I need to put my Mushkin ram through memtest, to see if it's bad at all, because even at 5:4 high latency fsb divider @ 213mhz, it causes consistent ati driver failures in 3dmark2001.
Do you have good information on the vdimm mod, i'd be interested to at least read about it.

Update: memtest seems ok with the mushkin (note: the ram speed is 5:4 low lat, 213mhz /ddr426... ever since I installed the modded bios, the ram speed reporting has been weird in some apps)

Stiletto wrote on 2021-04-08, 06:33:

It's because you need more posts before you can do that, so please keep posting. 😀

Got it, thanks! Will definitely try to stay more active here 🙂

RockstarRunner wrote on 2021-04-08, 08:00:

I need to put my Mushkin ram through memtest, to see if it's bad at all, because even at 5:4 high latency fsb divider @ 213mhz, […]
Show full quote

I need to put my Mushkin ram through memtest, to see if it's bad at all, because even at 5:4 high latency fsb divider @ 213mhz, it causes consistent ati driver failures in 3dmark2001.
Do you have good information on the vdimm mod, i'd be interested to at least read about it.

Update: memtest seems ok with the mushkin (note: the ram speed is 5:4 low lat, 213mhz /ddr426... ever since I installed the modded bios, the ram speed reporting has been weird in some apps)
IMG_20210408_125111.jpg

Best way to test the RAM for overclocking in my opinion if you have a different board that takes DDR, test it in that setup.

I'll have to dig up my notes on what I did for the vdimm mod, but I probably wouldn't recommend it since I seemed to have screwed something up with it 🤣

Infineon BE/CE memory can pass Memtest, but just as easily cause visual corruption or crashes in 3D applications.

The Serpent Rider wrote on 2021-04-08, 14:34:

Infineon BE/CE memory can pass Memtest, but just as easily cause visual corruption or crashes in 3D applications.

Yeah this is what I remember. At higher speeds it would pass memtest and superPi runs, but I'd still get instability and crashing during real world use.

I wanted to run memtest just to confirm the mushkin ram isn't faulty. Really, I've only seen issues with 3dmark series, as benching Crysis and fear go ok, but a failure is a failure, regardless of where it occurs. I'm very curious why you've been having much better results, with the same ram and motherboard... Without any better idea, I've put the HyperX ram back, and it actually runs a bit better, after using your setting suggestions.

Messing with this today... I'm not sure if I can push the fsb speed as high with the X6800 because it gets so hot even with the water cooled setup, but will see how it does.

Edit: These settings failed soon after I started running Prime95 🤣. It's probably going to take me a little while to get back up to speed with all the settings.

So I messed around with this for a little while. I'm hitting a hard limit of 280mhz on the fsb and 5:4 divider. Doesn't matter if I loosen up memory timings or mess around with PAT. It just won't go higher than that.

With 1:1 divider, 233mhz fsb is as much as I can go before I start hitting the limits of this RAM in this board. I know the RAM can reach higher speeds in a better board. This new P5PE-VM is an older revision than my other one was. I was also actively cooling the north bridge on my old setup. I don't have that going here.

That said, I'm pretty eager to swap out this board and see how the Asrock 4coredual-VSTA does in comparison with some DDR2, so I might shelf this P5PE-VM for a little while until I can get a separate rig set up so I can run both side by side.

Why in the world are you running the CPU voltage so high?

The x6800 I have in one of my systems runs at at least 3.2Ghz on stock volts with just a good air cooler and it doesn't get hot at all.

I'm maxing out at about 300fsb but that is because the board doesn't have any adjustment for voltage whatsoever. I had to do a pencil mod to bring the NB voltage up.

It's been a while and I would have to pull the system out to get the board brand/model but I think it is one of the ASRock boards.

I was using super PI. You shouldn't have any problem getting to DDR 533 on decent ram. You might have to loosen the timings to 3-3-3-8 which would be spec for DDR2 even though you are using DDR1, Id reccomend de clocking the CPU to make sure your errors are not the cause of the CPU being to far overclocked than it can handle. Throwing more voltage at the CPU doesn't help it when its already gone past what it will do.

i was able to easy hit 533mhz on my ram on 865 chipset.

cyclone3d wrote on 2021-04-10, 01:57:

Why in the world are you running the CPU voltage so high?

It doesn't need to be that high, I just crank it up while I'm trying other overclock settings to help eliminate CPU bottlenecks. Up to 1.5V is within spec for the X6800. Mine still gets pretty hot on stock voltage during prime95 runs though, at least according to ThrottleStop. I feel like the readings are a bit high, though. HwMonitor shows lower temps by like 5-10C.

