you can get special aux psus. a regular psu needs to be connected to a mobo and have a certain min load on the rails. The aux units usually fit in a drive bay.

why not just buy a single good psu

Davros wrote:

why not just buy a single good psu

Shouldn't take much to run an old athlon system, what pwer supply are you running now?

bushwack wrote:

Shouldn't take much to run an old athlon system

The problem we run into on our motherboards is that the power comes off the 5v rail instead of the 12v, like P4 and after. So getting a beefy processor going will take 30+ amps. Most modern supplies will only give you 20 or so, unless they are pretty large, 700+ watts. This isn't necessarily wrong either, your literally dealing with 2 different ATX specification revisions.

It's an interesting problem. It's only now that I realise and appreciate (from reading Prophase_j's postings) why the Intel Pentium 4 boards adopted that new 12v CPU dedicated power plug. I reckon they must have seen the problem coming by simply looking at the very last generation of old Athlon boards: the faster the Athlon CPU you put in them, the more the 5v rail on a typical PSU is pushed to its limit. And although these mobos boot up, they don't seem rock solid. (To solve this, why not give the P4 CPU their own 12v power source.)

The following system appears to be more or less OK, I think:

PSU: 350W Nexus; 3.3v=28A, 5v=30A (220w combined), 12v=18A
(Removed Corsair 450W; 3.3v=20A, 5v=20A [just 130w combined!], etc)
CPU: t-bred 2000+ rated
(Removed t-bred 2400+ rated)
FSB: 133
Gfx: FX5950 (71.84 drivers)
RAM: 1 stick of 256mb PC133 (Infineon)

However, just by swapping out the t-bred 2000+ for a t-bred 2400+ results in system instability when running the "strenuous" 3DMark2001 first edition.

The problem today is that you can't buy new PSUs with a really meaty 5v rail. New PSUs are all geared up for dedicated 12v CPU power and dedicated 12v power for graphics.

I've got a nutty idea! I can't afford a dedicated supplemental PSU ATM, but would this work instead:

Get 2 "normal" PSUs, and plug them both in to 2 mobos. 1 PSU powers a "dummy" mobo which is not used for anything, the other PSU powers the mobo you are actually working on. The "dummy" mobo is simply any old mobo, just to get some "load" on to the "dummy PSU", simply to get it to work. The idea is that you "borrow" most of this PSU's power by plugging in one of its molex power cables in to your working graphics card - which is plugged in to the *other* mobo - the mobo you are actually working with.

Yes, this is possible, but there are more elegant ways to switch on the "auxiliary" PSU.

ATX PSUs will turn on when you connect the green (PWR_ON) wire to any of the black ground wires. The most simple solution: a bent paper clip. like this:

If you want both PSUs to turn on when you press the power button connected to the mainboard, splice the green wire of the "auxiliary" PSU into the green wire of the "main" PSU.

Thanks a lot 5u3! 😀 I must just double-check something: I'm looking at my PSU power plug, and I can see that green wire. I'm looking at this green wire, as if it's on the "top row", as in your photo. Immediately to its left is a single black wire. Immediately underneath that single black wire, on the "bottom row", is a pair of black wires going in to the same "socket". One black wire is slightly thicker than the other. Shall I connect the paperclip from the one and only green wire, to the "bottom row" "double pair" of black wires socket? Thanks.

Doesn't matter. Any black wire will do, they are connected inside the PSU.

Thanks a lot, this works. I tried a simple set up to begin with: the "aux" PSU is simply providing power for the DVD-ROM drive. That's it. But it works, because I tested it in Windows 98. Now, I'm going to get the "aux" PSU to power a meaty graphics card. In fact, I'm going to try an nVidia 6800GT. The "aux" PSU is a Corsair VX 450W. It has mediocre 3.3 & 5 volt rails (20A each), but it seems to have 33A on its 12v rail. I reckon that ought to power the 6800GT, and if it works, I might be able to get one of the Abit KT7A boards I'm testing to run with either a 2000+ or 2400+ rated CPU, without stability issues.

