D820 Dual Core Overclocking

Leonardo

System Specifications:

Intel Pentium D 820 (Dual Core, 2.8GHz/core, 800MHz FSB, 2MB L2 Cache)
MSI 945P Neo-F motherboard
1 GB (2X512MB) Gigaram DDR2 667 (value RAM) (4-4-4-11 SPD)
ATI Rage 8MB PCI video card (it does 1280X1024, 32bit color!)
Robanton 600W PSU, modified
Chieftec (Antec 1030) case, modified
Zalman CNPS 7000-AlCu HSF
WD800JB HDD

Preliminary Results for Overclocking, each core:

CPU Frequency - (MBM5 & CPU Z) 3023MHz; 14 X 216

I've hit a brick wall at BIOS FSB 215 (CPU Z 216). So far, I have not tweaked the DDR2 except for switching between 533 and 667. I've enabled "CPU Ratio Unlock", which allows the CPU:FSB ratio to switch between 3:4 and 3:5 automatically. I don't know whether the FSB barrier is DRAM limited or CPU limitations. Interestingly, maybe the problem is the motherboard. After increasing the VCore from default 1.37v to 1.5v, the BIOS hardware monitor and CPUz still indicate the voltage as 1.37. Bad motherboard?

There may be other BIOS settings I've overlocked. I'm new to MSI overclocking and also new to DDR2. Ah, who knows, maybe 3GHz is the limit for this early release processor.

muddocktor

I went to msi's site and downloaded what I believe is the manual to your board, Leo. I notice there is a field to adjust the cpu ratio. With your ES proc, you should be able to do some adjusting of the multiplier I'm pretty sure. Check that out and see if you can adjust down on the multiplier. If you can clock down the processor with the multi, then you can see if you are being limited on your fsb speed by either the mobo or the ram. Also, they have that auto overclocking stuff in bios and you might try that out and see what you can get out of your new proc.

Leonardo

I've got the downloaded manual as well. I found nowhere in the BIOS where I can change the multipliers. I'm pretty sure they are locked. I'll look again - maybe something i missed.

Leonardo

Nope, I'm pretty sure this CPU is multiplier locked. There is nowhere in the BIOS to change the multiplier.

3024 (216FSB X 14multi) still seems to be the limit. I tried relaxing the RAM timings way down to 5-5-5-15, raised the vCore and memory voltages, shut off all unused IO devices. No dice. If anyone has suggestions, I'd be glad to try it. Heat does not seem to be a factor. Core temperature, reported by both MBM5 and the BIOS, did not exceed 44*C. (I thought these Smithfields were supposed to be hot?)

muddocktor

Leo, is there a setting in bios to run the ram asynch with the fsb, just to make sure it's not a ram limitation? On the manual I downloaded, it shows a setting in the core cell section of the bios that will let you select the memory speed setting. Try setting it to 400 and see if you can get a higher fsb speed. Also, they should have a tattletale on top in the same section saying if the multi is unlocked and also a line there to set a different multi if the core has some unlocked multipliers, according to the manual I d/l'ed.

Leonardo

The BIOS section in "Cell" showing multipliers shows one thing only: "200X14", and it does not allow highlighting like when you select a changeable item. This CPU is mulitplier locked. I've tried both 533 and 667 settings. Brick wall is 215 or 216FSB either in 533 or 667, even with timings relaxed to 5/5/5/14. Right, the manual does show an option for DDR400, but it is not available in my BIOS. BTW, I'm running the latest (11/25/05) BIOS. Apart from that, the system is rock solid. The dual core (yes, I know, not a true dual core) REALLY makes a difference in multitasking. After running this machine, I doubt I'll ever buy another single core chip.

TheLostSwede

http://www.digital-daily.com/motherboard/msi-945p-neo/index05.htm

Have a look at that page.
Can you run CPU-Z and see if the screen looks like from that review? The review shows that he is using an unlocked cpu.

Leonardo

Hi, Mack, thanks for dropping in. On the "Cell" page of the BIOS, mine is different from the one in the article you linked. Mine does NOT have:

High Performance Mode
CPU FSB & PCI-E Clock
Dynamic Overclocking
Adjust CPU Ratio

Also, on the CPUZ screen, mine shows "x 14.0" only, instead of a range. That tells me there is only the factory default multiplier available. I've attached the two images: 1) BIOS Cell Menu screen from linked article, and 2) screenshot of CPUZ from my computer.

ryko

i know it shouldn't matter, but i have heard of people hitting a wall b/c of msi's corecell application suite. maybe try uninstalling it and then go back to overclocking to see if it matters.

if not, then i think you just have a very poor overclocking motherboard. you did buy it from newegg's refurbished section, right? maybe the 215fsb limit is why it was returned?

