agent_x007 wrote on 2025-02-26, 20:08:

Latency isn't tested via memtest... You need a AIDA64 cache&memory test (for example) or Intel MLC GUI/yCruncher/"SuperPI_XS 1M" […]
Show full quote

Latency isn't tested via memtest...
You need a AIDA64 cache&memory test (for example) or Intel MLC GUI/yCruncher/"SuperPI_XS 1M".
It's about how much of perf. difference is between stable options.

EDIT : 1:1 isn't "best" or "fastest" - it's just is there for highest FSBs OC.

Soo... Let's say I have 800mhz ram, and I wanna OC the FSB from 266mhz to 333mhz... Is it better to downclock that ram to 667mhz, or just keep it at 800mhz?

(PS, I will also test latency with AIDA64 cache&memory, should be available for XP, right?)

That's the thing, it depends on what strap MB BIOS sets for your frequency 😉
I think latest AIDA64 might not work for XP, but older should be fine - just use .zip not .exe (just in case).

Fastest DRAM speeds for 1333MT/s FSB (on Intel chipset), by strap :
1. 200MHz = 1332/1111
2. 266MHz = 835/1002/1335
3. 333MHz = 667/800/1066
4. 400MHz = 667/887/999
^End performance also depends on "Performance Level" used by chipset (can be seen in memset program) and timings.

agent_x007 wrote on 2025-02-26, 20:20:

That's the thing, it depends on what strap MB BIOS sets for your frequency ;) I think latest AIDA64 might not work for XP, but o […]
Show full quote

That's the thing, it depends on what strap MB BIOS sets for your frequency 😉
I think latest AIDA64 might not work for XP, but older should be fine - just use .zip not .exe (just in case).

Fastest DRAM speeds for 1333MT/s FSB (on Intel chipset), by strap :
1. 200MHz = 1332/1111
2. 266MHz = 835/1002/1335
3. 333MHz = 667/800/1066
4. 400MHz = 667/887/999
^End performance also depends on "Performance Level" used by chipset (can be seen in memset program) and timings.

Ah dang, I don't remember all of those... I even saw them in BIOS when I did some OC and dowclock testing. Pretty sure I both saw the 667 and 800 with a 333mhz FSB, because I actually selected 667 on the DRAM for the first OC test I did at 333mhz FBS (just a boot test to see if everything reached the OS flawlessly, it did.)
But, then again, I'm also am pretty sure that I saw 800 even with the 266mhz, with the multiplier of DRAM showing at x1.50 in HWiNFO, and there was an 800mhz setting with a 266mhz FSB.

Pretty sure this MOBO is kinda flexible in ram timings, AFAIK it was a MOBO made with overclockers in mind. It was actualy chance that I got this model exactly.
The manual is pretty bare (maybe they took for granted that people getting this kind of MOBO knew how to use basic OC functions in BIOS).

I'm not that expert in DDR2 memory, I've only ever handled DDR4, wich thanks to FSB being integrated in modern CPUs, it's not something you need to worry about. Just slap the ram, load the XMP, and you're set. And also SDRAM, wich is much more simple, and usually just takes a jumper or two on the mobo to set it right.

Hello there, getting back to everyone with the Overclocking experience.
I first did a test run with stock MOBO settings:

• FSB 266 mhz
• Ram at 800mhz, 1.9 Volt and 5-5-5-18 cas

bench results:

• Passmark PerformanceTest 8.0: 1260.9
• Unigine Heaven 1.0: 968
• Cinebench 11.5: 2.63

I also opened AIDA64 to check and bench memory latency, but I have not recorded the results. But by looking at the score list, it appeard to be in line with similar settings ram.

Then I went with the OC, just like I had tested last week to check if it posted.

• FSB 333 mhz, stock volts for both CPU and NB/SB
• RAM at 667mhz, 1.9 Volt and 4-4-4-12 cas

Bench results:

• Passmark PerformanceTest 8.0: 1417,6
• Unigine Heaven 1.0: 972
• Cinebench 11.5: 3.30

Cinebench went on a run for 30+ minutes of continued rendering. No issues with stability and temps, with Tjmax reaching a peak of 66°, while being most of the times at 65.

I then decided to try again AIDA64.

And it reboot the system.

Try again launching AIDA64. The software is just about to open... and it reboot the system.
Weird. Let's roll back to stock CMOS settings and see if it change something.

