Can't get GA-EP45-DS3L host freq over 380MHz

vshourt

Distinguished
Sep 8, 2010
8
0
18,510
First, system specs:

GA-EP45-DS3L, Rev 1.0, F10 BIOS
C2Q Q9550, E0 stepping
4GB RAM, OCZ2P10662G
MSI GTX-470 (standard, not "superclocked")
530 Watt 80 plus power supply

Other stuff that I doubt matters:
WD 500GB HD
AVerMedia A188C PCIe Digital TV Tuner
LG DVD-RAM
Blu-Ray drive (Samsung I think)

Notes:
- I know the power supply might be considered cutting it a bit close, but I'm pretty sure that that isn't the issue here. I originally had a GTX 260, which has significantly lower power requirements, and I still had the same overclocking problems.

- I've tested the RAM thoroughly, and it passes Memtest86+ at 1066 with 5-5-5-15 timings with a host clock of 333 or 400.

Okay, now for the problem. I can't seem to make the system stable in Prime 95 with large FFTs with a host frequency of over 380MHz. With small FFTs, I've taken the CPU to 3.8GHz without a problem, but the large FFTs always have a couple of cores fail due to rounding errors or just blue screen on me.

I'm pretty sure that the host frequency is the problem here, because I lowered the CPU multiplier and raised the host frequency above 380 (while making sure that everything else wasn't overclocked), and I encountered the same problems.

Things I've tried adjusting:

Vcore: I've gone up to 1.4 volts. I don't plan on keeping it that high; I just wanted to see if that was the limitation.

VTT (CPU Termination): I've played with this a bit, and ranged it between the default and 1.36 volts. Truthfully, I don't really know exactly what this does and have been mostly following advice from other posts.

CPU PLL: I've tried adding a bit to this with no luck. This is another one that I don't entirely understand.

CPU Reference: I've tried pretty much every setting available. As far as I can tell it doesn't make a difference.

MCH Core: I've gone up to 1.4 volts. It seems to help stability somewhat (fewer blue screens), but it always crashes eventually.

MCH Reference: If I touch this, it refuses to boot. I haven't tried every value, though.

RAM Voltage: This defaults to 2.1 with my memory, which is an appropriate setting according to OCZ. I've tried raising and lowering it, but without much apparent effect.

I've disabled C1E and EIST. I've got the suspend setting on S1, but I never use it so I'm unsure if this really matters.

I'm able to push the host frequency to 360MHz without modifying ANY of the voltage settings. If I bump the Vcore and VTT up a few hundredths of a volt, I can do 370-380. After that, it seems like nothing I do helps.

The CPU doesn't seem to be the limitation, and after contacting OCZ support I seem to have the RAM working fine.

It seems like the issue might be the memory controller, since the crashes only occur with large FFTs, but that's just a guess.

I've had most of the system since March of '09. It originally had a Pentium Dual Core E5200 in it, which I had OC'd to 3.66 GHz with no problems. Of course, that OC only required a 333MHz host speed due to the Pentium's high multiplier. I upgraded to the Q9550 a couple weeks ago, since I didn't want to have to do a whole platform upgrade to get an i5. Now I'm questioning my decision :/

Please let me know if you need any further information. I probably forgot something, anyway.

It's worth noting that I had the system running with a 400MHz host clock, with the CPU at 3.4GHz and the RAM at 1066 for over a week before making this post, and only had one crash outside of Prime 95. I'm not sure if I'm being picky, but I feel like I need it to be rock solid. On top of that, most similar setups that I've seen have no problems hitting 3.4 GHz and beyond, so I feel like I must be missing something here.

Thanks!
 

d1rtyju1c3

Distinguished
Sep 1, 2010
874
0
19,060




You can take this with a grain of salt but I am not a fan of prime95. Im not saying its a bad program but it can be picky with certain CPUs and large FFTs but if you want you can try to bump the MCH to 1.4v and I wouldnt go over 1.45v. Your VTT is your FSB voltage and again try not to go over 1.4v. Your CPU PLL simply put helps keep your frequency from oscillating to much so if your frequency is jumping around alot under full load try to raise it a couple notches at a time, also to much CPU PLL can make your overclock more unstable. Vcore max for your chip I think is 1.36v so I wouldnt go much past that, if you do, no higher than 1.4v. Remember that once you get it stable start trying to lower voltages and still stay stable and make sure temps dont pass 70c. You dont need to mess with CPU or MCH reference voltage. Also not all cpu batches overclock the same one batch might hit 4.2ghz and another may only hit 3.8ghz so keep that in mind. Hope this helps. [:bohleyk:1]
 

