The con of SLI board is lane splitting rather than compatibility. You cannot do x12 + x4 but it does x8 + x8 for you automatically. I don't want to waste x8 on wifi card or sound card. Given that we are often running pcie 3.0 cards on these old boards I would prefer not to do the split for higher bandwidth.

AlexZ wrote on 2025-06-25, 10:06:

The con of SLI board is lane splitting rather than compatibility. You cannot do x12 + x4 but it does x8 + x8 for you automatically. I don't want to waste x8 on wifi card or sound card. Given that we are often running pcie 3.0 cards on these old boards I would prefer not to do the split for higher bandwidth.

I mean even modern boards don't do that sort of splitting unless its via the chipset and integrated periphials so its just one of the many things you just have to live with.

Also I personally don't bother with WIFI cards for older systems . .a USB dongle works perfectly fine and solves the lack of PCIe 1x slots, I find cards to be a waste of a PCIe slot if they have a USB equivalent which is already using the PCI bus.

But I also avoid SLI on these older boards ...they usually don't have the grunt to do justice to SLI/XFire, found it pointless back in the day and like you would have liked more PCIe 1x/4x slots. But ..it was all the craze back then...even with all its issues and lack of support for games.

Trashbytes wrote on 2025-06-25, 08:16:

AFAIK all PCIe 16x slots are fully backwards compatible with 8x, 4x and 1x cards, I would imagine that if it can run a GPU then it'll be able to run everything else. Only time compatibility was ever an issue with PCIe was between PCIe 1.0 and 2.0/3.0 where the newer 2.0 cards would have issues with PCIe 1.0 slots.

Even the funky boards with that little SLI/Xfire card should be compatible also, that card simply splits the 16x lanes of the primary slot into two 8x slots to allow dual cards, but itll still be electrically compatible with all PCIe cards but will just share bandwidth with the GPU.

I personally have never heard of a PCIe 16x slot being only compatible with GPUs and I'm not even sure how they would go about doing that without causing issues, even a PCIe 1x slot can in theory run a 16x GPU due to how PCIe works.

They are electrically compatible, sure. But i've stuck a bunch of x4-x8 cards into this slots and i've seen a few issues.

One option is - motherboard does not POST at all or hangs on POST if something funky is installed in "main" pci-e x16.

Another option - when switching between x16 and x8x8 is involved it does not always work if smaller card installed into second slot. Main slot remains x16 and the card in second one simply is not detected. Swap the cards (videocard in second slot) and everything works, as long as first issue does not happen.

This is probably 99% software (BIOS), but still...

AlexZ wrote on 2025-06-25, 10:06:

The con of SLI board is lane splitting rather than compatibility. You cannot do x12 + x4 but it does x8 + x8 for you automatically. I don't want to waste x8 on wifi card or sound card. Given that we are often running pcie 3.0 cards on these old boards I would prefer not to do the split for higher bandwidth.

Does it really matter? I've seen whole lot of tests, realistically the difference will be within margin of error, a few % at best. At least as long as it is 2.0.

It also depends on how you use the system. If it is games only then sure. But multiple large pci-e allow doing fun stuff like using large SAS controllers, FC controllers, etc. Outside of actual servers which are usually quite unsuitable for running at home.

But if you want whole lot of pci-e the most fun thing i've played around with are those Chinese LGA2011 boards. 40 lanes, all exposed. 2 x16, 2 M.2 (x4). And bifurcation in BIOS, so each X16 can be split into x8x8 or x4x4x4x4. That's not counting a few x1 slots which use pci-e from PCH, not CPU.

Technically XP compatible too...

Archer57 wrote on 2025-06-25, 12:51:

They are electrically compatible, sure. But i've stuck a bunch of x4-x8 cards into this slots and i've seen a few issues. […]
Show full quote

Trashbytes wrote on 2025-06-25, 08:16:

AFAIK all PCIe 16x slots are fully backwards compatible with 8x, 4x and 1x cards, I would imagine that if it can run a GPU then it'll be able to run everything else. Only time compatibility was ever an issue with PCIe was between PCIe 1.0 and 2.0/3.0 where the newer 2.0 cards would have issues with PCIe 1.0 slots.

Even the funky boards with that little SLI/Xfire card should be compatible also, that card simply splits the 16x lanes of the primary slot into two 8x slots to allow dual cards, but itll still be electrically compatible with all PCIe cards but will just share bandwidth with the GPU.

I personally have never heard of a PCIe 16x slot being only compatible with GPUs and I'm not even sure how they would go about doing that without causing issues, even a PCIe 1x slot can in theory run a 16x GPU due to how PCIe works.

They are electrically compatible, sure. But i've stuck a bunch of x4-x8 cards into this slots and i've seen a few issues.

One option is - motherboard does not POST at all or hangs on POST if something funky is installed in "main" pci-e x16.

Another option - when switching between x16 and x8x8 is involved it does not always work if smaller card installed into second slot. Main slot remains x16 and the card in second one simply is not detected. Swap the cards (videocard in second slot) and everything works, as long as first issue does not happen.

This is probably 99% software (BIOS), but still...

AlexZ wrote on 2025-06-25, 10:06:

The con of SLI board is lane splitting rather than compatibility. You cannot do x12 + x4 but it does x8 + x8 for you automatically. I don't want to waste x8 on wifi card or sound card. Given that we are often running pcie 3.0 cards on these old boards I would prefer not to do the split for higher bandwidth.

Does it really matter? I've seen whole lot of tests, realistically the difference will be within margin of error, a few % at best. At least as long as it is 2.0.

It also depends on how you use the system. If it is games only then sure. But multiple large pci-e allow doing fun stuff like using large SAS controllers, FC controllers, etc. Outside of actual servers which are usually quite unsuitable for running at home.

But if you want whole lot of pci-e the most fun thing i've played around with are those Chinese LGA2011 boards. 40 lanes, all exposed. 2 x16, 2 M.2 (x4). And bifurcation in BIOS, so each X16 can be split into x8x8 or x4x4x4x4. That's not counting a few x1 slots which use pci-e from PCH, not CPU.

Technically XP compatible too...

