Trouble overclocking Athlon II X4 630 on Asrock 880GMH/U3S3

[doveman]

While you have it off, if you have a jigsaw available cut a hole in the motherboard tray so you don't have to remove it next time to mount an aftermarket heatsink. In addition it will help reduce the socket temps. if you do it just make sure you don't have any of the edges sticking up, a file will also come in handy to remove them.

I'm not going to start cutting up the case thanks. Did that before with another one and it was a hell of a job trying to tidy up the sharp edges and cleaning up the shavings stuck in the cracks.

I think you mean a hole behind the CPU and the case already has one of those but that doesn't help me.

The reason I had to remove the board is because I can just about get the little spanner in there to undo the nuts holding the heatsink on but there's not enough room to get my hand in there to put the left hand one back on (the right-hand one's easy enough once I remove the RAM). Even if I'd managed to get the nut back on somehow, the likelihood is that it would have taken ages, with much cursing and it's just easier to remove the board so that I can easily get to the nuts. If I cut the rear exhaust fan grill out and removed the fan, that would probably give me enough room to get my hand in for the left-hand nut, but again it would be very sharp and I'd have to line it with rubber before going near it and it's just not worth the effort when I can remove the board in a couple of minutes. It's not like I'll be doing it again once I've got it fitted properly anyway and nor will my brother (he's not technical at all and wouldn't know where to start).

Anyway, I tried your suggestion and it looks like it's worked a lot better. I did use a bit too much TIM on the left-hand middle pipe as it came out a bit quick (new tube probably had something to do with that, nothing to do with my shaky hands and poor co-ordination!). I'm not sure it's quite good enough yet though and it looks like there might be some uncovered area in the middle of the left-centre pipe. It's hard to tell exactly what part of each pipe is actually in contact with the CPU. I don't think the TIM pattern tells us much and fitting the heatsink requires putting it on as square as possible but then when fitting the ACK-U01 crossbar, I have to move it around a fair bit to line up the holes with the heatsink, so any analysis of the TIM spread would probably be skewed by that.

Looking at my previous heatsink photo, I'm guessing the dark patches of TIM are where it's heated up, in which case it's seems like the heatsink is only contacting the CPU on one side (the right) for some reason, which can't be good.

 

[Mandrake4565]

looks like you may want to put a little bit less TIM and your good.

 

[dostov]

Looking much better now, doveman. Let us know if the remount/reorientation helps with temperatures!

 

[doveman]

OK, so either the TIM re-application or the re-orientation or a combination of both has definitely helped.

Stock (FSB 200, CPU 2.8Ghz):
Exhaust fan (Xilence) 920 RPM, Front Intake Fan 610 RPM, CPU Fan 1035 RPM
Idle: CPU 35, Core 24, MB 39
IntelBurnTest (Standard, 5 runs): CPU 50, Core 42, MB 42. CPU Fan 1454 (spun up at around 48c with target temp 45c). Front fan didn't speed up from 600RPM (target temp 55c). CPU-Z shows 1.36v

Overclock 20% (FSB 240, CPU [email protected] in BIOS), CoolnQuiet disabled. HT 1920Mhz, NB [email protected]. RAM 960Mhz@5-5-5-15-19-2T
Fans as above.
Idle: CPU 38, Core 28, MB 41 (CPU-Z @1.368v)
IntelBurnTest: CPU 56, Core 48, MB 42, CPU Fan 1480 RPM, Front Fan 610 RPM (CPU-Z: 1.416-1.424v)

So the Core has come down from 62 to 48c, which is great, but the CPU has only come down from 62 to 56, which seems a bit strange.

IBT passed the 5 runs when overclocked but it did that before and then I got a BSOD when using Iron Portable, so obviously IBT isn't sufficient to establish stability.

 

[RGone]

So the Core has come down from 62 to 48c, which is great, but the CPU has only come down from 62 to 56, which seems a bit strange. Why so strange? They are two different temps because they are in two different locations. Good contact or whatever can lower the internal core temps much more easily than the cpu temp also known as the socket temp in some circles.