Leonardo

Ryko, you may very well be correct in both your points. I don't see how the Core Cell software would interfere with the BIOS options and preliminary overclocking, as the BIOS is embedded, and the Core Cell software is in Windows. By "preliminary", I mean the 215FSB brick wall is met at the beginning of POST. At 215, when I hit F10 to save BIOS changes the screen is black - nothing happens except spinning fans. Concerning the refurb motherboard. Yes, it is. It was a money saving measure, which I do not regret. What overclock I can get out of this system is fine, be it modest or impressive. The goal here was to have a Folding beast at a modest price, and to have fun learning a new CPU-DRAM-Chipset protocol. But I'll try your suggestion about uninstalling the Core Cell software.

ryko

yeah, i know it a windows thing vs. a BIOS thing, but this is just what i have read in the past...not even sure it was your exact board...worth a try at least!

i also remeber something about artificial ocing limits being placed on 915/925/945/955 chipsets to like a 10% max oc unless extreme measures were taken by the mobo maker. i think asus and abit got around the 10% thing, but i don't remember if msi did....or if they did which of their motherboards circumvented it.

Leonardo

FSB BARRIER IS SMASHED!

I'm working my way up incrementally, 1 or 2MHz at a time. Here's where it stands now:

3220MHz - 14/230
FSB: 230/920
vCore: 1.55v
vMem: 2.05v
Memory Freq: 383.7 (CPU Z)
DRAM Timings: 4/4/4/11
CPU-FSB Ration -- 3:5

I am overclocking in Windows using the MSI CoreCenter software. I'll report back on progress. Time to observe for stability. Two instances of folding are running, one each per core.

csimon

holy cow you need to start posting some bandwidth benches!
what multi are you using now?

Leonardo

The only multiplier that is available - 14. Core temperature stands at 47*C (MBM5), 48*C (SpeedFan), 65*C (MSI CoreCenter). 65 can't be right. I don't think I'd have an 400MHz + overclock if the core were that high. Gotta few more things to do before bumping up the FSB again. I'm presently typing on the the 820! Had a browser crash a couple minutes ago. I've bumped up the voltage a little bit.

Do you think 2.1v is too much for this RAM? Default voltage is 1.8.

csimon

gee that's a good question! Mfg's site only says 1.8v ...I guess everything above that is chancing it but I dunno if I would go past 2.1v be honest. That's a .3v increase ...my redline is rated from 2.6-2.9 ...a .3v increase from default stock.

Maybe Mack can shed some light on that.

Leonardo

I put my fingers on the RAM chips a while ago while I had a slideshow running. They were very warm, but certainly not hot. The heatsink is throwing off heated air very fast. Love those Zalmans.

OK, time to run up the FSB some more.

Leonardo

3304MHz - 14x236
FSB: 236/44
vCore Set/vCore Indicated: 1.56/1.47 (default 1.37)
vMem: 2.00v
Memory Freq: 393.3 (CPU Z)
DRAM Timings: 4/4/4/11
CPU-FSB Ratio -- 3:5

Don't know how much more the PSU will support. I had reported earlier that the 12v rail was very health. Hmm, well it's only good according to MBM, which shows it to be 12.22. Core center and BIOS both show it at about 11.75-80.

(Stability testing is nothing sophisticated. I'm running two instances of Folding, browsing the internet, and turning on a slideshow of 1 to 3MB digital photos.)

csimon

Leonardo wrote:

(Stability testing is nothing sophisticated. I'm running two instances of Folding, browsing the internet, and turning on a slideshow of 1 to 3MB digital photos.)

Yeah but will it do pron?

Leonardo

Observations on overclockings through software:

It's really nice not having to enter the BIOS, save settings and reboot after every tweak. I was really against using the overclocking utilility. As my only experience previous was the good old fashioned line-by-line changes in the BIOS, software overclocking just seemed wimpy. A while ago I tested to see what a shutdown would do. Upon rebooting the system defaults to the settings that were last saved manually in the BIOS. CoreCenter only holds settings when Windows is running.

I suppose tomorrow I'll remove all the MSI software and return to in-BIOS, manual overclocking. Ryko suggested I try uninstalling the OC software to see if somehow it was the cause of the previous 215 FSB "brick wall".

ryko

so what did you do to get past 215? just more voltage?

Leonardo

I had read in a motherboard review, here, that the MSI 945P Neo might have problems with cold boots after resetting the BIOS. It just seemed to me there had to be some FSB headroom left. So I just tried the MSI software overclocking, "CoreCenter".

Still working my way up on the FSB ladder. Currently stable (I think?) at 2332MHz, 14X238/952, memory 396.7MHz. The only instability I've encountered so far in this overclocking foray has been the keyboard occasionally not working. That may just be the keyboard though. It's ten years old.

I'm watching the vCore readings closely through SpeedFan and CoreCenter. Not good. The setting I input was 1.55v; readings are fluctuating between 1.45 and 1.49. CPU core temperature is good - only 47*C at full load on both cores.