That issue is possibly related to PCIe 2.0 cards not liking PCIe 1.0 boards and vice versa, there was a period of time where this was a big issue especially with GPUs but also affected other cards.

The other issues is cards that say they are compatible with 16x but ..well they did some shifty shit and the cards themselves are the problem and not the board.

Its a similar issue you might face with AGP 4x cards in an 8x slot ...not all 4x cards will work this way and require a legit 4x slot to work, same with 2x cards in a 4x slot. I have a bunch of 4x cards that refuse to post on a 8x board but work just fine in a 2x/4x board.

I have never actually used PCIe SLI or Crossfire as it's just not very appealing, but usually don't go out of my way to avoid these motherboards unless it's a micro ATX board. I rarely need more than 1 card aside from the GPU these days and on old systems I prefer to use the onboard controllers as part of the experience but perhaps add an X-Fi Ti.

On the PCIe compatibility front, I've had fun trying to run Radeon RX 6500XT, RX 550, and R7 240 on nForce 4 with varying degrees of success. I also found the VOGONS favorite Datapath VisionRGB-E1S capture card, a bridged PCI-X design that worked on my Z370 motherboard, causes problems with POST and is not detected on my newer Z690 system.

I played around with various BIOS versions on Gigabyte GA-MA770-UD3 v2.0 and noticed that in the last few BIOS versions Gigabyte started to hide OC related settings.

FKD version doesn't have Advanced Clock Calibration anymore.
FI version and later are missing DDR VTT Voltage Control
FG version and later are missing SB/HT Voltage Control

So the best BIOS version to use is FF which has all OC settings. This holds for v2.0 and v2.1.

Interestingly users reported problems with running memory at DDR800 when paired with Brisbane CPU having non-integer multiples - Re: Which Athlon 64 X2 should I chose?
In Gigabyte GA-MA770-UD3 BIOS there is CPU Host Clock Control set to 200Mhz and memory multiplier which can be set to 4x and 5.33 among other options. I don't see a reason for not being able to run at full DDR800 speed.

I put together a temporary AM2 build on a shoe box to run some benchmarks. I used motherboard Gigabyte GA-MA770-UD3 v2.0 and CPU is Athlon 64 X2 6000+ ADX6000IAA6CZ (Windsor). I also have 6400+ but chose to benchmark 6000+ first as it will be more common, is cheap and is probably what people will be buying for AM2 builds. Interestingly its rated TDP is 125W just like 6400+. There exists also ADA6000IAA6CZ which has TDP 89W. I used 4x DDR2 ADATA Vitesta 1066 2GB sticks and Samsung 850 EVO SSD. I use 8GB RAM as this build is eventually meant to run Windows 7 as well. It is a temporary build because I do not have a case yet as a good semi-modern mesh case is very hard to find.

I didn't add VRM cooling to the board yet. I have VRM heatsink from another modern Asus board that can be modified to fit this board. Therefore, I used a large fan blowing air from side as it's quite hot these days.

The CPU boots without thermal paste but very quickly reaches 55'C in BIOS. As this CPU has higher TDP, I used Arctic Freezer 64 Pro which is a little cooler with 6 heatpipes, better than AMD stock coolers without heatpipes. I would not run any workload on this CPU without thermal paste, just do a quick test that it still works and we can enter BIOS setup. With MX-4, the temperature is much lower and the chosen cooler is more than sufficient.

I installed Windows XP SP3 on a partition prepared by gparted (Ubuntu live cd) with SATA controller in AHCI mode.

Findings:
- when AHCI mode is enabled, we get to see AMD AHCI BIOS 2007 during boot. Detected SATA drives are not displayed in the main BIOS setup, only in the AMD AHCI BIOS (ctrl+A) - similar solution I have seen in Athlon XP and some Athlon 64 boards.
- with Windsor CPU, BIOS setup allows selection of DDR speed, not multiplier - e.g. DDR533, DDR 667, DDR 800. When DDR 800 or AUTO is selected, what happes is divisor 8 is selected that is used to divide CPU clock speed to determine real memory speed (3000 / 8 = 375Mhz) . There is no 7.5 divisor, therefore memory runs as DDR 750 instead of DDR 800. This could be worked around by OCing the CPU a little bit. Memory multiplier is supposed to be selectable with Phenom CPU.
- we cannot configure Ganged/Unganged memory controller mode for Windsor CPU. This means both controllers form a 128bit bus rather than working independently. Not a big issue for X2 CPU and probably better for Windows XP.

I ran 3d mark 2003, 2005, 2006 in 1024x768 for comparison with Athlon 64 3400+ and in 1600x900, which I see as ideal resolution for a late Windows XP/Vista build.

Athlon XP 64 3400+ measurements were done to compare it with Athlon XP ( Re: High-end Socket462/A build. ). We don't look at 3d mark score but FPS of individual games which is a better metric. Identical GeForce 9800 GT was used for both Athlon 64 3400+ and Athlon 64 X2 6000+.

3d mark 2003 breakdown, 1024x768, Athlon 64 3400+, GeForce 9800 GT:

• Wings of Fury - 318 fps
• Battle of Proxycon - 255 fps
• Troll's Lair - 202 fps
• Mother Nature - 203 fps

3d mark 2003 breakdown, 1024x768, Athlon 64 X2 6000+, GeForce 9800 GT:

• Wings of Fury - 525fps
• Battle of Proxycon - 297fps
• Troll's Lair - 230fps
• Mother Nature - 224 fps

3d mark 2003 breakdown, 1600x900, Athlon 64 X2 6000+, GeForce 9800 GT:

• Wings of Fury - 447fps
• Battle of Proxycon - 207fps
• Troll's Lair - 165fps
• Mother Nature - 190 fps

3d mark 2003 represents games with low CPU load that were already well playable on Athlon XP / 64.