So far, so good. I've got the "aux" PSU providing power for just the nVidia 6800GT graphics card. Everything else is being powered by the other "primary" PSU. I ran 3DMark2001 first edition 4 times, and get these scores:

10025, 10132, 10131, 10146 (Consistent values)

I max out all of the 3DMark2001 f.e. settings (incl. 1280x1024), and rerun it 3 times:

8418, 8390, 8412 (Consistent values)

(That's using the 1.2 revision board, with a t-bred 2000+ rated CPU, @133fsb)

I reckon that's a success. The next thing to try is the 1.3 board with a t-bred 2400+ rated CPU, at 133fsb. That's the configuration I never got working properly with using just 1 "fairly average" PSU. Actually, before I try that, while the 1.2 board is up and running, I'll swap out the 2000+ rated t-bred, and put in a 2400+ rated t-bred, run it @100fsb, and rerun 3DMark2001 f.e. a couple of times...

Edit: That worked - scores are 8993, 9075, 9102, 9124 (Consistent values)
3DMark2001 f.e. maxed out - scores are: Oh that's a pity, it was running OK then it dropped back to the desktop.

It's interesting to note that using 2 PSUs makes less noise than using 1 PSU. With the 1 PSU set up, there's more of a "strain sound" on the PSU - that is, when 3DMark2001 is running.

Nice! Now, get a decent PSU... 😉

The values printed onto PSU stickers often are greatly exaggerated. They seem to denote the wattage you can pull off the unit for a few seconds without it bursting into flames, but that says nothing about the voltage stability on the rails.

Getting a decent PSU is now my top priority! 😀

Following on from the test set up above, I decided to alter the PSUs configuration before rerunning the tests. I did this:

On the mobo, I removed 2 of the 3 chassis/CPU fans.
I removed the HDD power cable from the "mobo PSU"
I removed the DVD-ROM power cable from the "mobo PSU"

Then, I attached 2 case fans to the "graphics PSU", and also attached the HDD & DVD-ROM reader to it as well. (Basically, I'm trying to remove as much strain from the "mobo PSU" as possible. I did think about using a 3rd PSU, just to power the HDD, DVD and fans, but I couldn't be bothered to set that up.)

I rerun 3DMark2001 f.e. and get 8993. I can't be bothered to do another 3 of these tests, so I just max out 3dMark2001 and run that 4 times instead. I get: 7646, 7676, 7664, 7658, (very consistent values!)

I reckon that's a success! Please notice the ~10% loss of speed from the faster 2000+ *133fsb* set up. My final experiment will be to set up the 1.3 revision board, with the 2400+ t-bred, and run it at 133fsb...

Edit: I tried to set the fsb to 133 using the t-bred 2400+. No POST. I set it back to 100fsb, but added 10mhz to the fsb instead. That worked. So, the real clock speed is now 2200mhz. I get 9803 in 3DMark2001 f.e. That's still not as fast as a 2000+ t-bred @133fsb. It looks like the 2000+ t-bred @133fsb (1666mhz) is the winner on this 1.2 revision board.

Edit 2: I tried an XP-M mobile barton 2200+ rated (1666mhz) CPU, and ran it @133fsb. I get the best results yet - it beats the t-bred 2000+ rated (1666mhz) CPU @ 133fsb. I guess it is faster, because of the "double L1 cache" on the bartons; 512kb as opposed to 256kb.

3DMark2001 f.e. = 10823, 10897
3DMark2001 f.e. (maxed out) = 9009, 9001

Edit 3: I gave the FSB a small boost inside the BIOS, and increased it by 5mhz. The FSB is now 138mhz. It seems to work OK. I also gave the CPU core voltage a small increase to 1.40v (from 1.35v). I'm not sure if it needed to be increased - I just guessed that it needed to be. 3DMark2001 f.e. (maxed out) now scores 9187

Final thoughts: the xp-m mobile barton 2200+ rated (1666mhz) @ 133fsb is the new winner, whether you overclock it or not.

