I think this is as good of a score I will manage to get using an all air cooled 100% year 2006 system with a single (non modified) Geforce 8800GTX only cooled with the Nvidia stock reference cooler! 😀

The system is now GPU limited just as much as CPU limited. Using a Geforce GTX 580 or similar would probably produce a score near 370 FPS.

Non tweaked XP-SP3 (only "security center" and "themes" disabled). All settings set in BIOS, no further HW tweaks. What I think isn't the best driver for Doom 3 by far. No driver tweaks or the like. The ambient temp in my cramped new computer tinkering area was about 25C as I had been tinkering for a while.

340.3 FPS, Skyscraper, Core2Duo X6800 @4.0, Asus P5B Deluxe P965, 2GB DDR2 1000 CL4,4,4,12 2T, 8800 GTX 675/1107 197.45, X-Fi XtremeMusic XP-SP3

Doom3 demo1 1024x768 Ultra. (Doom 3 ver: 1.0)

The list is up to date.

Decided to join in on the fun since I was testing out a socket 939 motherboard anyway. Feels good sticking with AGP, this is more like I remember it (originally played it with a Radeon 9800 Pro).
These are period correct entries. Might try out a newer PCI-E build at a later date.

System specs:
MSI K8N Neo2 Platinum
Athlon FX 55 (San Diego)
2x512MB OCZ Platinum EL
BFG Geforce 6800 Ultra
Audigy 2 ZS

Stock Settings

Overclocked

lost77 wrote:

Decided to join in on the fun since I was testing out a socket 939 motherboard anyway. Feels good sticking with AGP, this is mor […]
Show full quote

Decided to join in on the fun since I was testing out a socket 939 motherboard anyway. Feels good sticking with AGP, this is more like I remember it (originally played it with a Radeon 9800 Pro).
These are period correct entries. Might try out a newer PCI-E build at a later date.

System specs:
MSI K8N Neo2 Platinum
Athlon FX 55 (San Diego)
2x512MB OCZ Platinum EL
BFG Geforce 6800 Ultra
Audigy 2 ZS

Great!

Finally someone posted a score using a FX55 and Doom 3 version 1.0!

That's the motivation I needed to bother rebuilding a socket 939 system to be able replace my FX55 scores made with the slow Doom 3 version 1.3 and try to beat your 109.2 FPS! 😀

Also nice to see another 6800 Ultra card with 425 MHz stock frequency in the list. We have had an entry with an "Ultra+" before but lower in the list. Now I think I will clarify that 425 MHz GPU speed is fine in the stock list as 6800 Ultra cards with those stock clocks did exist late 2004. I know there were later "Ultra+/Extreme" cards with 1200 MHz stock memory speed and even faster GPU speeds but I think those came in Q2 2005 and so far no one has contributed a score with one anyhow. I will do some reading up on exactly what factory overclocked cards did exist late 2004 but a few ones with 425 MHz core were available for sure.

Your overclocked result will probably be hard to beat (My A64 4000+ @3120 score on the overclocked list was made with Doom 3 version 1.0) but I will try. AGP Geforce 6800 cards are faster than the PCI-E ones but the difference is small and I think nForce4 boards have a good chance of reaching better CPU and memory overclocks compared to nForce3 boards. I don't have a Geforce 6800 card able to hit the same clocks as yours but perhaps that can be overcome by pushing CPU and memory.

Your scores have been added to the lists! 😀

Both of my 6800 Ultra AGP cards have a 425Mhz core clock, so didn't think much of it.

I tried to push for a higher HTT but my RAM tops out at about 215Mhz at 2-2-2-5. Maybe at higher voltage but the MSI board tops out at 2.85v.
Dropping timings to something like 2.5-3-3-7 would drop the score quite a bit. I would probably need a HTT of atleast 250Mhz to offset the looser timings.

Good luck overclocking 😁

I know I'm a bit off contemporary specs, but here's a Pentium D 945 and a 1800XT. It's a Dell Optiplex 745. Swapped CPU and smashed in the X1800XT. It fits well enough to work. Adding E6700 soon. 4GB DDR2 800.