3d mark 2005 breakdown, 1024x768, Athlon 64 3400+, GeForce 9800 GT:

• Return To Proxycon - 35fps
• Firefly Forest - 28fps
• Canyon Flight - 74fps

3d mark 2005 breakdown, 1024x768, Athlon 64 X2 6000+, GeForce 9800 GT:

• Return To Proxycon - 57fps
• Firefly Forest - 45fps
• Canyon Flight - 107fps

3d mark 2005 breakdown, 1600x900, Athlon 64 X2 6000+, GeForce 9800 GT:

• Return To Proxycon - 56fps
• Firefly Forest - 45fps
• Canyon Flight - 93fps

3d mark 2005 represents more CPU heavy games from year 2005 where Athlon XP / 64 is overloaded. "Return To Proxycon" is CPU bound as we got the same score for 1024x768 as we did for 1600x900. The fps we get at 1600x900 is playable unlike on Athlon 64 3400+, although it isn't that great.

3d mark 2006 breakdown, 1024x768, Athlon 64 X2 6000+, GeForce 9800 GT:

• Return To Proxycon - 42fps
• Firefly Forest - 43fps

3d mark 2006 breakdown, 1600x900, Athlon 64 X2 6000+, GeForce 9800 GT:

• Return To Proxycon - 41fps
• Firefly Forest - 41fps

3d mark 2006 represents even more CPU heavy games from 2006. We are again CPU bound and lowering resolution would not help.

A period correct match for Athlon 64 X2 6000+ would have been a GeForce 8800 GT which I do not have, therefore I used 9800 GT (even though I have a GTX version too) - just a basic one as these are available, very cheap. They are slim cards that need just one 6pin pcie connector and do not produce much heat. GeForce 280 is very rare, many of them probably died from running hot, just like GeForce 8800 GT.

I also tested a few games that caused trouble to Athlon 64 3400+
- Test Drive Unlimited - runs about 2x smoother, Athlon 64 3400+ gets to about 23fps on highways which is not great although still somewhat playable.
- Need for Speed: Most Wanted (2005) - we get much lower CPU load, can refresh car reflections faster and run in higher resolution. There is less stutter although it doesn't feel like it's gone completely. Probably a game engine issue as it happens even at 60-70fps.
- The Sims 2 - SSD and faster CPU help to avoid occasional music cut off Athlon 64 3400+ experienced. Athlon 64 3400+ did get sufficient fps, but the music cut off could be annoying.
- Lock On: Modern Air Combat didn't run that great on Athlon 64 3400+, only about 23fps on max details. Athlon 64 X2 6000+ runs much better, we get about 35fps.

More demanding games were not tested yet as I only re-tested demanding games I have on Athlon 64 3400+.

I have many more GPUs and CPUs that I will test. I will also test a few more demanding games with built-in benchmarks.

Meanwhile got my "free" 6400+ with missing pins, only one actually important pin is missing related to memory channel B, so the CPU works, as long as only channel A is used.

Any tips/experience soldering pins on this CPUs? Looks like the pad is still there, so should be able to steal one of the unused pins and replace the important one.

One advantage 6400+ has over 6000+ is memory frequency - in 6400+ it works out to be exactly right - it'll be DDR2-800, not 750.

Also yeah, those "AMD AHCI controller". Have this on my phenom2 system and had plenty of issues with it. In fact it was the reason i retired this sustem and replaced it with LGA1155 back then - SSD compatibility issues, driver issues in win8, etc...

Funnily enough nforce based boards work better with storage, be it socketA or AM2...

I would also try to test some ECC RAM, it should apparently be supported and... why not? I have the sticks anyway...

I would just get another 6400+. I received a CPU with missing memory channel B pin as well. It ended in rubbish bin. Seller claimed they tested everything, it worked fine. After mentioning I would be leaving negative feedback they agreed to refund. It was a seller with lot of items. Those rarely test items properly. GeForce GTX 780 I received first wasn't tested properly either and seller ended up sending another one that actually worked. And then lastly the FX CPU intended for my Phenom II build didn't work either.

Back in the day the RAM speed defeciency wasn't that of a big deal as people overclocked CPUs. So they would run it at 3.1Ghz and get DDR775. Lower DDR divisors than 8 do exist. When cool&quiet is enabled and special AMD CPU driver used in Windows XP to underclock the CPU, DDR divisor is lowered as well. There may be a tool that can set it independently.

Additional games tested:
- Far Cry (2004) - playable in 1600x1200 with max settings, water on ultra. About 60fps when looking at the whole island outside cave at the beginning of campaign. Ultra water quality decreases fps by about 15.
- Quake 4 (2005) - runs fine
- F.E.A.R. (2005) - runs fine, in 1600x1200 with max settings we get in built-in benchmark 78 fps average, 47 fps minimum
- DiRT (2007) - runs fine
- Need for Speed: ProStreet (2007) - runs fine
- Call of Duty 4 Modern Warfare (2007) - in 1600x1200 with max details and textures we get 65-130 fps (lowest when aiming at tower with scope) outside the training ground.
- World in Conflict (2007) - in 1600x1200 with most settings on max (couldn't set all on max due to game having an issue with applying settings) we get 32 average fps in built-in benchmark. We get 60-70 fps in the training mission. Playable.
- Crysis (2007) - needs a stronger GPU, possibly CPU as well. Can only use high settings on 9800 GT, not ultra. In 1280x960 we get in built-in benchmark 39 fps average, 21 fps minimum. Borderline playable.
- STALKER (2007) - in 1600x1200, first mission outside we get about 27-38 fps. Borderline playable.
- Far Cry 2 (2008) - with max settings, in 1280x960 we get about 30-35 fps in the initial jeep ride. Borderline playable. About 450MB of GPU memory is used.
- Witcher (2008) - in 1600x1200 we get about 40-45 fps during the initial fight. Borderline playable.
- Need for Speed: Undercover (2008) - needs stronger CPU and possibly GPU. FPS is not high enough for pleasant experience even in 1280x960.
- Need for Speed: Shift (2009) - seems to be less demanding than previous version. Playable from inside car in 1600x1200, CPU utilization about 70% so with better GPU we should be able to get more.