Diagram of current PSUs configuration -

I'd like to say a big 5 amperage thank you to prophase_j, who correctly advised me (on more than one occasion) that the primary problem was a lack of power. Now that I've got 2 PSUs feeding power to the Abit board (I'm on the 1.3 revision ATM, but this also applies to the 1.2 board), the mobo + CPU are not under too much strain, and consequently all of the 3DMark testing has succeeded.

1.3 board
2400+ t-bred
133 fsb
2000mhz real clock speed
nVidia 6800GT (77.72 drivers)
3DMark2001 f.e.

Default settings = 10608, 10734, 10731
Max settings = 8898, 8894

Edit: This is a bit odd - the mobile barton 2200+ CPU won't POST correctly in the 1.3 board. If you set up the CPU correctly in the BIOS, and also enable AGP 4x, it won't POST. Oh well.

Edit 2: I'm still puzzled and surprised that the ATI Radeon 9800 Pro (and also the 7500) won't work correctly if the CPU rated speed is greater than 2000+. I just tried the 9800 with a 2400+ CPU, and the driver doesn't "engage" - I just see the "basic VGA" screen when the desktop appears. On other tests, the driver does "engage", but the desktop is corrupted and unstable. As soon as I remove the fast 2400+ rated CPU and replace it with a slower 2000+ rated CPU, the problem vanishes.

Also, using a "slower" 2000+ rated CPU, I tried 3DMark2001 f.e., with all settings maxed out (including 6x AA), and it froze during testing. I am still using the "aux" PSU, so it could just be the buggy ATI 6.2 driver. But it's strange that the "aux" PSU doesn't make much of a hissing sound, to indicate that it's really powering the ATI card. When I tested the nVidia 6800GT, the "aux" PSU was really making a "straining" sound, as if it was doing all it could to power the thing. I wonder if the ATI card uses less of the "aux" PSU 12v juice, and is instead using more of the "mobo PSU" 5v juice? (That might explain why the aux PSU isn't making much noise, and it froze during testing!)

You have tried a few different XP 2400+ ?

ux-3 wrote:

You have tried a few different XP 2400+ ?

Amazing! I removed my current 2400+ CPU, and replaced it with a back up 2400+, and that completely fixed the problem! This replacement CPU is running at "top speed": 2000mhz @133fsb, and the Radeon is happy with it, running at 4x AGP with no corruption on the desktop.

On the first "broken" CPU, do you think there's some kind of crazy bug inside it? The thing is, it seems to work just fine with other graphics cards, eg nVidia 6800GT, and FX5950.

BTW, both 2400+ CPUs are green and have the same CPU ID code of AXDA2400DKV3C. The "broken" CPU has a stepping code of AIUCB, whereas the "working" CPU has a stepping code of AIUHB.

Thanks a lot for your suggestion - to try a replacement! 😀

I think at 30 amps your right on the threshold. The closer you get to the max output, the more the power level will fluctuate, and this will translate through the voltage regulation circuit to the processor. I think one of your processors is less sensitive than the other. I'm willing to bet that if you had a little more juice, the suspect one would level out. If you haven't already purchased a new power supply, I would get at least 35 amps if you want to continue playing around with high end Athons XPs. I think this is even more important if you start testing motherboards with different chipsets, since they will use even more power by having things like faster RAM and FSB speeds.

Ideally, I would like to buy a PSU from about 5 years ago, but which is "new old stock". An Antec or Enermax - one of the greats, with excellent 3.3 & 5 volt rails. I had a quick browse online at new and modern PSUs and found this one -

http://www.kustompcs.co.uk/acatalog/info_1710.html

It has 40A on both 3 and 5 rails, but it's incredibly expensive.