The Serpent Rider wrote:3200+ usually can be undervolted. 1.5v (default 1.65v) will noticeably decrease the stress to VRM/PSU.
I thought this was a pretty good idea so gave it a go.
Test configuration:
-fresh install of windows xp pro with only drivers and 3d mark 01 & aquamark 3 installed
-mobo out of case on motherboard box
-all bios voltages initially set to default
-3200+ installed
-2Gb ram consisting of 1gb & 2 x 512mb sticks
-6600gt
-2.5inch sata hdd connected to IDE controller through an adapter ( I had heaps of trouble with the onboard nforce2 sata )
-PSU: Corsair CX430M (all I have spare thats new/modern)
Test case - lower vcore from 1.65V to 1.5V
result: immediate hard lock on aquamark.
Test case: vcore @ 1.5V change GPU to GF4 MX440L - low profile & passive.
result: immediate hard lock on aquamark.
Test case: vcore returned to default @ 1.65V GPU GF4 MX440L
result: aquamark run for about half the benchmark before locking - reboot - windows hung @ loading screen
Test case: vcore @ 1.65V GPU GF4 MX440L - attempt 3D mark 01
result: PASS! - first time this computer has ever passed this benchmark fully. Now try aquamark again - massive artifacting all of a sudden, lock in the scene artifacting happens
Test case: vcore 1.65V reduce ram to single 512mb stick
result: hard lock on windows loading screen - reboot - hard lock on aquamark - reboot - stuck at bios screen cannot enter bios - power cycle.
Test case: vcore to 1.525V with GF4 MX440L single 512mb stick
result: locked on windows loading screen - reboot - locked on aquamark loading screen - then dropped back to windows with the message "termination exception caught the thread tried to read or write from a virtual address which is does not have the proper access for" - seems to be an aquamark thing https://community.hwbot.org/topic/16789-aquam … ception-caught/
-reboot - no image on screen - power cycle
Test case: vcore @ 1.675V single 512mb stick
result: aquamark locked at loading screen - reboot - 3d mark PASS! - aquamark fail again
I don't know. I can't really see much of a pattern there. Less voltage made things worse. The right amount made no difference with minimal hardware and more vcore didn't seem to help either. I don't think I have a less powerful agp card than the geforce 4 mx440l - single fps in aquamark 🤣. Surely that 430W psu is enough, the only power hungry thing left is the cpu and the TDP for that isnt even 100W.
shamino wrote:Have you made sure the Vcore setting is correct for your 3200+? Are any of the other configurable voltages set to anything odd? […]
Show full quote
Have you made sure the Vcore setting is correct for your 3200+? Are any of the other configurable voltages set to anything odd?
It is an interesting idea to actually try undervolting and see if the reduced power usage helps stabilize it. On the other hand, maybe voltages need to be increased.
Does the 3200+ work okay with the Prime95 torture test? That stresses the CPU and RAM but it won't stress the graphics card.
Does the 3200+ work any better if you minimize the rest of the setup? The lowest power graphics card you have, minimal RAM, just one drive to be able to run tests, etc.
Far fetched, but: What about setting more conservative RAM timings? I'm guessing the 2600+ is using the same FSB, but even so, maybe there's an interaction with the stress of the 3200+ that makes the timings more sensitive. Other than repeating the tests that have previously crashed, you could also try memtest86.
I don't remember if nForce2 allows asynchronous clocks, but if it does, you could also try setting the RAM to DDR333 or lower and see if it gets any more stable.
You could try measuring Vcore on the board with your meter. There's a group of 6 MOSFETs in the Vcore regulator area (between the CPU and the rear ports). Put the positive probe on the large tab on the back of those MOSFETs, and the black probe at ground (side of the PSU or somewhere easy). At the tab on half of the MOSFETs you should see 12.0V and the other half should show you Vcore. Make sure both of those values are where they should be. Also try watching the Vcore as stress is applied and see if it appears to be stable. If it's not, then check the 12.0V side for stability also.
If you do this then be careful with the positive probe. You can partially insulate it with tape to reduce the risk of shorting anything.
xarg0n wrote:On one of the previous restarts the nvidia control panel warned me performance was degraded due to the gpu not getting enough voltage. I put a multimeter across the spare molex connected to the gpu power (6600gt) and observed a 0.02V change on the 5V & 12V which is nothing really.
The fluctuation looks minimal, but what are the voltages?
That nVidia warning about voltage definitely makes it seem like there's a power issue.
.legaCy wrote:
afaik +/- 0.25V ripple on 5V rail is within the limits of the atx specification, and 0.60v ripple on 12v rail is within the limits too.
That's the tolerance for the DC voltage level (+/-5%), but that's not the spec for high frequency ripple. I don't know what the ripple spec is, but I think it's much less than that. Anyway, it can't be measured easily.
You're right that the tolerance technically allows the +5V rail to be between 4.75-5.25V, and +12V to be between 11.40V-12.60V, but in reality, I don't think you can trust the stability of a system that dips much into the lower half of that range. Issues can start to appear, even though it's supposed to work.
A very conservative, server/workstation quality board might get away with something like a 5V rail running at 4.80V, but I think most enthusiast and budget brand boards aren't really designed to be reliable in that situation.
-all voltages have been on defaults
-I can prime95 overnight if you still think it might be insightful given the above results.
-I haven't looked at memory settings at all but they would be default, I reset all defaults when I got the board.
-I will measure the vcore regulation area tomorrow with a DMM and get back to you 😀
-the original voltages were bang on 12V and 5V fluctuating to 5.01, 5.02V and no more, very solid. 12V was similar.
-the nvidia warning has only happened once. I immediately checked the molex connector to the card and it was connected, but I noticed there was a bit of play, could have been making poor contact?
I measured up all the caps this morning - capacity/deminsions. Can grab direct replacements except for one rubycon series, they would need to be polymers since they stopped making them in 2010. Any more ideas? I can only think caps but it bothers me they all look great. The only common hardware throughout all the testing has been the mobo, hdd and one single ram stick, everything else has had a spare part dropped in so it must be the board?