Conclusion about Athlon 64X2 6000+ (Windsor) with GeForce 9800 GT :
- it is a good choice for Windows XP era games (2002-2006). Athlon 64 3400+ (socket 754) covers 2002-2005. Athlon XP covers 2002-2004 due to limitations of AGP cards.
- it covers the Windows Vista era (2007-2009) poorly. Titles from 2007 should be playable, some with reduced settings.
- 30-40 fps in FPS games in 1600x1200 is a big problem. We need to get to 50-60 at least.
- GeForce 9800 GT was a mid-range graphics card so the results are not surprising

Next steps:
- upgrade GPU to GeForce GTX 260 used by Athlon 64 3400+ system to see how it helps
- retest 3d mark and problematic titles where GeForce 9800 GT struggled

AlexZ wrote on 2025-07-04, 15:54:

I would just get another 6400+. I received a CPU with missing memory channel B pin as well. It ended in rubbish bin. Seller claimed they tested everything, it worked fine. After mentioning I would be leaving negative feedback they agreed to refund. It was a seller with lot of items. Those rarely test items properly. GeForce GTX 780 I received first wasn't tested properly either and seller ended up sending another one that actually worked. And then lastly the FX CPU intended for my Phenom II build didn't work either.

Why?

I got this one on purpose. The seller was actually very nice, i bought the CPU from him, he then discovered its state, cancelled/refunded and after a bit of a discussion offered to just ship it for free if i want to. I bought a few things from them before...

This pins are not super hard to replace, the biggest issue that the stuff is very small and hard to work with. I could get another CPU, but why not at least try fixing this one if i can? Fixing stuff like this is half of the fun messing with old hardware...

AlexZ wrote on 2025-07-04, 15:54:

Back in the day the RAM speed defeciency wasn't that of a big deal as people overclocked CPUs. So they would run it at 3.1Ghz and get DDR775. Lower DDR divisors than 8 do exist. When cool&quiet is enabled and special AMD CPU driver used in Windows XP to underclock the CPU, DDR divisor is lowered as well. There may be a tool that can set it independently.

Yeah, it is not a very big deal and it can be OCd, but still a small extra difference. Also 6400+ is supposedly unlocked, but i've seen different opinions on that. Need to try it...
And then again - i am not a huge fan of overclocking this old HW outside of short experiment...

AlexZ wrote on 2025-07-04, 15:54:

Conclusion about Athlon 64X2 6000+ (Windsor) with GeForce 9800 GT : - it is a good choice for Windows XP era games (2002-2006). […]
Show full quote

Conclusion about Athlon 64X2 6000+ (Windsor) with GeForce 9800 GT :
- it is a good choice for Windows XP era games (2002-2006). Athlon 64 3400+ (socket 754) covers 2002-2005. Athlon XP covers 2002-2004 due to limitations of AGP cards.
- it covers the Windows Vista era (2007-2009) poorly. Titles from 2007 should be playable, some with reduced settings.
- 30-40 fps in FPS games in 1600x1200 is a big problem. We need to get to 50-60 at least.
- GeForce 9800 GT was a mid-range graphics card so the results are not surprising

I'll probably just pair mine with GTX660 and see how it compares to E8600. I may be weird, but to me once it's pci-e it feels modern and i see no point trying to mess with period-correct pci-e cards, especially considering bumpgate...

My guess would be - with fast enough GPU this can run pretty much any XP game. What would be interesting to see if there is any practical difference between it and E8600 for XP games...

Archer57 wrote on 2025-07-04, 17:00:

Yeah, it is not a very big deal and it can be OCd, but still a small extra difference. Also 6400+ is supposedly unlocked, but i've seen different opinions on that. Need to try it...
And then again - i am not a huge fan of overclocking this old HW outside of short experiment...

It's just 100Mhz to fix the RAM issue. Buying a 5600+ and trying to turn it into 6400+ makes no sense.

Testbench 2:
- Gigabyte GA-MA770-UD3 v2.0
- Athlon 64 X2 6000+ ADX6000IAA6CZ (Windsor)
- Gainward GeForce GTX 260 Golden Sample (factory clocks, 625 GPU, 1100 memory, 1350 shader). This card has two big fans, unlike many period correct cards, which is probably why it survived. NVidia driver 177.83

3d mark 2003 breakdown, 1024x768, Athlon 64 3400+, GeForce GTX 260:

• Wings of Fury - 326 fps
• Battle of Proxycon - 334 fps
• Troll's Lair - 247 fps
• Mother Nature - 240 fps

3d mark 2003 breakdown, 1024x768, Athlon 64 X2 6000+, GeForce GTX 260:

• Wings of Fury - 561 fps
• Battle of Proxycon - 433 fps
• Troll's Lair - 320 fps
• Mother Nature - 274 fps

3d mark 2003 breakdown, 1600x900, Athlon 64 X2 6000+, GeForce GTX 260:

• Wings of Fury - 523 fps
• Battle of Proxycon - 331 fps
• Troll's Lair - 257 fps
• Mother Nature - 245 fps

3d mark 2003 breakdown, 1600x1200, Athlon 64 3400+, GeForce GTX 260:

• Wings of Fury - 307 fps
• Battle of Proxycon - 244 fps
• Troll's Lair - 196 fps
• Mother Nature - 206 fps

3d mark 2003 breakdown, 1600x1200, Athlon 64 X2 6000+, GeForce GTX 260:

• Wings of Fury - 504 fps
• Battle of Proxycon - 286 fps
• Troll's Lair - 223 fps
• Mother Nature - 228 fps

3d mark 2003 represents games with low CPU load that were already well playable on Athlon XP / 64. Scores in 1600x1200 were included for comparison with Athlon 64 3400+. Because the scores are too high, it makes little sense to use 3d mark 2003 for testing of further GPU upgrades.

3d mark 2005 breakdown, 1024x768, Athlon 64 3400+, GeForce GTX 260:

• Return To Proxycon - 35 fps
• Firefly Forest - 28 fps
• Canyon Flight - 76 fps

3d mark 2005 breakdown, 1024x768, Athlon 64 X2 6000+, GeForce 9800 GT:

• Return To Proxycon - 57 fps
• Firefly Forest - 45 fps
• Canyon Flight - 107 fps

3d mark 2005 breakdown, 1024x768, Athlon 64 X2 6000+, GeForce GTX 260:

• Return To Proxycon - 57 fps
• Firefly Forest - 45 fps
• Canyon Flight - 120 fps

3d mark 2005 breakdown, 1600x900, Athlon 64 X2 6000+, GeForce GTX 260:

• Return To Proxycon - 56 fps
• Firefly Forest - 45 fps
• Canyon Flight - 113 fps

3d mark 2005 represents more CPU heavy games from year 2005. We see a performance improvement over GeForce 9800 GT in Canyon Flight only.