Leonardo

Running up the FSB, I started hitting Windows shell loading problems at 14X240. I shutdown, re-entered the BIOS and backed off the RAM timings from the SPD of 4/4/4/11 to 5/5/5/13. Fixed the problem. Above 245 FSB, CoreCenter became unstable and refused to take my inputs. The system though, did not crash or freeze. I've backed it off now to:

CPU Freq: 3360MHz - 14x240
FSB: 240/960
vCore Set/vCore Indicated: 1.53/1.45
vMem: 2.00v
vPCI/PCIe/AGP: 1.7
Memory Freq: 400 (CPU Z)
DRAM Timings: 5/5/5/13
CPU-FSB Ratio: 3:5

I guess 2 X 2.8GHz @ 3.36Ghz isn't so bad. Seems stable. We'll see.

muddocktor

Leo, that's great news on how your overclocking that 820 is going.

About your temps; I wouldn't dismiss the core cell readings without verifying by using throttlewatch, which checks to see if the processor is throttling (due to temps). Just google throttlewatch and it should get you a link to d/l it. If it shows the proc is throttling, then your temps are probably running in the 60+ C range. And it wouldn't surprise me if they are running that high, just cooling it with the Zalman. If it's throttling it won't hurt the proc, just reduce it's folding efficiency. Since you value a quiet machine, you might consider watercooling that machine too, if it's throttling. You might check out Dangerden for a waterblock and radiator, since they are located in Washington or Oregon to save a little on shipping.

Leonardo

Both CPUZ and SpeedFan show the processor running at 3360. Wouldn't throttling reduce the frequency? Also, CoreCenter shows the CPU running at 65*C regardless of the CPU clock or load. It even indicates 65*C at idle at default clock of 14X200. BIOS core temp readings are within *1C of MBM5's and SpeedFan's. I've been attentively monitoring Electron Microscope, and it shows no slowdown of time/frame. Nevertheless, I'll download and run the software you recommend. I'm always eager to try new monitoring utilities.

ryko

Leonardo wrote:

...Above 245 FSB, CoreCenter became unstable and refused to take my inputs. The system though, did not crash or freeze....

that's what i was talking about earlier...if you want to get around 250+fsb, you have to ditch the corecenter stuff. for minor overclocking it doesn't seem to interfer.

good job on the oc!

Leonardo

Mudd, Ryko, thank you both for working with me here. I consider this endeavor YOUR overclock as well as mine. Let's deal with clock throttling first, then with overclocking sans MSI software. Mudd, I've attached a screenshot of three monitors. Throttlewatch does indeed show what looks like significant throttling. But, the CPU Frequency History graph shows the frequency to continually maintain at 3360 except for minor blips. What does this mean - throttling appearing to significant, but CPU clock holding steady? Seems like a contradiction?

Also, which temperature monitors should I trust? CoreCenter's monitor is 17*C higher than MBM5's, SpeedFan's, and the motherboard's BIOS.

Leonardo

After some more testing and monitoring, I am inclined to think that maybe MSI's temperature monitoring may be more accurate than I had thought. Twice now I have quickly shut down the system and entered the BIOS' hardware monitor. Each time it showed the CPU core in the high 50's *C. Ouch. I think I will swap the Zalman 7700 in System No. 1 for the 7000 in the D820 system. Aaargh, looks like I've nearly run out of cooling tolerance for this beast. I'm glad I had not disabled thermal throttling in the BIOS when I set up this machine! I find it remarkable though, that I could get a 560MHz overclock at core temperatures approaching the CPU design limit of 70*C!

ryko

yeah these little intel heaters really require water cooling if you want to get a high oc. i would be happy with anything around/over 230fsb on air.

the single core 3.0e in my g/f's pc can easily get to 3.6ghz (240fsb), but even with an xp-90/sf-2 it was climbing to 70+* C under full load---totally stable mind you, but my uguru monitoring software would go nuts and flash red and beep an alarm. plus the pwm's were climbing to 75+* C which can't be healthy.

it now sits comfortably and more importantly quietly at 3.2ghz (215fsb) and never goes above 60* C under full load.

Leonardo

So then, you are implying that the MSI software temp monitor is probably more accurate than MBM and SpeedFan? If so, I am now inclined to agree with you.

ryko

i gotta tell you that i would have thought the exact opposite b/c most of what i have read says that the corecenter in inaccurate.

but since the 820 is essentially 2 prescotts taped together, i would think it has very similar characteristics to a single core prescott (if not worse). since my prescott behaves pretty much like yours at the 600mhz overclock level.....

it would seem that corecenter is right on. msi must have got their stuff together on the most recent corecenter update.

Leonardo

Two Prescotts! Run for the hills! To play it safe, I will now start relying on the MSI CoreCenter reading. And yes, it is pretty close to what the BIOS reports. The BIOS reports a few degrees cooler, but that could be explained by the cooling the CPU experiences from the time it takes to close all the programs and reboot into the BIOS. Oh well, I'm going to swap heatsinks and see if I get any more thermal headroom. The Zalman 7700 is not top of the line, but it should cool about 3*C better than the 7000 that is on the CPU now. I'll post results later today.