vshourt

Distinguished
Sep 8, 2010
8
0
18,510
I tried moving the MCH up to 1.4v. It did seem to help stability, but it still always crashed eventually. I don't think that the CPU is the problem here, since I lowered its multiplier to 6 and set the host frequency to 400 (for a CPU frequency of 2.4GHz, less than the 2.83 stock), and it still had issues with Prime 95. It's worth mentioning that I DID test it fully with Prime 95 at entirely stock settings.

As to not liking Prime 95: do you have any alternatives in mind? Like I said, I only had one crash outside of Prime 95, and that was in Crysis. I've actually spent around 5 hours in Crysis in total under theses settings. Additionally, the computer is also a media center PC, so I leave it on 24/7, and it seems stable with the exception of the large FFTs test. Granted, it's never really under load while it's just sitting there.

Is Prime 95 just an unrealistic benchmark for stability?
 

d1rtyju1c3

Distinguished
Sep 1, 2010
874
0
19,060
This is what I have come to realize Prime95 is great for stability, most of the time but realy who utilizes 100% of the CPU all the time unless you only run benchmarks 24/7. If you can play games, run the occasional futuremark bench, leave it running 24/7, and your temps stay under 70c without a crash you are fine.
I have a couple of questions though. What is the PSU brand and model, and why do you want 4.0ghz so bad? If you are just gaming 3.6ghz or 3.8 ghz will probably net you the same framerates in a game as 4.0ghz the only time you may see a big difference would be in 3dmark, and like I said not all CPU batches will overclock the same so no matter what your settings are in the bios you may never hit 4.0ghz. You didnt make a mistake by not upgrading the whole platform, the price to performance isnt worth it for an i5 upgrade your set up will perform just as well in gaming as an i5 the only place it may lack is in benchmarks and video encoding. Remember make sure you core temps dont go over 70c heat will kill a CPU before voltage will. Use core temp and let me know what your max temps are during a 30min run of Prime95. Yes I said Prime95, it is great for getting max CPU temps.
 

vshourt

Distinguished
Sep 8, 2010
8
0
18,510
My core temps never hit 70C. I've had the good fortune of having a wide range of weather lately, and when the temperature was in the mid 30s C, the Processor never went over 63 C under extended load on about 1.35 volts. That being said, the temperature here rarely exceeds 22C, even in summer, and at that temperature it stays under the mid 50s C, and idles in the high 20s. I also lock the CPU fan at full. It's a little louder, but I like the peace of mind :)

I'm not desperate for 4.0 GHz. Honestly, I'd settle for a solid, stable 3.4-3.6. The problem is that my CPU has a maximum multiplier of 8.5, and I haven't been able to get the host frequency very stable over about 370 MHz, so my maximum OC at this point is about 3.15 GHz, which isn't a hell of a lot better than the stock 2.83. I can get the thing to be stable in small FFTs to about 3.8 GHz, but I haven't tested that speed with games and other apps.

As to the actual, noticeable performance: you're totally right. In games, the 3.4 GHz clock has netted me about a 10-15% increase in framerates AT BEST. In most, it isn't really noticeable at all. Hell, at the stock clock Crysis never falls below 30 fps on its highest settings at 1080P, so I guess I don't have much room to complain. I guess I'm just annoyed that my old Pentium Dual Core managed to scream along at damn near 4.0 GHz on this same motherboard (it had a much higher multiplier; I never had to take the host frequency past 333 MHz), while I'm having trouble overclocking this one at all. Sure, this one performs better at stock clocks then my old one did when it was practically melting. What I do notice are better boot times and load times for programs and games, but frankly that's not the end of the world. Maybe I'm just nuts :)

In any case, I've got it running at stock speeds at the moment (with the exception of the video card; that baby took a 30% overclock without blinking). I'd really like to be able to get it stable in the mid 3 GHz range, but I doubt I'll lose my mind if it doesn't work out.