3d mark 2006 breakdown, 1024x768, Athlon 64 X2 6000+, GeForce 9800 GT:

• Return To Proxycon - 42 fps
• Firefly Forest - 43 fps
• Canyon Flight - 59 fps
• Deep Freeze - 53 fps

3d mark 2006 breakdown, 1024x768, Athlon 64 X2 6000+, GeForce GTX 260:

• Return To Proxycon - 42 fps
• Firefly Forest - 43 fps
• Canyon Flight - 81 fps
• Deep Freeze - 53 fps

3d mark 2006 breakdown, 1600x900, Athlon 64 X2 6000+, GeForce 9800 GT:

• Return To Proxycon - 41 fps
• Firefly Forest - 41 fps
• Canyon Flight - 47 fps
• Deep Freeze - 50 fps

3d mark 2006 breakdown, 1600x900, Athlon 64 X2 6000+, GeForce GTX 260:

• Return To Proxycon - 41 fps
• Firefly Forest - 42 fps
• Canyon Flight - 70 fps
• Deep Freeze - 52 fps

3d mark 2006 represents even more CPU heavy games from 2006. We get benefit in SM3.0 benchmarks only, SM2.0 is the same. The upgrade may pay off in titles using shader model 3.

We do not use 3d mark vantage as that is a dx10 benchmark.

Performance gains from GeForce GTX 260 are unconvincing in benchmarks but we will see in real games. In Athlon 64 3400+, GeForce GTX 260 does help with CPU light scenes like stadium views in Fifa 2007.

Games tested:
- F.E.A.R. (2005) - in 1600x1200 with max settings we get 119 fps average, 58 fps minimum in built-in benchmark
- Call of Duty 4 Modern Warfar (2007) - in 1600x1200 we get 85-200 fps (lowest when aiming at tower with scope)
- World in Conflict (2007) - in 1600x1200 we get 38 average fps in built-in benchmark with same settings as before. The game does allow us to increase visual quality further compared to 9800 GT, we then geet 30 average fps in built-in benchmark.
- Crysis (2007) - in 1280x960 we get in built-in benchmark 49 fps average, 27 fps minimum. We are getting into playable territory in 1280x960, but our target is 1600x1200. In 1600x1200 average fps in benchmark is 45, min fps 24.
- STALKER (2007) - in 1600x1200, first mission outside we get about 45-80 fps. We are in playable territory.
- Far Cry 2 (2008) - with max settings, in 1280x960 we get about 30-40 fps in the initial jeep ride.
- Witcher (2008) - in 1600x1200 we get about 40-45 fps during the initial fight.
- Need for Speed: Undercover (2008) - in 1600x1200 we get 32-40 fps in the city at the main menu. 100% cpu utilization and lowering resolution does not help.
- Need for Speed: Shift (2009) - 1600x1200 we get 55-70 fps during race from inside car

Conclusion about Athlon 64X2 6000+ (Windsor) with GeForce GTX 260 :
- it is a good choice for Windows XP era games (2002-2006)
- mediocre coverage of Windows Vista era (2007-2009) due to slow CPU. We didn't improve Need for Speed: Undercover or Far Cry 2. Crysis improved but not enough.
- we saw improvement in titles such as F.E.A.R., Call of Duty 4, World in Conflict, Crysis, STALKER, Need for Speed: Shift. GeForce GTX 260 helped to make the gaming experience visibly more enjoyable over 9800 GT.
- GeForce GTX 260 may be sufficient for Windows Vista era with a fast Intel core 2 duo.
- we are not happy with performance in Crysis, Far Cry 2, Need for Speed: Undercover.

Next steps:
- upgrade GPU to GeForce GTX 480 to see how it helps
- retest 3d mark 2005/2006 and problematic titles where GeForce GTX 260 struggled
- very minor improvement is expected as specs of GeForce GTX 480 are only marginally better than Gainward GeForce GTX 260 golden sample

Archer57 wrote on 2025-07-04, 17:00:

I'll probably just pair mine with GTX660 and see how it compares to E8600. I may be weird, but to me once it's pci-e it feels modern and i see no point trying to mess with period-correct pci-e cards, especially considering bumpgate...

My guess would be - with fast enough GPU this can run pretty much any XP game. What would be interesting to see if there is any practical difference between it and E8600 for XP games...

I do it gradually as I would like to see effect of both CPU and GPU upgrades. I will not be testing all combinations. The last GPU for Athlon 64 X2 6000+ will probably be GeForce GTX 480 as game benchmarks clearly show we need a faster CPU. I will then switch to Brisbane and Athlon 64 X2 6400+ with GTX 580. That should be more comparable to your setup. Brisbane has slightly higher L2 cache latency, lower clocks and smaller cache. GTX 770 will be more of Phenom 1 territory.

good testing AlexZ thanks for sharing the results,
I always wondered about the experience with "maxed out Am2", with the 6000 or 6400, AM2+ motherboard, and 1066 DDR2,

a question, does the AM2+ board allow for HT overclock over something like 1100Mhz? I never had success going over that but always had AM2 boards only

We will definitely see a maxxed out AM2. I will try a temporary OC while I still have the Athlon 64 X2 6000+ CPU in it. I will push it to 3.2Ghz max. I will also try slight HT OC. I don't want to push it too much as I haven't installed VRM heatsink yet and it tends to be hot these days.

My socket 754 Athlon 64 3400+ board allows to select 1000Mhz HT in BIOS (with 5x HT multiplier), but the cpu is rated only for 800Mhz and it wasn't stable above about 880Mhz and would refuse to boot. It ran fine at 2.3Ghz but had occasional instability at 2.4Ghz (I didn't increase voltage though). Could also be nforce4 chipset running too hot as it doesn't have a fan yet.