Edit: Her's the same with a E6700:

Skyscraper wrote:

Also nice to see another 6800 Ultra card with 425 MHz stock frequency in the list. We have had an entry with an "Ultra+" before but lower in the list. Now I think I will clarify that 425 MHz GPU speed is fine in the stock list as 6800 Ultra cards with those stock clocks did exist late 2004. I know there were later "Ultra+/Extreme" cards with 1200 MHz stock memory speed and even faster GPU speeds but I think those came in Q2 2005 and so far no one has contributed a score with one anyhow. I will do some reading up on exactly what factory overclocked cards did exist late 2004

The EVGA 6800 Ultra Extreme existed in 2004 as per https://hothardware.com/reviews/evga-geforce- … extreme-edition review. The review states that Nvidia announced it in 2004 aswell as a response to the X800XT PE so 6800 Ultra Extremes at clocks of 450 / 1200 should definitely be allowed 😀 plus I will be posting a score up with one at some point in the near future!

Xplo wrote:

The EVGA 6800 Ultra Extreme existed in 2004 as per https://hothardware.com/reviews/evga-geforce- … extreme-edition review. The review states that Nvidia announced it in 2004 aswell as a response to the X800XT PE so 6800 Ultra Extremes at clocks of 450 / 1200 should definitely be allowed 😀 plus I will be posting a score up with one at some point in the near future!

As soon as you post your score with the Geforce 6800 Ultra Extreme I will update the first post with the new allowed "stock clocks" so people who aren't lucky enough to own a real "Ultra Extreme" can pretend! 😉

Most Geforce 6800 Ultra should be able to hit 450/1200 but for people like me who only own a couple of lousy Geforce 6800GT cards it's a challange! 😁

Some time ago I said that I would try to regain the top positions in the period correct lists. I have been a bit buisy but yesterday I set up a test system and today I did some benching. I don't really know where my San Diego CPUs are so I decided to test a DFI motherboard and CPU combo I bought a couple of years ago but never tested. This is not my only DFI nF4 SLI-DR and all together I have at least 5 working DFI nForce4 motherboards but every DFI board has it's own quirks. 😀

The CPU turned out to be a NewCastle 3500+, not very exciting. I saw this as an opportunity to show that there isn't much of a performance difference between the original 0.13um Clawhammer/NewCastle cores and the newer San Diego/Venice.

NewCastle Athlon 64 3500+

Venice Athlon 64 3500+

Half of a percent differnce, within the margin of error.

Then I did some overclocking with the Venice A64 3200+ as I know from earlier testing that this chip can do 3+ GHz. I have owned this chip since "back in the day" and truth be told I don't think I would find a better Venice even if I started buying A64 CPU lots...

Overclocking was still a bit of a challange as I used the less than optimal Zalman CNPS7700-Cu cooler included with the the DFI-motherboard bundle, too lazy to find anything better. I could not use more than 1.55V or the CPU would get too hot! 😜 The HDD used is a very period correct 250GB drive, loading XP after every BIOS change took ages! 🤣

I'm pretty happy with the results. Here is how the benching went. Not bad for a CPU with only half the cache compared to a San Diego.

DFI nForce4 SLI DR

Doom3 v1.0 1024 Ultra

A64 3500+ (0.13 um 512kb) 2x1GB DDR400 3-4-4-8 1T Geforce 6800GT stock = 89.2 FPS

A64 3500+ (0.13 um 512kb) 2x1GB DDR400 3-4-4-8 1T Geforce 6800GT @Ultra = 92.4 FPS

A64 3500+ (0.13 um 512kb) 2x1GB DDR400 3-4-4-8 1T Geforce 6800GT @Ultra+ = 93.1 FPS

A64 3500+ (0.13 um 512kb) 2x1GB DDR400 2.5-3-3-7 1T Geforce 6800GT @Ultra+ = 96.5 FPS

A64 3500+ (0.13 um 512kb) 2x512mB DDR400 2-2-2-5 1T Geforce 6800GT @Ultra+ = 99.7 FPS