 

[dostov]

Good news, doveman! Test for stability and you'll be good to go.

Next you can work to get that RAM up to speed!

 

[doveman]

So the Core has come down from 62 to 48c, which is great, but the CPU has only come down from 62 to 56, which seems a bit strange. Why so strange? They are two different temps because they are in two different locations. Good contact or whatever can lower the internal core temps much more easily than the cpu temp also known as the socket temp in some circles.

That's fair enough but whenever I've asked, I'm sure I've been told it's the CPU temp I need to keep under 55c and if we can't easily control that and it's not affected by good contact, TIM, etc then it seems a bit hard to do that, in which case surely the Core temp is what we should be concerned with?

 

[RGone]

As I understand it, the cpu itself can shut itself off just like Intel. That temp is usually around 90c. Now that temp is calculated between Tcase and Tjunction which is not a direct read by any software that I am aware of. So the worry over CPU/Socket temp is valid but not a thing for undue paranoia. CPU temp can be lowered by directing air around the base of the CPU socket or when we run FX processors we direct air onto or even away from the rear of the mobo at the socket, thru the access whole cut in the mobo tray in the better cases of today.

 

[doveman]

As I understand it, the cpu itself can shut itself off just like Intel. That temp is usually around 90c. Now that temp is calculated between Tcase and Tjunction which is not a direct read by any software that I am aware of. So the worry over CPU/Socket temp is valid but not a thing for undue paranoia. CPU temp can be lowered by directing air around the base of the CPU socket or when we run FX processors we direct air onto or even away from the rear of the mobo at the socket, thru the access whole cut in the mobo tray in the better cases of today.

So I take it you're saying that it's the Core temp I should be aiming to keep under 55c and not to worry too much about the CPU temp?

 

[doveman]

Thanks Blaylock, that clears that one up for me.

Regarding the VRM/Mosfet temps, I don't think my board reports those and the MB temp is probably from the NB chip or something, so it's hard to know what they're at. My Asrock 880GMH/U3S3 board is probably not ideal for overclocking and is only has a 4+1 stage I think, so what do you suggest I should limit it to. For example, will it be OK to go to 1.45v for the CPU or should I go no higher than 1.40v? For the NB, will it be OK to push it to 1.30v or should I stay at 1.25v? Anything else I should be mindful of?

 

[doveman]

Actually it doesn't seem to say that it's 4+1 so perhaps I should e-mail them to check that before doing anything else http://www.asrock.com/mb/AMD/880GMHU3S3/?cat=Specifications

I'm also not clear what the Load Line Calibration setting should be on when Overlclocking, as the description's a bit confusing, so I'll see if they can tell me that as well.

HwInfo does report some things wrong at the moment, although I generally find if I report them they get fixed in the next update. It shows the GPU frequency wrong (966Mhz when it's 497.6Mhz), the GPU utilisation is showing 0% when GPU-Z shows 8-10%, NB frequency is correct at stock but when overclocked and set to 1920Mhz it shows 1600Mhz. So it may well not be as accurate as HwMonitor. On the other hand, it shows a lot more readings and it has the handy HwInfoMonitor sidebar gadget, so it's good for generally keeping an eye on things in day-to-day use but perhaps not ideal for testing overclocks.

HwMonitor doesn't appear to show any other temps. The Aux is obviously wrong and can be discounted (in HwInfoMonitor as well). Note that I had to crop the window but all it shows below Power is the HDDs.

Since my last message, I tried the IBT High test and something went funky on the second run as it took much longer but didn't fail but then I was having programs crash later so it's obviously not stable. This was with the CPU at 1.425v as well. On this test, I guess because it took longer, the Front fan did ramp up which helped push the temps back down a couple of degrees from 59/49 to about 57/47 but the test ended shortly after that. At 960Mhz, it automatically uses the 800Mhz 5-5-5-15 timings when in theory I suppose it should use the looser 1066Mhz timings so I don't know if that could be causing the problems.