The pencil mod is something I haven't tried yet. Maybe I'll give that a shot and add some active cooling to the northbridge.

If you haven't already you might consider removing the NB heatsink and applying some proper thermal paste.

1m super PI is not reliable. 2M or higher for proper test.

Warlord wrote on 2021-04-10, 02:19:

If you haven't already you might consider removing the NB heatsink and applying some proper thermal paste.

1m super PI is not reliable. 2M or higher for proper test.

Eh... I only use superpi to run quick benchmarks. I typically use prime95 for stress testing, as it does a much better job especially with multi-core CPUs.

And yes, I will apply some fresh thermal paste to the northbridge for sure if I end up pulling it out to add a fan to it.

You shouldn't have any problem getting to DDR 533 on decent ram.

With 512 mb RAM sticks - yes, plenty of RAM, which is capable.
With 1 gb RAM sticks - no, only Samsung or Infineon chips and your luck may vary, 3-3-3-8 is bit too optimistic.

P5PE-VM VDIMM mod:

50K or 20K variable resistor between pin #2 of LM358 (or just "358", as mine is labeled) and ground. Alternatively you can use Pin 4 as ground. With the VR set to around 10K ohms, I'm getting 2.8v to the RAM.

Make sure you start with high ohms, then decrease the ohms to increase voltage. Always best to test these mods with no RAM sticks.

VDDR measurement points, for board revs. 1.02G and rev. 1.03G:

I managed to fire up my "dead" P5PE-VM board over the weekend, not exactly sure what the issue was but I'm happy to have that one working again. The Rev. 1.03G has some better polymer caps around the VRMs for the CPU and it used to give me slightly better overclocks than I'm getting on this Rev. 1.02G board. I also ordered a test bench case and new power supply so I can start doing some benchmarks and comparisons without having to do full disassembly each time. I dug out an old set of 2x512MB "PQI" brand RAM (which I believe is TCCD) and I'm waiting on a Pentium E6500K I ordered from China along with a Pentium E6700 from here in the US to add to my CPU collection (already have X6800, E7600, and QX6700). For graphics, I have HD3850, HD4670, and 7800GS (anyone know someone with a 7950GT AGP? 😁). Will be interesting to compare the P5Pe-VM results to the 4CoreDual-VSTA with all these different combinations of hardware!

P5PE-VM vagp mod:

50k trimmer / variable resistor between pins 6 and 11 (ground) of LM324, which is behind the blue IDE header. Decrease resistance to increase voltage. This also drives up the northbridge voltage, so active cooling on the northbridge is a good idea, and I also put some small heatsinks on the two mosfets pictured. You can see discoloration there from heat, and that happened before I did this mod.

Green: pin 11 / ground
Purple: pin 6
Blue: vagp measurement point

I'm still stress testing, currently running agp/nb voltage at 1.6, but between this and the vdimm mod it got me just above the 300MHz fsb barrier. Currently using the 5:4 divider with loose timings (3-4-4-8), e7600 @ 11 multiplier.

I dropped the CPU multiplier down to 10 to see how far I could push the fsb, and had it stable at 315. I tried to crank it up to 318 before prime95 threw up an error. Not bad!

One thing I noticed, the 33MHz PCI lock in the modded bios seems to work fine on my rev. 1.03G board. I noticed the lock doesn't seem to work on my other rev. 1.02G board. It moves up along with the fsb when using clockgen or setfsb. I thought it might have been a software glitch, but I expect this the main reason why I'm able to push the speeds a lot higher (even higher now with these volt mods) on the 1.03G revision.

Stable at 315, nice one.
I stopped trying to push my system, after your tips let me get it stable with OK timings, but it's fun to see what you are managing.

RockstarRunner wrote on 2021-04-24, 13:06:

Stable at 315, nice one.
I stopped trying to push my system, after your tips let me get it stable with OK timings, but it's fun to see what you are managing.

Just having fun with it! There seems to be something strange going on as soon as I set the FSB slightly above 303. The system remains stable, but open windows start taking a lot longer to refresh if I drag them around the screen or even just switching tabs in CPU-Z. I'm not sure if that's something to do with the HD4670 or if it's just beyond the limits of what the 865G chipset can handle. I'll try to swap in another video card later on and see if that fixes it. It's definitely not heat because I can literally go up and down a single notch in SetFSB to observe it happening.

I also tried swapping in my quad core (I thought I had a QX6700, but it's actually a QX6800) and it won't post at all. I know I had this CPU working at some point but I can't remember what I did to get it working. I think I'll try resetting the CMOS and swapping in the X6800, configure the BIOS settings, then swap in the QX6800.