A64 3500+ (0.09 um 512kb) 2x512mB DDR400 2-2-2-5 1T Geforce 6800GT @Ultra+ = 100.2 FPS

A64 3500+ (0.09 um 512kb) @3000 2x512mB DDR400 2-2-2-5 1T Geforce 6800GT @Ultra+ = 112.2 FPS

A64 3500+ (0.09 um 512kb) @3000 2x512mB DDR400 2-2-2-5 1T Geforce 6800GT @440/1130 = 114.3 FPS

A64 3500+ (0.09 um 512kb) @3100 2x512mB DDR414 2-2-2-5 1T Geforce 6800GT @440/1130 = 116.2 FPS

A64 3500+ (0.09 um 512kb) @3200 2x512mB DDR427 2-2-2-5 1T Geforce 6800GT @440/1130 = 117.8 FPS

A64 3500+ (0.09 um 512kb) @3245 2x512mB DDR433 2-2-2-5 1T Geforce 6800GT @440/1130 = 118.4 FPS

A64 3500+ (0.09 um 512kb) @3245 2x512mB DDR433 2-2-2-5 1T Geforce 6800GT @450/1150 = 119.6 FPS

A64 3500+ (0.09 um 512kb) @3245 2x512mB DDR433 2-2-2-5 1T Geforce 6800GT @450/1160 = 119.8 FPS

A64 3500+ (0.09 um 512kb) @3245 2x512mB DDR433 2-2-2-5 1T Geforce 6800GT @460/1160 = 120.8 FPS

A64 3500+ (0.09 um 512kb) @3245 2x512mB DDR433 2-2-2-5 1T Geforce 6800GT @460/1175 = 121.2 FPS

A64 3500+ (0.09 um 512kb) @3289 2x512mB DDR438 2-2-2-5 1T Geforce 6800GT @460/1190 = 121.9 FPS

A64 3500+ (0.09 um 512kb) @3300 2x512mB DDR440 2-2-2-5 1T Geforce 6800GT @460/1190 = 122.4 FPS

A64 3500+ (0.09 um 512kb) @3300 2x512mB DDR440 2-2-2-5 1T Geforce 6800GT @460/1200 = 122.7 FPS

A64 3500+ @3300 (Venice 512KB), DFI nF4 SLI-DR, 2x512MB DDR 440 2-2-2-5 1T, GF6800GT (460/1200) 66.93, Audigy 2, XP-SP3 (No driver tweaks or the like)

1024*768 Ultra 122.7 FPS

Now I just need to remember where I put the FX55...

The list is more or less up to date.

I found the San Diego FX55! 😀 The Clawhammer FX55 is still missing, it's probably mounted on a board in some random motherboard box in a stack of random motherboard boxes in storage.

First I did a run where I fully mimicked the settings member lost77 used for his stock settings period correct record run a few posts above.

Athlon 64 FX55 (San Diego), DFI nF4 SLI-DR, 2x512MB DDR 400 2-2-2-5 1T, "Geforce 6800 Ultra+" 66.93, Audigy 2, XP-SP3 (No driver tweaks or the like)

1024*768 Ultra 111.2 FPS

It seems my DFI nForce4 motherboard is a tiny bit faster than lost77s MSI nForce3 one. The fact that the default (or manual for that matter) 200 MHz referece clock setting on this board results in an actual clock of 200.9 MHz helps the score but that is something most nForce Athlon64 boards seem to have in common. Member lost77s MSI board also sets the reference clock to 200.9 MHz. 😀

The memory I'm using is Corsair CMX512-3500c2 rated for 434 MHz with 2-2-2-5 timings so naturally the next step was to up the memory clock.

Athlon 64 FX55 (San Diego), DFI nF4 SLI-DR, 2x512MB DDR 434 2-2-2-5 1T, "Geforce 6800 Ultra+" 66.93, Audigy 2, XP-SP3 (No driver tweaks or the like)

1024*768 Ultra 112.6 FPS

Every FPS counts! 😜 This memory divider isn't supported with the original Clawhammer core FX55 so this is almost cheating. The memory kit is from early 2004 though and there are other ways to increase A64 memory speed.

