'd be pairing with an Athlon 64/X2) PCI-e x1 for my Creative X-Fi. I had it with an Athlon 64 x2 until I bent a load of pins. It was a great middle step between a Pentium 4 system and Core 2 system. I could never get it to work on Windows 7 though (chipset issues). Having HDMI output right on the motherboard with a socket from 2004 is pretty cursed though.

That AsRock is indeed something else! Curious about Windows 7; I did a test install using a Kingston SSD on Abit AN8 SLI and X2 4800 with 4gb of RAM and 7 worked out of the box with all the drivers installed from the get go! The problem was performance and not stability. I did used a period incorrect GTX 560TI.

And we reached the end of the tests! The standings are as follows - Abit KU8 socket 754/ULI M1689 gets a 100% base score:
1. ABIT AV8 - VIA K8T800 PRO - SOCKET 939: 105.87%
2. ABIT UL8 - ULI M1689 - SOCKET 939: 105.52%
3. ABIT KV8 PRO - VIA K8T800 PRO - SOCKET 754: 101.28%
4. ABIT KU8 - ULI M1689 - SOCKET 754: 100%
There is only around 5.5% difference between socket 754 and socket 939, courtesy of dual channel and 1.000 MHz HTT! This make the performance rating of the Athlon 64 3500 unsustainable. The "real"PR should be Athlon 64 3400 and not 3500! This would not seem much - only 100 PR points between the AMD rating and "my rating".
However, in the mean time, I received the Athlon 64 3800 Newcastle and started testing it against the Athlon 64 Newcastle for socket 754 - same clock rate and L2 cache. I did only the code creatures, Aquamark and 3dmark tests but the difference in % is exactly the same as the difference between Athlon 64 3500 and Athlon 64 3200. That would indicate that the "real"PR of the 3800 should be 3600 and, by consequence, the PR of Athlon 64 4000 should be 3800!
So dual channel adds 200 PR points to the Athlon 64 - reflected in the lower clocked CPU's: Athlon 64 3000 & 3200 for socket 939 lose 200 MHz and got the same rating as their socket 754 counterparts.

For PCIe platforms I would suggest to add Cinebench R10. It still runs on Windows XP and should support SSE3. Keep Cinebench 2003 for comparison.