Testbench 3:
- Gigabyte GA-MA770-UD3 v2.0
- Athlon 64 X2 6000+ ADX6000IAA6CZ (Windsor, released in 2007)
- Asus GeForce GTX 480 (factory clocks, 701 GPU, 924 memory, 1401 shader, released in 2010). NVidia driver 197.41

We skip 3d mark 2003 as the scores are too high and we get no practical benefit from running 2003-2004 era games with higher fps.

3d mark 2005 breakdown, 1024x768, Athlon 64 X2 6000+, GeForce GTX 260:

• Return To Proxycon - 57 fps
• Firefly Forest - 45 fps
• Canyon Flight - 120 fps

3d mark 2005 breakdown, 1024x768, Athlon 64 X2 6000+, GeForce GTX 480:

• Return To Proxycon - 57 fps
• Firefly Forest - 45 fps
• Canyon Flight - 126 fps

3d mark 2005 breakdown, 1600x900, Athlon 64 X2 6000+, GeForce GTX 260:

• Return To Proxycon - 56 fps
• Firefly Forest - 45 fps
• Canyon Flight - 113 fps

3d mark 2005 breakdown, 1600x900, Athlon 64 X2 6000+, GeForce GTX 480:

• Return To Proxycon - 56 fps
• Firefly Forest - 45 fps
• Canyon Flight - 122 fps

3d mark 2005 represents more CPU heavy games from year 2005.

3d mark 2006 breakdown, 1024x768, Athlon 64 X2 6000+, GeForce GTX 260:

• Return To Proxycon - 42 fps
• Firefly Forest - 43 fps
• Canyon Flight - 81 fps
• Deep Freeze - 53 fps

3d mark 2006 breakdown, 1024x768, Athlon 64 X2 6000+, GeForce GTX 480:

• Return To Proxycon - 42 fps
• Firefly Forest - 43 fps
• Canyon Flight - 93 fps
• Deep Freeze - 53 fps

3d mark 2006 breakdown, 1600x900, Athlon 64 X2 6000+, GeForce GTX 260:

• Return To Proxycon - 41 fps
• Firefly Forest - 42 fps
• Canyon Flight - 70 fps
• Deep Freeze - 52 fps

3d mark 2006 breakdown, 1600x900, Athlon 64 X2 6000+, GeForce GTX 480:

• Return To Proxycon - 41 fps
• Firefly Forest - 42 fps
• Canyon Flight - 88 fps
• Deep Freeze - 52 fps

3d mark 2006 represents even more CPU heavy games from 2006. We see improvement only in SM3.0 benchmark Canyon Flight as the system is CPU bottlenecked. Users would have done better by upgrading the CPU instead of GPU in 2010. Athlon 64 X2 6000+ (Windsor) was released in 2007 but aged very quickly and is underpowered for Windows Vista era.

We do not use 3d mark vantage as that is a dx10 benchmark.

Games tested:
- Need for Speed: Hot Pursuit 2 (2002) - we see a major performance regression observed in Athlon 64 3400+ as well. It basically makes the game unplayable.
- F.E.A.R. (2005) - in 1600x1200 with max settings we get 137 fps average, 53 fps minimum in built-in benchmark. Very minor improvement over GTX 260.
- Call of Duty 4 Modern Warfar (2007) - in 1600x1200 we get 140-250 fps (180 when aiming at tower with scope). Noticeable improvement over GTX 260.
- World in Conflict (2007) - in 1600x1200 we get 37 average fps in built-in benchmark with same settings as before. The game does allow us to increase visual quality further, we then geet 29 average fps in built-in benchmark. Runs at 60-70 fps in training mission. No improvement over GTX 260.
- Crysis (2007) - in 1600x1200 average fps in benchmark is 45, min fps 11. 45-80 fps in the initial mission after swiming out of water to beach. Playable, but we saw no action yet. For some reason we get much lower min fps than with GTX 260.
- STALKER (2007) - in 1600x1200, first mission outside we get about 50-120 fps. Noticeable improvement over GTX 260.
- Far Cry 2 (2008) - with max settings, in 1280x960 we get about 35-45 fps in the initial jeep ride. Very minor improvement over 9800 GT/GTX 260
- Witcher (2008) - in 1600x1200 we get about 40-45 fps during the initial fight. We are CPU bottlenecked as the game cannot take advantage of 2 cores.
- Need for Speed: Undercover (2008) - in 1600x1200 we get 32-40 fps in the city at the main menu. 100% cpu utilization and lowering resolution does not help.
- Need for Speed: Shift (2009) - 1600x1200 we get 55-65 fps during race from inside car. No improvement over GTX 260.

Conclusion about Athlon 64X2 6000+ (Windsor) with GeForce GTX 480 :
- it is a good choice for Windows XP era games (2002-2006)
- mediocre coverage of Windows Vista era (2007-2009) due to slow CPU. We didn't improve Need for Speed: Undercover, Need for Speed: Shift, Far Cry 2 or Crysis.
- we saw improvement in titles such as F.E.A.R., Call of Duty 4, STALKER.
- GeForce GTX 480 may be sufficient for Windows Vista era with a fast Intel core 2 duo.
- we are not happy with performance in Crysis, Far Cry 2, Need for Speed: Undercover. We saw a regression in min fps in crysis.
- further GPU upgrades are of limited value as the remaining titles are CPU bottlenecked
- Athlon 64X2 6000+ (Windsor) seems to be a good match with GTX 260 for Windows XP gaming and partial coverage of Windows Vista era. Due to regression in Crysis in min fps and Need for Speed: Hot Pursuit 2 I would recommend GTX 260/280 over GTX 480 with Athlon 64X2 6000+ (Windsor).

Next steps:
- OC Athlon 64X2 6000+ to 3.1Ghz and 3.2Ghz and see how it helps.
- retest 3d mark 2006 and Crysis benchmark

So just for comparison a few rough benchmarks on those C2D E8600+GTX660 (driver 368.69):

Crysis, 1920x1080 everything on high (very high not possible in xp). Benchmark - 31min, 61avg, 84max. 70-100 in the beginning of new game. Load displayed by MSI Afterburner is GPU - 100%, CPU (Total) - 60-80%. Can not run 1600x1200 because of the monitor...