Perhaps I also did a run using the Geforce 6800 Ultra Exreme stock clocks as this Geforce 6800 GT card actually can do it now after I found a Thermalright HR-03 in one of my random junk boxes. 😎

I will withhold that score for now so member "Xplo" has a chance to post his score with a "real" Geforce 6800 Ultra Extreme first.

I just got reminded how bad of an overclocker my San Diego FX55 is, even my Clawhammer FX55 is better. There isn't much of a point in even trying with this super hot running chip. The temperature issue is why the voltage shown in the screenshots for the FX55 is low as the CPU was running a bit too hot for my liking at the default 1.45V even at stock speed. Perhaps it's bad contact between the IHS and the die and a delid would help or perhaps the chip just is shit.

While we (well mostly me I guess 😜 ) wait for member Xplo to post his score with the Geforce 6800 Ultra Extreme I found that going from PC3500 memory to PC3700 memory gained another 0.9 FPS. This is sadly the highest memory divider availible at the 13x stock multiplier for the FX55 on this motherboard with this BIOS. At 10x, 11x and 12x the maximum divider avalible works out to a slightly higher memory speed up to a maximum of (10/8)*200 = 250MHz = DDR500 = PC4000. In case of the FX55 and it's 13x multiplier (13/10)*200 = 260MHz/DDR520 should be avalible but isn't nor is (12/9)*200 = 266Mhz/DDR533 with the 12x multiplier.

Athlon 64 FX55 (San Diego), DFI nF4 SLI-DR, 2x512MB DDR 475 2-2-2-5 1T, "Geforce 6800 Ultra+" 66.93, Audigy 2, XP-SP3 (No driver tweaks or the like)

1024*768 Ultra 113.5 FPS

So I’m coming up a little short at the moment for some reason. Specs are

FX57 @ FX55
2x512mb Corsair Xpert TCCD @ 2-2-2-5 (200mhz)
DFI Lanparty NF3 Ultra-D
6800 Ultra Extreme AGP

Best run was 104.xx fps. I’m putting it down to an old hdd that i had laying around with XP already installed, only did a quick driver wipe and re-installed drivers so I could get some runs in. Going to do a fresh install of XP and that should sort it hopefully.

Failing that, I will roll out the bigger guns

FX57 @ FX55
2x512mb Geil ONE BH-5 @ 1.5-2-2-5 (200mhz)
DFI Lanparty NF4 Expert
6800 Ultra PCI-E @ Ultra Extreme

I also have a 3.46ghz Gallatin, that will go for a spin too

So Skyscraper you’ve made some benching runs with your memory overclocked/async. Does this qualify for the stock settings run (if) the memory is stocked overclocked i.e pc4000 etc?

May take me some days to try again. I have a 6 month old and a demanding missus.

Xplo wrote:

So I’m coming up a little short at the moment for some reason. Specs are […]
Show full quote

So I’m coming up a little short at the moment for some reason. Specs are

FX57 @ FX55
2x512mb Corsair Xpert TCCD @ 2-2-2-5 (200mhz)
DFI Lanparty NF3 Ultra-D+
6800 Ultra Extreme AGP

Best run was 104.xx fps.

...
...

May take me some days to try again. I have a 6 month old and a demanding missus.

That is pretty much spot on for what I would expect the score to be with your hardware if using a Doom 3 vesion other than v1.0. Are you sure you are not using the slow crappy v1.3 one (Steam version) ?

If you don't have access to a Doom 3 disc with v1.0 and can't find one send me a PM.

Xplo wrote:

So Skyscraper you’ve made some benching runs with your memory overclocked/async. Does this qualify for the stock settings run (if) the memory is stocked overclocked i.e pc4000 etc?

It dosn't really matter if your memory is advertised as beeing capable of running at PC4000 or not. Such kits were avalible at the time even if people with the original Clawhammer core A64 had to juggle a few more settings to make use of them. There is no rule against running the memory using a divider.