I noticed that at stock, HwInfoMonitor shows the Core VID at a constant 1.40v, whilst CPU-Z (and HWMonitor) shows the Vcore at 1.368v rising to 1.4 something under load. However, when overclocked with the CPU at 1.425v, at idle HwInfoMonitor showed the Core VID at 1.38v whilst the Vcore was at 1.432v





As it's not stable at FSB 240/[email protected] I'm not sure what I should do just now. I could probably run the IBT Standard test overclocked to produce the screenshot you asked for as so far that's worked, even though other programs crash or I could run the test at stock clocks, although I'm not sure that would be very useful to you. Let me know anyway.

I did also try increasing the RAM from 960Mhz to 1237Mhz (which was the next available step) and despite manually setting the timings for 1333Mhz and then 1600Mhz, it wouldn't post reliably, so I don't seem to be able to increase it from 960Mhz. Certainly when not overclocking it runs fine at 1333Mhz, even though technically the specs only allow for 1066Mhz with 2 slots populated on a 4-slot board. I'm pretty sure I couldn't get it to work at 1600Mhz though.

I also tried increasing the FSB from 240 to 250, whilst keeping the NB and HT at 2000Mhz and even with 1.45v it wouldn't POST, so I don't think I'll be able to go any higher than 240. At 250, it set the RAM to 1000Mhz. Overclocking doesn't affect the onboard GPU frequency (checked with GPU-Z), which I guess is tied to the PCI-E 100Mhz setting, so I think that rules that out as the problem.

 

[Mandrake4565]

Vrm Phase count database info.

 

[doveman]

Thanks Mandrake4565,

My board is actually the 880GMH/U3S3 which doesn't appear on that list but maybe it's safe to assume it's similar enough to the 880GMH/USB3 to also have a 4+1 Phase VRM. I've e-mailed Asrock to check anyway.

It would be preferable to have a heatsink on the VRM of course but hopefully my top-down CPU cooler and rear exhaust fan will help keep the temps down a bit at least.

 

[doveman]

OK, Asrock have replied to say

"Power phase design as far as I can tell without checking with Taiwan is 4+1.

Unfortunately I do not have a list that tells me which load line calibration setting fits bets to which CPU. Please start with default settings and check if you have considerable drop of the Vcore under load. If you do, then change the setting and try again…"

So I guess I'll check for Vcore drop on Auto and if so, then try it on Disabled (which seems to be what the BIOS suggests for AM3 and then on Enabled if necessary. So far though, I don't think I've seen Vcore drop, only increase from 1.368v to 1.414v (at stock when using Cool'n'Quiet, overclocked without CnQ I think it stayed at a steady 1.432v).

 

[RGone]

LLC or Load Line Calibration is not an AMD specification until the FX processors. That is why most bioses say to disable LLC when using non-FX processors (non-AM3+).
RGone...

 

[doveman]

LLC or Load Line Calibration is not an AMD specification until the FX processors. That is why most bioses say to disable LLC when using non-FX processors (non-AM3+).
RGone...

Ah OK, thanks. Reading up about it, it seems there's a risk of overvolting with LLC if you're not careful, so it's probably best to disable it.

This damn thing still isn't reliably booting when overclocked.

I tried manually adjusting the timings for the auto-selected 960Mhz DDR3 and it wouldn't boot. The auto timings as far as I can tell from setting it to manual and looking at the column which appears to show them are:

5-5-5-15-2T-19-6-110-4-4-4-16 (these match the JEDEC 800Mhz timings except the fourth number, 15, which is 12 and the last five numbers are 64-3-3-3-12. I don't think it's even possible to set 64 manually as the lowest option is 90.

So I tried the 1066Mhz timings which are:

6-6-6-16-1T-25-8-90-4-4-4-16

and that wouldn't boot. It actually shows 86 in JEDEC but the closest option is 90. I tried 110 as well and it still wouldn't boot.