PS-
I've been running Prime 95 for the past 15 minutes or so, and the CPU temp appears to have stabilized at 46-47C (it keeps oscillating between them). Stock clocks.
 

d1rtyju1c3

Distinguished
Sep 1, 2010
874
0
19,060
Everything looks good. What is your PSU manufacturer and model? Does your PSU have a 4-pin cpu connector or an 8-pin? Because Quad core use more juice they need an 8-pin CPU connector. I know what your saying about your E5200, my E8400 I could get to 4.6ghz water cooled, those thing were overclocking monsters. Have you tried setting the clock speeds down on the 470 and only overclocking the CPU.
 

vshourt

Distinguished
Sep 8, 2010
8
0
18,510
It's a Rosewill RG530-2. Continuous at 40C, 80 plus certified, etc. The PSU has an 8 pin (Split into two fours), but the motherboard only has 4.

I didn't have the 470 overclocked when I was testing stability with Prime 95, although it shouldn't really matter as far as I know because Prime 95 puts no load on the graphics card, and the GPU scales back down to about 50MHz when not under load. I also locked the PCI Express frequency at 100MHz, just to be safe.

Anyhow, I'm pretty sure the problem lies with the northbridge, not the CPU. At least that's the impression that I get, based on my testing. The memory itself checks out with Memtest, and the CPU does just fine under full load with small FFTs (when the main memory utilization is minimal). It's only when both are under heavy load that failure occurs, and only then when the host frequency is set above 370-380MHz. To me, all of that points to the northbridge. I'm just not sure how to stabilize it.
 
Here's a summary of my three systems:
GA-EP45-UD3P | Q9550 OC'd to 3.6 GHz (425 MHz X 8.5) C3 stepping :(
GA-EP45-UD3L | Q6600 OC'd to 3.6 GHz (400 MHz X 9)
GA-EP35-DS3P | [strike]E5200[/strike] OC'd to [strike]3.78[/strike] GHz ([strike]315[/strike] MHz X [strike]12[/strike])
GA-EP35-DS3P | E7300 OC'd to 4.07 GHz (370 MHz X 11)
The P35 previously had the Q6600 in it running at 3.6 GHz. All have 4 GB of RAM; 2 sets Crucial, one of Patriot. All are 24 hour Prime95 stable on both small and large fft's. Came into possession of a not bad E7300. :)

Are you running your RAM at 1:1? If not, go into the BIOS and change the System Mem Mult to 2.0.

Oh, check the Intel documentation. Intel's recommended max core voltage is 1.3625 volts. Intel's absolute max voltage is 1.45 volts.
 

d1rtyju1c3

Distinguished
Sep 1, 2010
874
0
19,060



It is a good PSU. Do you have line load calibration on. I dont know man it is hard to say it could be the north bridge or like I said it could be the batch just get it stable where you like it and dont worry you have a hell of a system overclocked or not, enjoy it. Run 3dmark and let me know what kind of scores you get ill compare it to mine, and let you know the difference in performance. [:bohleyk:1]
 

vshourt

Distinguished
Sep 8, 2010
8
0
18,510
I generally run the RAM at 1:1 with loose timings when I'm trying to OC the CPU. I've also tried using whatever multiplier gets it closest to (and below) 1066 MHz, since I've heard that this kind of RAM has certain settings that it might "prefer." In any case, the RAM seems to check out just fine when I run Memtest86 on any of these settings. I don't know if that ensures that they're truly okay though.

My MB doesn't have load line calibration, sadly. I monitor the voltages, though. I don't think I've ever set this CPU to above 1.4 volts, and I'm aware of the 1.3625 volt recommended max, and have no intention of running it higher than that in the long term. It's worth noting that I ran my Pentium Dual Core E5200 (which is a 45nm Woldale) at 1.45 volts for about a year and a half with the PC running 24/7, and it never flinched. Of course, that CPU only cost about $60, so I wasn't exactly terrified of the thought of it burning out. I'm planning to be much more careful with the q9550, since it set me back a couple of Benjamins.

With the CPU on stock settings, here's what I got on 3dmark Vantage on the standard "performance" setting:

With PhysX: 20883
No PhysX (PPU disabled): 15506

The GPU score for both of those was about 17600.