Very nice getting that Venice to 3.3Ghz, didn't think they went that high with regular cooling.

The memory I'm using is Corsair CMX512-3500c2 rated for 434 MHz with 2-2-2-5 timings so naturally the next step was to up the memory clock.

As I recall Corsair just stated CL2 for this RAM at 343Mhz, supported timings being something like 2-3-3-?.I am not entirely sure, it has been a few years 😉

lost77 wrote:

Very nice getting that Venice to 3.3Ghz, didn't think they went that high with regular cooling.

The memory I'm using is Corsair CMX512-3500c2 rated for 434 MHz with 2-2-2-5 timings so naturally the next step was to up the memory clock.

As I recall Corsair just stated CL2 for this RAM at 343Mhz, supported timings being something like 2-3-3-?.I am not entirely sure, it has been a few years 😉

My Venice 3500+ is probably a 1 in a 100 if not 1000 chip. 😀 With a delid and a simple water cooling loop it would probably do 3500 MHz bench stable.

I actually not 100% sure what Corsair promised for this exact kit oher than CL2 but the SPD timings are indeed 2-3-3. Memory kits with Winbond BH5 memory should be running at 2-2-2 timings though, anything else would be a total waste! 😀

Perhaps its best I try to explain why the very liberal rules are like they are.

There is no rule for the stock period correct list that the memory should be running at 200MHz/400DDR or less and we have had entries since the start with memory running faster. The issue with Clawhammers not supporting faster memory than 400 MHz DDR at stock might have warrented a ban on memory dividers greater than 1:1 on the AMD side but it's a bit late for that now. The performance difference between PC3200 1.5-2-2 and PC3700 2-2-2 is probably between 1 and 2 FPS at the top end so the difference isn't huge. In the overclocked list the Clawhammer memory divider limitation dosn't matter but the ~200 MHz difference in top overclock compared to San Diego undeniably does.

On the Intel side it's possible to use any DDR2 kit along with Prescott steppings not available in 2004 and it's all just about convenience. To keep track of which core steppings and memory modules are kosher/halal and which are not is alot harder than just keeping track of video card generations, motherboard chipsets and CPU models. If someone posts a score with a CPU model that existed in 2004 I don't want to tell him that his CPU isn't allowed because of the core stepping and so on. There is also the fact that it would severely decrease the number of entries in the period correct list.

If we would have been really hardcore we would have only accepted period correct results with Windows XP up to SP2 as SP3 (which for unknown reasons adds about 1% performance to Doom 3) was released in 2008. As I said ealier in the thread when just these exact issues came up for discussion the extra performance gained from newer core steppings and such in the overclocked list offsets the fact that we aren't using Vapochill or Prometeia phase change coolers which many performance minded people actually did back in 2004.

When it comes to "factory overclocked" video cards they are allowed in the stock non overclocked list and as soon as a score has been posted with one it's also allowed to mimic those clocks with a non overclocked model or even lesser model as long as the spec otherwise is the same except for stock clockspeeds. The same goes here, it wouldn't be very nice to tell someone that his video card isn't allowed because it dosn't follow Nvidias or ATIs guidelines for the model. Those who bought such a card "back in the day" for sure diddn't lower the clocks just to follow the spec.

There are no rules against water cooling, setting the AC to max or simply opening a window letting a bit of cold air into the room either but moving your benchmark rig outside in a cold climate or using a 1000+W water chiller hidden under a table might be a bit of a stretch.

I can see how it would be hard to keep track of memory kits, there were a lot of different ones. It will be interesting to see what results faster memory speeds can get us.

In terms of memory and CPU I just kept settings at auto in the bios for my stock result. I can see now that left me with a 800Mhz HT instead of 1000, but the correct memory timings got recognized.

BTW I found a review of the CMX512-3500C2 that states 2-3-3-6 for 434Mhz. Guess my memory isn't as bad as I thought 😀

I might still have some of those Corsair sticks in the basement to try out, and maybe some other high speed kits.