Opening AIDA64, loading the software, it reboots the system before it can open.
Now that's really weird.

I delete the software folder. I manually reboot the system, I extract it again in a different folder. And now it works. I close it, reopen it... Now it doens't, rebooting the system. When the system is back on again, I try again and this time it works. (Almost) every time.
The fudge?

I now try setting up FSB at 333mhz (with ram at stock settings), and launching AIDA64 again. Nothing. No matter what I try, every time it tries to load the software, it reboots the system.

AFAIK this is really weird behaviour. It doesn't even appear to be system instability, becasue any other stress test with both Unigine Heaven and Cinebench went on without a single problem.
I don't know if it's worth it to open a dedicated thread in the "software" section, but I still can try an older version of AIDA64 (or Everest), and see if it gives me similar issues. I don't even know why it should cause instabilities just by launching the program. Does it perform some bench or stress test when it opens itself?

I currently reverted to stock settings, even if I was alredy pretty happy with the CPU uplift. But the gaming uplift in Unigine Heaven was in the margin of error, 4 points difference (also, while keeping MSI afterburner open with Rivatuner statistics, I drop from 970 points in unigine to 920. Is this normal?)

If AIDA64 crashes on the same test, but doesn't under stock - somethings up (example : too high power usage or unstable voltage = PSU cuts power, alternative is too high temps or unstable chipset settings).

When it crashes regardless of everything - drop it/use different program.

agent_x007 wrote on 2025-03-04, 20:26:

If AIDA64 crashes on the same test, but doesn't under stock - somethings up (example : too high power usage or unstable voltage = PSU cuts power, alternative is too high temps or unstable chipset settings).

When it crashes regardless of everything - drop it/use different program.

The thing is: It's not a test. It crashes the whole PC (or, simply, hard reset it) before it can even open itself. I manage to open it only with stock settings, and after a few reboots or moving the software in other locations. Because sometimes it hard resets itself even on stock settings when trying to open it.
And it's not like I wasn't able to open it before, since I tested the ram latency (it works, even with tighter OC timings)

I'm gonna try an older version in the next days, and maybe not the portable one.
Because it's weird I'm encountering unstable voltages or high power usage by just opening AIDA64, and not while stress testing cinebench 11.5 for half an hour + another half an hour of unigine heaven.

If nothing works, I can try a slight overvolt, since up until now I was doing everything on stock mobo voltage settings (even the ram, at 667mhz 4-4-4-12 and 1.9v)

I'd chuck it to AIDA64 acting up if the other programs are OK.

If you want a good CPU test, use Intel Linpack stress test. This will really really load down your CPU properly.
OCCT has the Intel Linpack stress test. To get it to run, first run the stock OCCT CPU stress test for a few seconds, then stop it. Then select the Linpack test and make sure it runs on all 4 cores. If that doesn't crash your CPU, nothing else will. I find Intel Linpack stress test to make CPUs pull about 10-20% higher power consumption than other CPU stress tests I have tried. So it really does work it out.

Mondodimotori wrote on 2025-03-03, 13:03:

Cinebench went on a run for 30+ minutes of continued rendering. No issues with stability and temps, with Tjmax reaching a peak of 66°, while being most of the times at 65.

I wonder if that is the same as the CPU core temperature, CPU Tcase temperature, or something else.
Intel lists the maximum Tcase temperature for the Q6600 as either 63C or 71C, IIRC, depending on which stepping you have.
Either way, if your core is reaching 65C under load, see if you can have your CPU fan rev up under load. Given the huge CPU heatsink, I find these temperatures a little high. With a Dell 4-heatpipe cooler and 92 mm fan, I can keep my OC'ed Q6600 below 63-64C most of the time under max load (my cooler does get loud, though, as I have set up a pretty aggressive cooling curve.)

momaka wrote on 2025-03-05, 17:02:

I'd chuck it to AIDA64 acting up if the other programs are OK. […]
Show full quote

I'd chuck it to AIDA64 acting up if the other programs are OK.

If you want a good CPU test, use Intel Linpack stress test. This will really really load down your CPU properly.
OCCT has the Intel Linpack stress test. To get it to run, first run the stock OCCT CPU stress test for a few seconds, then stop it. Then select the Linpack test and make sure it runs on all 4 cores. If that doesn't crash your CPU, nothing else will. I find Intel Linpack stress test to make CPUs pull about 10-20% higher power consumption than other CPU stress tests I have tried. So it really does work it out.