3Dmark 06, 1024x768. Should be completely CPU limited, as in your case, which is good:

- Return To Proxycon - 58 fps
- Firefly Forest - 61 fps
- Canyon Flight - 120 fps
- Deep Freeze - 76 fps

And also ran 3dmark 06 at 1920x1080, showing it is CPU limited indeed....

- Return To Proxycon - 58 fps
- Firefly Forest - 61 fps
- Canyon Flight - 108 fps
- Deep Freeze - 77 fps

So there is an improvement, but it is not massive. This CPUs are roughly a year apart - 6000+/6400+ is 08.2007, E8600 is 08.2008 if i am not mistaken.

We are going to explore various OC settings.

BIOS settings:
- the CPU has max multiplier 15 in BIOS setup, but it can be lowered.
- memory clock can be set to max DDR800. Other options are DDR667, DDR533, we cannot set the divisor. Memory clock is then determined by CPU clock divided by a DDR divisor.
- HT link setting allows us to set 1Ghz, 800Mhz, 600Mhz, not multiplier. Internally it seems to set HT multiplier.

Experiment 1:
- we bump up the base clock to 208Mhz and reduce CPU multiplier to 14.5, we get 1040Mhz HT. So it turns out we can OC the HT. CPU clock is still 3Ghz.

Experiment 2:
- we bump up the base clock to 215 and reduce CPU multiplier to 14, we get 1075Mhz HT. CPU clock is 3009Mhz
- interestingly, the DRAM divisor is now 7 instead of 8 and we get DRAM frequency 430Mhz instead of 375Mhz with standard settings. This corresponds to DDR2 860. In Bios, DDR800 is selected.

Experiment 3:
- we bump up the base clock to 232 and reduce CPU multiplier to 13, we get 1160Mhz HT. CPU clock is 3015Mhz
- DRAM divisor is still 7, DRAM frequency is 430Mhz

Experiment 4:
- we bump up the base clock to 251 and reduce CPU multiplier to 12, we set HT link 800 in BIOS (to reduce HT multiplier). We get 1000Mhz HT. CPU clock is 3012Mhz
- DRAM divisor is still 7, DRAM frequency is 430Mhz
- this is a nice way to stay at 3Ghz, 1Ghz HT but have DDR2 860 instead of the original 750

Experiment 5:
- we bump up the base clock to 251 and reduce CPU multiplier to 12, we set HT link 1000 in BIOS. We get 1255Mhz HT. CPU clock is 3012Mhz
- DRAM divisor is now 6, DRAM frequency is 502Mhz
- DRAM divisor is somehow linked to CPU multiplier and HT multiplier
- this is a nice way to stay at 3Ghz, get DDR2 1000 and HT 1.25Ghz
- I configured memory timings to 5 5 5 15 and 2.2V as written on ADATA Vitesta 1066 memory modules. We are using 4x 2GB memory sticks.
- lower DRAM divisor is not possible as 3000/5 = 600

The results of experiment 5 look like ideal settings if you do not want to OC your CPU but want to take advantage of DDR2 1066. OC of HT is a byproduct. Base frequency is not too high and should not cause instability. Therefore we are going to benchmark this configuration.

Testbench 4:
- Gigabyte GA-MA770-UD3 v2.0
- OCed Athlon 64 X2 6000+ ADX6000IAA6CZ (Windsor, released in 2007) with DDR2 1000 and HT 1.25Ghz.
- Asus GeForce GTX 480 (factory clocks, 701 GPU, 924 memory, 1401 shader, released in 2010). NVidia driver 197.41

3d mark 2006 breakdown, 1024x768, Athlon 64 X2 6000+, GeForce GTX 480:

• Return To Proxycon - 43 fps
• Firefly Forest - 44 fps
• Canyon Flight - 95 fps
• Deep Freeze - 54 fps

3d mark 2006 breakdown, 1600x900, Athlon 64 X2 6000+, GeForce GTX 480:

• Return To Proxycon - 42 fps
• Firefly Forest - 43 fps
• Canyon Flight - 91 fps
• Deep Freeze - 53 fps

3d mark 2006 breakdown, 1600x1200, Athlon 64 X2 6000+, GeForce GTX 480:

• Return To Proxycon - 42 fps
• Firefly Forest - 42 fps
• Canyon Flight - 87 fps
• Deep Freeze - 53 fps

There is a very small 1-3 fps performance benefit observed in synthetic benchmark.

Games tested:
- Crysis (2007) - in 1600x1200 average fps in benchmark is 51, min fps 17 without full screen anti aliasing. Everything else was set to max. 52-80 fps in the initial mission after swiming out of water to beach. About 6 fps gain over DDR 750. When 4x anti aliasing is enabled, we get 49 average fps, 19 min fps. Performance drop is minimal therefore 4x anti aliasing is recommended.

Conclusion about Athlon 64X2 6000+ (Windsor) with GeForce GTX 480 and DDR2 1000:
- noticeable improvement in Crysis in average, min fps and in game
- no instability was observed in the selected configuration

continuation of exploration of various OC settings.

Experiment 6:
- we bump up the base clock to 239 and reduce CPU multiplier to 13, we get 1195Mhz HT. CPU clock is 3107Mhz
- interestingly, the DRAM divisor is now 7 instead of 8 and we get DRAM frequency 444Mhz instead of 375Mhz with standard settings. This corresponds to DDR2 888. In Bios, DDR800 is selected.