I found a few kits when looking for parts for my benching. I do own alot of legacy hardware but saying that it isn't very well organized is an understatement.

Some of of my performance kits are 2x256MB ones so useless for benching Doom 3 at 1024 Ultra (needs 750 - 800 MB or it will be a stutter fest). Others are Promos kits and also useless. A couple are Samsung TCCD (one kit pictured XMS3202v4.1) and probably usable for trying to get good scores for the overclocked list. If I remember right the A-Data DDR500 kit uses Hynix chips and won't do anything over it's rated frequency. I think the Kingston KHX3700K2 kit uses some crap chips.

In fact the only decent kit pictured except for the Samsung TCCD one is the CMX512-3200LL v1.2 kit. It has Winbond CH5 chips so it probably won't run 2-2-2 timings at speeds much greater than 200 MHz but should be able to scale 2-3-2 with voltage. It's not hard to see why the CMX512-3500C2 v1.1 kit ended up in the test rig. That kit will probably do 250+ MHz 2-2-2-5 at 3.5V but as it's the only 2x512MB BH5 kit I was able to find in my DDR memory box I won't take any risks with it and will stay at 3.3V.

A bigger issue for me is CPUs.

I have a single Athlon 64 San Diego capable of speeds greater than 3GHz and I have no idea where it is. All I can find is a 3700+ capable of 29xx MHz, the crappy San Diego FX55, the fantastic Venice, two untested Venice and about 10 older 0.13um CPUs. My Clawhammer FX55 can at least be pushed to ~3 GHz but it's on a board in a box somewhere just like my decent San Diego 4000+.

Still can’t figure how you guys are getting much more FPS than me in the stock period correct list.

FX57 @ FX55
Lanparty NF4 Expert
BH-5 @ 200mhz 2-2-2-5
6800 Ultra @ 450/1200
Fresh install of XP SP3, all updated drivers

No fancy OS tweaks, optimized defaults in BIOS, best return I’ve had is 105.4 fps

Lost77 has got 5ps more than me, with 6800 clocks of 425/1100. Go figure!

Tell us what else you two are tweaking behind the scenes so it’s a level playing field 😉

Oh and I’ve got an Audigy 2 ZS chucked in there to reduce the CPU overhead, and I installed using original CD’s so I’m on version 1.0.something

Lol. Something not right here.

108.6 fps

FX57 @ 3100mhz (310 x 10)
1gb TCCD @ 1:1 (310mhz @ 2.5-4-4-8 1T)
Lanparty NF4 Expert
6800 Ultra @ 450/1200
Audigy 2 ZS
Forceware 66.91
Latest DFI BIOS
Full clean install XP SP3
All drivers updated

Heavily overclocked, yet im still 0.5 fps short of Lost77’s *STOCK* period correct system with an FX55 @ 2.6, slower ram and a slower clocked 6800U. Hmm 😒

My benching system runs a fully updated Windows XP SP3 Professional. Some OS tweaks but I wouldn't call them fancy and their impact on the performance is slim to none.

No Security Center, no Themes, no driver update softwares/services running in the background or other such bloat and the virtual memory is disabled. I have never seen any of this making a difference on the FPS at all. Disabling virtual memory makes sure I notice if I start running out of memory when benchmarking and disabling bloat makes it less likely to happen. I do not touch the video driver settings at all but I do disable Nvidias driver helper service as it's wasting memory, it has no impact on performance. The system is not connected to the network while benchmarking, especially USB Wifi dongles will have an impact on performance. I do sometimes disable stuff I don't use in the BIOS but after a BIOS reset I often forget and it dosn't seem to affect performance.

For me when using an Athlon 64 system it dosn't matter if I run the benchmark right after a reboot or after I do a 3dmark 2001 run or similar, the result is about the same. When running the benchmark the first run is fast, if I do a second run it's at least 1 FPS slower but the third run is about as fast as the first run, after that the runs stay consistant. If I close Doom 3 and open it again the second run will again be slower than the first, third, forth and so on.

With a really fast system like an overclocked Core 2 Duo or faster you want to run the benchmark directly after a reboot and the first run will be the fastest.