I then tried reducing the FSB from 240 to 230 for 3.22Ghz, with the NBx9=[email protected], HT 1840Mhz and RAM auto selected 920Mhz. This wouldn't boot with the CPU at 1.4 or 1.425v, even though it was booting OK before at FSB 240.

I then returned to stock settings but set the RAM to 1600Mhz, which seems to use timings of

9-9-9-24-2T-37-12-110-5-6-6-24

which match the JEDEC timings except for 110 should be 128 but that's not an option. I'm currently running with that but had an exception in Iron so I don't think it's stable and will have to drop back to 1333Mhz.

 

[doveman]

You should leave your ram at a known stable setting while OCing your CPU. Once you find stability with your CPU then start on your ram.

The point is I tried to reduce the FSB from 240/3.36 to 230/3.22 to see if that was more stable and couldn't find any RAM settings that would POST, even the Auto setting of 920Mhz, which is lower than the 960Mhz it was using at the higher FSB.

And at 240 FSB, it won't POST if I leave it on 960Mhz but try loosening the timings.

So it's currently unstable at 240/3.36, won't boot if I loosen the timings and won't boot if I reduce it to 230/3.22Ghz, so how can I find stability with my CPU?

EDIT: Another thing I've noticed is sometimes the BIOS shows the NB voltage (not CPU-NB) as 1.10v and other times 1.19v, as far as I can tell without my changing any settings. I think the only program that shows this voltage in Windows is Asrock Xtreme Tuning Utility and currently at stock settings it shows 1.19v (without a label):



So I presume that's what it should be using and not 1.10v, so maybe I need to set it to 1.19v manually when overclocking to ensure it uses that and not 1.10v?

 

[doveman]

There definitely seems to be something a bit dodgy about the NB (not CPU-NB) voltage setting on this board. At stock, in the BIOS it shows 1.19v but when overclocking, it shows 1.10v, even if I manually set it to 1.19v or even 1.30v.

I wasn't able to get it to POST properly, even just setting the FSB to 240 (or 230), the CPU to 1.45v, NB to 1.25v, reducing the multipliers to keep the the CPU, NB and HTT at stock clocks, and manually setting the 1066Mhz timings with the RAM at 960 (or 920Mhz), so I think I give up!

I'm working on underclocking it now instead, so at least I can keep it cooler and quieter most of the time. It seems stable at [email protected] (default 1.1v) and [email protected] (default 1.2v) so I just need to find what the lowest voltage it can run stably at 2.8Ghz is and then fine tune them. Is it sufficient to just reduce the VID bit by bit and run IBT Standard test each time or is there something else I need to do?

With a CPU fan target temp of 55c, running the IBT tests (5 runs) I get maximum temps of:

Standard: CPU 55, Core 45, MB 43 (CPU fan hits 1040 RPM, Idles at 800)
High: CPU 58, Core 48, MB 44 (CPU fan hits full 1480 RPM around 58c)
V. High: CPU 58, Core 48, MB 45

I did notice that Vcore dropped to 1.344v for the High and V.High tests. Is that something I need to do anything about? With the VID set to 1.35v at 2.8Ghz, Vcore shows 1.296v running IBT Standard.

 

[doveman]

Yeah, I found it was passing IBT but BSOD with OCCT (Linpack).

I found there was a fair bit of vdroop with LLC on Auto so I changed that to disabled, which fixed that and now I've got it stable at 2.8Ghz/1.3v, [email protected], [email protected] and [email protected].

Probably could have got it down a notch or two more if I had more time but I had to get it packed up.

 

[trents]

IBT doesnt stress as much as Prime95. I would stress with prime and yes just drop your voltage a bump or two at a time till you drop workers or BSOD. Then raise it a bump or two.

I disagree with that. I have found that IBT actually stresses slightly more and the temps are driven up a little higher than with Prime95. The offset is, however, that the default IBT is very short in duration and if you want to compare apples with apples and oranges with oranges then you need to configure the IBT to run longer.