Mondodimotori wrote on 2025-03-03, 13:03:

Cinebench went on a run for 30+ minutes of continued rendering. No issues with stability and temps, with Tjmax reaching a peak of 66°, while being most of the times at 65.

I wonder if that is the same as the CPU core temperature, CPU Tcase temperature, or something else.
Intel lists the maximum Tcase temperature for the Q6600 as either 63C or 71C, IIRC, depending on which stepping you have.
Either way, if your core is reaching 65C under load, see if you can have your CPU fan rev up under load. Given the huge CPU heatsink, I find these temperatures a little high. With a Dell 4-heatpipe cooler and 92 mm fan, I can keep my OC'ed Q6600 below 63-64C most of the time under max load (my cooler does get loud, though, as I have set up a pretty aggressive cooling curve.)

Sorry, it was the hottest core for Hardware Info, 66°. HwInfo also list tjmax for the junction at 105. It sure wasn't Tcase the temp I was reading, otherwise I would've come very close to that limit even during the stock stress test.

I'll have to get some fan control software, that's silent in the background and not taking up lots of resources, because I couldn't find anything in BIOS, and was also unable to locate the DFI managing software that came onthe motherboard CD (even on internet archive and The Retro Web wasn't present). It supposedly was able to both perform OC from windows and fan control.

As a note, I didn't find those temp that strange, considering the case. It's not the Fractal Torrent I sit my main rig in, but a Focus G with only two intake fans, and no exhaust one. So it may be an airflow issue.

I still have my 2008 build. I went big, so Intel X38 chipset it was. The Gigabyte board I have came with the full compliment of legacy I/O as well, so makes for a good bridge machine.

I had a good experience replacing the Q6600 I bought in 2007 with a Xeon E5450. So good in fact I bought a used aliexpress mobo and another E5450 to put in it. Both ran stable at 3.6Ghz (stock is 3.0) with no voltage changes. I bought the Xeons (and 2nd mobo) in 2018 and they had the pin swap and notches already cut, so it was just a quick replacement after modding the bios on both boards. That was the only stressful part, but it worked without a hitch. The only issue I had was with one board, the one I'm still using. The memory divider wasn't calculated correctly-- right answer, but the wrong way and it crashed until I set it manually with one almost trivial change. It's been completely stable for years at 3.6Ghz with lower TDP than the stock Q6600.

Nematocyst wrote on 2025-03-09, 04:17:

I had a good experience replacing the Q6600 I bought in 2007 with a Xeon E5450. So good in fact I bought a used aliexpress mobo and another E5450 to put in it. Both ran stable at 3.6Ghz (stock is 3.0) with no voltage changes. I bought the Xeons (and 2nd mobo) in 2018 and they had the pin swap and notches already cut, so it was just a quick replacement after modding the bios on both boards. That was the only stressful part, but it worked without a hitch. The only issue I had was with one board, the one I'm still using. The memory divider wasn't calculated correctly-- right answer, but the wrong way and it crashed until I set it manually with one almost trivial change. It's been completely stable for years at 3.6Ghz with lower TDP than the stock Q6600.

Yeah, a real shame you need to mod some late Xeon chips to work on LGA 775.

Anyway, I finally tested a different version of AIDA64 (from 2011), and it worked without problems with the OC in place.
Getting good results from both memory and CPU benchmarks, but I still need to tinker with Speedfan. The PC is basically silent, and I don't think Speedfan is doing anything to both CPU and Case fans. Only the GPU fan responds to the program, making a ruckus when going over 30% speed. I'll probably keep it on low speeds until I get a 560ti.

But I still have found some weirdness in this version of AIDA64: It reports lower temps than HwInfo during stress tests. Now, I didn't do a long stress yet (10 minutes), but I noticed that there's a pretty sizable gap in reported temps from the CPU sensors: in HwInfo they reached 60°, with peaks at 65° (so just like it used to do with Cinebench). But, in AIDA64, they top a 50 something during the same stresstest. Wich is weird, considering the sensor should be the same. I alredy checked for eventual offsets in both program's settings, but to no avail.
Also, I found the reporting from the MOBO sensor, wich (I think) should be used to monitor the TCase temp. Never seen it over 45° even during stress tests.