Lost77s results are spot on for his hardware compared to other both overclocked and non overclocked results in the lists. I switched to the driver he was using but I'm almost sure driver 91.31 that member havli used for his overclocked San Diego run is a tiny bit faster. Havli got 116.2 FPS at only 2.8 GHz with DDR466 and very loose memory timings using a Geforce 6800GT at 425/1100 and I think that his result probably still is the best considering the hardware settings. Back when he first posted that score we others had not yet realized how much of a performance impact the Doom 3 version makes and were still using the slow v1.3 one from Steam but after some head-scratching we figured it out. I used to use driver 92.91 only because that's the driver I first used and I wanted to be able to compare results between different hardware platforms I tested. Many others ended up using that driver for the benchmark as it's a popular one.

By now I can tell directly if a result is off and your results are 10% lower than they should be running Windows XP SP3 and Doom 3 version 1.0. I don't know what the issue is if it isn't the Doom 3 version.

Just the fact that you own TCCD memory capable of 310 MHz makes me confident you will figure it out what ever the issue is! 😀

For reference

You should come close to this score at 2.6 GHz and probably even beat it if you run your BH5 memory faster using a divider while still maintaining 2-2-2 timings.

Optimized defaults. Memory timings adjusted as my kit defaults to 2-3-3.

Athlon 64 FX55 (San Diego), DFI nF4 SLI-DR, 2x512MB DDR400 2-2-2-5 1T, "Geforce 6800 Ultra Extreme" (450/1200) 66.93, Audigy 2, XP-SP3 (No driver tweaks or the like)

1024*768 Ultra 114.8 FPS

A few posts up on this page I compared a stock Newcastle A64 3500+ (0.13um 512KB 2200Mhz ) with the newer Venice A64 3500+ (0.09um 512KB 2200MHz) with the Venice winning out achieving an extra half of a FPS in the Doom 3 benchmark (probably by coincidence). Today I found my Clawhammer FX55 so now we can compare the Clawhammer core against the newer San Diego core too see if there is any difference. The Zalman heat sink seems to have serious issues with FX55 CPUs regardless of the core version so when testing the Clawhammer I again had to lower the voltage to get acceptable temps.

Athlon 64 FX55 (Clawhammer), DFI nF4 SLI-DR, 2x512MB DDR 400 2-2-2-5 1T, "Geforce 6800 Ultra+" 66.93, Audigy 2, XP-SP3 (No driver tweaks or the like)

Doom 3 1024*768 Ultra: 111.1 FPS

Athlon 64 FX55 (San Diego), DFI nF4 SLI-DR, 2x512MB DDR 400 2-2-2-5 1T, "Geforce 6800 Ultra+" 66.93, Audigy 2, XP-SP3 (No driver tweaks or the like)

Doom 3 1024*768 Ultra: 111.2 FPS

The answer to the question is: No there is no performance difference between a Clawhammer and a San Diego in Doom 3 at all.

Yesterday evening I really REALLY tried to find my good Athlon 64 4000+ CPU. I thought it was mounted on the DFI nF4 Ultra-D motherboard I used for benching back in May 2015 but I checked every s939 motherboard and every loose K8 looking AMD CPU I could find as I was also on the lookout for other CPUs that might overclock decently. I searched through EVERYTHING and it took many many hours but I diddn't find the CPU nor any other good San Diegos only a truckload of AMD X2 3800+ CPUs, a few poor Venice CPUs and a San Diego Opteron 146 that ended up topping out at 29xx MHz using the crappy Zalman heat sink.

While searching yesterday I did stumble upon an Asus M2N32-SLI Deluxe nForce 590 motherboard, awesome I thought this will be great for benching AM2 for the "somewhat period correct" list. Today after work I dismantled the socket 939 system and went to find the M2N32-SLI so I could benchmark an A64 X2 6400+, a CPU not yet represented in the Doom 3 benchmark results. It turned out that now I couldn't find this board either but instead I found a DFI nF3 AGP motherboard in a box I missed yesterday... on that board sat a San Diego A64 4000+...