Experiment 7:
- we bump up the base clock to 247 and set CPU multiplier to 13, we get 1235Mhz HT. CPU clock is 3210Mhz
- interestingly, the DRAM divisor is now 7 instead of 8 and we get DRAM frequency 458Mhz instead of 375Mhz with standard settings. This corresponds to DDR2 916. In Bios, DDR800 is selected.
- this is a beefed up Athlon 64 X2 6400+ (Windsor) with running memory at DDR916 instead of DDR800 and HT at 1.23Ghz

Testbench 6:
- Gigabyte GA-MA770-UD3 v2.0
- OCed Athlon 64 X2 6000+ ADX6000IAA6CZ to 3.2Ghz (Windsor, released in 2007) with DDR2 916 and HT 1.23Ghz. Stock CPU voltage.
- Asus GeForce GTX 480 (factory clocks, 701 GPU, 924 memory, 1401 shader, released in 2010). NVidia driver 197.41

3d mark 2006 breakdown, 1024x768, Athlon 64 X2 6000+, GeForce GTX 480:

• Return To Proxycon - 45 fps
• Firefly Forest - 46 fps
• Canyon Flight - 100 fps
• Deep Freeze - 57 fps

3d mark 2006 breakdown, 1600x900, Athlon 64 X2 6000+, GeForce GTX 480:

• Return To Proxycon - 44 fps
• Firefly Forest - 45 fps
• Canyon Flight - 95 fps
• Deep Freeze - 56 fps

3d mark 2006 breakdown, 1600x1200, Athlon 64 X2 6000+, GeForce GTX 480:

• Return To Proxycon - 44 fps
• Firefly Forest - 45 fps
• Canyon Flight - 91 fps
• Deep Freeze - 56 fps

The gain is about 3-7 fps over standard Athlon 64 X2 6000+ with DDR 750.

Games tested:
- World in Conflict (2007) - in 1600x1200 we then get 32 average fps in built-in benchmark with the best visual quality settings. 3 fps improvement over standard Athlon 64X2 6000+. Benchmark is quite extreme and real game works better. Playable.
- Crysis (2007) - in 1600x1200 average fps in benchmark is 53, min fps 20 without full screen anti aliasing. Everything else was set to max. When 4x anti aliasing is enabled, we get 50 average fps, 27 min fps. Performance drop is minimal therefore 4x anti aliasing is recommended. Playable.
- Far Cry 2 (2008) - with max settings, in 1600x1200 we get about 35-45 fps in the initial jeep ride with 40 fps most of the time. Not enough for enjoyable experience.
- Need for Speed: Undercover (2008) - in 1600x1200 we get 35-45 fps in the city at the main menu. 100% cpu utilization and lowering resolution does not help. Not enough for enjoyable experience.

Conclusion about Athlon 64X2 6000+ (Windsor) OCed to 3.2Ghz with GeForce GTX 480 and DDR2 916:
- this OC configuration represents a user who bought Athlon 64X2 6000+ instead of Athlon 64X2 6400+ as those may be too expensive and OCed it to slightly exceed 6400+.
- noticeable improvement in Crysis in average, min fps and game. Min fps in Crysis is too random and not useful. We can state Crysis is in the playable range now.
- no instability was observed in the selected configuration
- it is a good choice for Windows XP era games (2002-2006)
- mediocre coverage of Windows Vista era (2007-2009) due to slow CPU, noticable in games from 2008 and probably untested 2009 games (makes little sense to introduce new games if we haven't solved 2008 yet)
- we are unable to improve performance of Need for Speed: Undercover or Far Cry 2 to acceptable levels

Archer57 wrote on 2025-07-05, 10:00:

So just for comparison a few rough benchmarks on those C2D E8600+GTX660 (driver 368.69):

Crysis, 1920x1080 everything on high (very high not possible in xp). Benchmark - 31min, 61avg, 84max. 70-100 in the beginning of new game. Load displayed by MSI Afterburner is GPU - 100%, CPU (Total) - 60-80%. Can not run 1600x1200 because of the monitor...

The above is the sort of performance you can expect from the top AM2 (not plus) build. The CPU is still too slow to cover 2008 comfortably and to take advantage of GTX 480. I'm not going to put in GTX 580 for Windsor as we are still CPU bottlenecked. GTX 580 may help in case of maxxed out full screen anti-aliasing. There may be games with low CPU utilization where it would help such as sports games. It is definitely a usable system given the right expectations about Windows Vista era.

Next steps:
- test stock Athlon 64X2 6400+ (Windsor) and stock Athlon 64X2 6000+ (Brisbane) with GeForce 480 GTX
- the intention is to benchmark the top stock CPUs

AlexZ wrote on 2025-07-05, 12:43:

- test stock Athlon 64X2 6400+ (Windsor) and stock Athlon 64X2 6400+ (Brisbane) with GeForce 480 GTX
- the intention is to benchmark the top stock CPUs

Does 6400+ brisbane exist?

From what i remember 6000+ 3.1Ghz is the best one, making 6400+ windsor the fastest CPU for AM2.

So for brisbane vs windsor comparison 6000+ windsor seems to be more appropriate...

Also yeah, this results may have convinced me not to bother with AM2 for now...

Archer57 wrote on 2025-07-05, 12:58:

AlexZ wrote on 2025-07-05, 12:43:

- test stock Athlon 64X2 6400+ (Windsor) and stock Athlon 64X2 6400+ (Brisbane) with GeForce 480 GTX
- the intention is to benchmark the top stock CPUs

Also yeah, this results may have convinced me not to bother with AM2 for now...

I mean, I kind of get it. This wasn't exactly a great time period for AMD performance wise, but I think it's better to focus on just how wide of a target you can hit with this setup. You get everything from fairly low end single core A64s or Semprons, clear through Phenom. Then again, if we consider AM2+ you can add some really potent Phenom IIs to the mix.

Archer57 wrote on 2025-07-05, 12:58:

Does 6400+ brisbane exist? […]
Show full quote

Does 6400+ brisbane exist?

From what i remember 6000+ 3.1Ghz is the best one, making 6400+ windsor the fastest CPU for AM2.

So for brisbane vs windsor comparison 6000+ windsor seems to be more appropriate...

Also yeah, this results may have convinced me not to bother with AM2 for now...

It was a typo, fixed it. Brisbane is expected to be a slight regression, but if someone OCed it to 3.2Ghz it could match Windsor.

AM2 is basically Windows XP era, no matter what hardware you throw at it. CPUs that came out in 2007 were obsolete next year. AM2+ CPUs may fare better. We will see.

DrAnthony wrote on 2025-07-05, 13:27:

I mean, I kind of get it. This wasn't exactly a great time period for AMD performance wise, but I think it's better to focus on just how wide of a target you can hit with this setup. You get everything from fairly low end single core A64s or Semprons, clear through Phenom. Then again, if we consider AM2+ you can add some really potent Phenom IIs to the mix.

Phenom and Phenom II will be included as part of AM2+ coverage. The expectation is they should be able to cover Windows Vista era.