I'll tweak the OC a bit and see if it'll run 3dmark stably, just for the hell of it...
 

vshourt

Distinguished
Sep 8, 2010
8
0
18,510
Okay, put the machine back into my 3.4GHz "stable in everything but large FFTs" overclock.

New 3dmark score: 16821, with 18442 GPU

I didn't bother running it with PhysX enabled this time, since I think the PhysX score bump is kind of cheap and not really representative of any real performance increase, and I didn't feel like sitting through that whole battery of tests again :)
 

d1rtyju1c3

Distinguished
Sep 1, 2010
874
0
19,060
My highest score in Vantage with PPU disabled, total 23336 and a gpu score of 22495 what was your cpu score. I am running an i7 860 overclocked to 4.1ghz and I have not run prime95 and I have had no problems with stability. I am also running 2 GTX 275s. My CPU score is 26437. At 4.0ghz my cpu score was much lower at 18355.
Here is a link to compare.
http://service.futuremark.com/compare?3dmv=2215216


I will try to do a run later at 3.4ghz and see what I get but I am pretty sure at stock my cpu overclocks to 3.4ghz and my total score was in the 18000-19000 range
 

vshourt

Distinguished
Sep 8, 2010
8
0
18,510
LOL, that's pretty funny. My CPU score is 13653 at 3.4 GHz, but I seriously doubt that an i7 (even at 4.1 GHz) is twice as powerful. I mean, I'm sure that the higher clock speed and newer architecture would make it at least around 30-40% faster. Maybe even 50-60% on the outside. But twice as fast? I guess that's why they call these tests "synthetics." :)

BTW: DAMN! 4.1 GHz! Those i7s must be absolutely ridiculous overclockers. How does running the processor that fast affect games and apps? When I bump my q9550 to 3.4 from 2.83, most games barely show a bump in fps. The only one where it's really noticeable is Crysis.

I'm actually a bit surprised that my single 470 is even remotely close to two 275s in SLI. Of course the one I got seems to be a bit of a gem, and has let me overclock the ever-loving snot out of it, LOL. I suppose I'm lucky in that respect; if there's one component that I want to overclock well, it's the GPU.

That setup sounds pretty power hungry. Do you have one of those nutty 1000 watt PSUs? It must get the room pretty toasty, too. I mean, I thought my setup was a decent space heater, but, well, damn!

Lastly, out of curiosity, is there anything that causes your setup to get choppy? I mean, there aren't many things I can throw at my new setup that bog it down to the point where it's noticeably slow, and yours sounds much, much faster.

Edit: I just noticed that the 4.0GHz score seems much more in line with my estimates. Actually, it seems really doubtful that an extra 100MHz could improve the score so much, unless there is something screwy with 3dmark. Did you change any other settings between those two tests?
 

d1rtyju1c3

Distinguished
Sep 1, 2010
874
0
19,060
I dont know why the scores are so different but my total scores were only 1000 points different and my GPU was lower to, But 3dmark scores are never consistent but if run at the same speed multiple times the total scores are usually pretty close. I think the reason my score is so much higher is because 3dmark vantage utilizes threads so it is like I am running 8 cores almost, now if that would make my score double yours, maybe but I dont know, but all the other 860 results are about the same as mine.
Im not surprised your 470 is so close to my 275's the 470's are beasts I would be extremely happy to have only one 470, my cards might beat yours in higher resolutions but at 1680x1050 and lower they probably perform about the same. Yes the thing is a heater it is almost to much to sit by it while I game for long periods of time but it is worth it, it isnt that power hungry I was a little worried my 850w PSU wasnt going to be enough but it has handled my setup great about a year and a half. I havent noticed things get choppy. Are you talking about it being sli? Drivers have come along way so sli isnt so bad anymore when you get into tri and quad sli things start to become less stable. As for in games I dont notice more than maybe 10-15fps increase from stock to 4.0ghz.


Your score is good for you set up and in games you should net about the same frame rates I do, depending on the game it could be higher or lower.
Here are Vantage scores to compare to yours the first one is at 4.5ghz and only netted about 2200 points higher than yours.

http://service.futuremark.com/compare?3dmv=2435927

And the second is at 3.4ghz and scored 2400 points less than you.

http://service.futuremark.com/compare?3dmv=2441600

So I would say for you setup your score great, maybey try to get it to 3.6ghz and see if the score goes up. [:bohleyk:1]