I could have moved the good A64 4000+ to this nF3 board to test the motherboard after I bought it at an unknown date from an unknown source or it's another San Diego 4000+... I guess we will find out. Not finding the Asus M2N32-SLI nForce 590 board eventhough I saw it yesterday is a bit irritating but I do know exactly where my M2N(non32)-SLI Deluxe nForce 570 board is so I guess I will use that motherboard for AM2 benching instead. It's not like the Geforce 7900 GTX will be bottlenecked much by 8x PCI-E and I think the CPU is too weak for two 7950GX2 cards to make sense. It will be more fair this way as candle_86 used the M2N-SLI when he made his top score for the list.

Athlon 64 San Diego 4000+ and DFI nF3

As I did a rebuild of the DFI nF4-SLI DR bench setup I took the opportunity to upgrade the cooling a bit. The best cooler that can easily fit Socket 939 is Thermalright Ultra-120 eXtreme. I do not have the appropriate backplate or other mounting hardware for s939 but combining the "bar" part from the older version of the Socket 775 mounting kit with Zalmans mounting kit for their CNPS series of heat sinks worked better than expected.

[Edit][Rant]

No this A64 4000+ CPU turned out to be rather poor, it won't even do 3000 MHz. I guess the princess is in another castle.

Whats even worse is that the Thermalright Ultra-120 eXtreme performs poorly and I have no idea why. I have tried two different identical heat sinks with the same result. I have tried different mounting pressures, I have checked that the heat sink makes good contact by checking the thermal paste imprint and everything looks good. I get the same if not worse tempratures when I compare with the Zalman CNPS 7700 Cu and that dosn't make any sense at all.

I just tested the Ultra-120 with the good Venice 3500+. The CPU would do 3300 MHz at 1.55V with a couple of extra fans blowing air at the crappy Zalman heat sink but it will only do 3200 with the Ultra-120. The ambient temprature was ~18C when I did the 3300 MHz run and it's ~23C now but the difference in cooling capacity should easily make up for that.

I guess it's possible that the thermal paste AMD used between the core and the heat spreader really has started to degrade now after almost 15 years and it's acting like a serious cooling "bottleneck". I don't remember having any issues with keeping Athlon 64 CPUs cool before but now I see 45C in the BIOS hardware monitor at stock and ~55C @1.55V when overclocked and thats without any load on the CPU.

[/Edit][/Rant]

[Edit2]

I switched back to the Zalman CNPS 7700 Cu and suddenly the San Diego 4000+ started working a bit more like you would expect a decent stepping stepping San Diego would... The temperature sensor is reading at least 5 degrees lower and stability is greatly improved.

I guess the Ultra 120 with it's convex base simply puts too much pressure at the center of a s939 K8 making it perform poorly somehow. I could whip out a water cooling loop but if I can match the 3120 MHz I did when I benchmarked this CPU (or another like it) back in 2015 I won't bother as it should be enough.

[/Edit2]

[Edit3]

Now I'm pretty sure the A64 4000+ I found isn't my "good" 4000+ as this CPU wouldn't do much more than 3000 MHz unless I really pushed voltage and it still wouldn't reach the 3120 MHz the "good" 4000+ previously did at only 1.475V.

The memory wouldn't go over 255 MHz 2-3-2-7 (the rest on auto) with the memory voltage set to 3.2V in the BIOS (reading ~3.1V if one would believe the onboard sensor) so I settled for 12x255. I'm kind of a coward when it comes to memory as it's much much harder to find good memory compared to CPUs. I run with both sets of memory voltage jumpers in the default "safe" position so the memory gets its power from the 3.3V rail and I guess with the voltage limiting jumpers set to safe mode all you get is <3.1V.

A64 4000+ @3060 (San Diego), DFI nF4 SLI-DR, 2x512MB DDR 510 2-3-2-7 1T, GF6800GT (460/1200) 66.93, Audigy 2, XP-SP3 (No driver tweaks or the like)

1024*768 Ultra 124.5 FPS

[/Edit3]