Overclocking my Phenom II X6 1090T Black Edition

[PetrolHead]

First things first, this is what I'm working with:

CPU: AMD Phenom II X6 1090T BE
GPU: NVidia GTX 650
MB: ASRock 970M Pro3
RAM: 2 x 8 GB Kinston HyperX Fury Black DDR3 (official numbers 1600 MHz CL10, stock values 1333 MHz CL9 due to Phenom II)
PSU: Cooler Master V650S
HDD: Samsung 850 EVO SSD 250 GB
Case: BitFenix Prodigy M
CPU Cooler: Arctic Cooling Freezer A11
Fans: 2 x BitFenix Spectre 120 mm
OS: Windows 8.1 64-bit & Linux Mint 17.2 Cinnamon 64-bit

Although I've been using computers since 1986 this is the first computer I've tried to overclock, partially because this is my first own desktop computer since the early 00's. This particular machine I obtained a couple of months ago. I was on a pretty tight budget, but luckily a couple of friends wanted to get rid of some parts at a very reasonable price and so I ended up with the Phenom II and GTX 650. Of these two only the GPU could be considered a bottleneck for my needs, and even if I went out and got a GTX 980 I wouldn't have to worry about the CPU wasting its talents. So, there was really no need for me to try overclocking. However, it was something that I had wanted to try for a long time and since a modern BIOS/UEFI makes it very easy, I decided give it a go. However, there were some limitations from the start:

-I have a mini ATX case, which means there's less space for a proper CPU cooler.
-The current CPU cooler is no more efficient than the stock cooler, only quieter.
-Running Prime95 in max heat mode for about 20 minutes brings my socket temp to 68 degrees Celcius (see picture) even at stock speeds (70, if the room gets warm from running Prime95).

So, I read some of the articles written about overclocking a Phenom II and decided to follow these basic principles:

-Keep the voltages at stock values in order to keep temps down.
-Concentrate on overclocking the Turbo Core mode instead of all six cores.
-Don't get too greedy.

The justification for the second principle is as follows: The Turbo Core mode has soft limits (in other words doesn't switch off immediately when a fourth core is needed) and most programs do not stress all six cores. Furthermore, the voltage is higher in Turbo Core mode than in normal mode, which means that there's a good chance of braking 4 GHz just by increasing the multiplier. The Turbo Core mode also keeps temps down, since only some of the cores will be running under full load. In theory I should be able to reap higher rewards in everyday usage while at the same time keeping temperatures well below 50 C.

Before I registered onto this forum, I had already overclocked my computer as follows:

CPU: 3,2 GHz / 3,6 GHz (Turbo) -> 3,5 GHz / 4,1 GHz (Turbo)
RAM: 1333 MHz CL9 (9-9-9-25) -> 1600 MHz CL9 (9-9-9-25)
CPU-NB: 2000 MHz -> 2200 MHz
HT link: 2000 MHz -> 2200 MHz (I read somewhere that Phenom II's prefer both running at the same clock frequency)

The system could run Prime95 for an hour or two without issues, but then I suffered a couple of crashes while running GPU tests with Phoronix test suite (on Linux) and a crash or two while running 3DMark 11, so I started so suspect something was wrong. Also, I noticed my socket temperature was going up to 71,5 C, which is less than optimal. Now I'm starting from scratch and trying to do some more benchmarking and temperature checking at each step. I'll follow the following procedure and report my findings as I go along (and whenever I have the time to write):

1. OC RAM to 1600 MHz CL9 (9-9-9-25)
2. OC CPU-NB/HT to 2200 MHz
3. Take CPU-NB/HT back to 2000 MHz, OC CPU to 3,2 GHz/4,1 GHz (Turbo)
4. Bring CPU-NB/HT back up to 2200 MHz

Finally and if no problems arise before finishing this set, I'll test the system by running Prime95 overnight.

Wish me luck!

 

[trents]

I see a number of problems here with regard to overclocking the 1090T on the system you have.

1. It's a micro ATX motherboard with only a 4+1 power phase rating. If you are going to overclock that 6 core Thuban very much you really need a motherboard with 6+2 or 8+2 power phase. And all of those are full-size ATX form factor which also helps dissipate the heat. Your first big problem you have already alluded to and that is the CPUTIN (socket area/VRM) is already maxed out at 68c. Your only hope there is to put a spot fan to blow on that little heat sink just west of the socket. A spot fan blowing on the backside of the socket would also help.
2. Kingston Memory. It's a well-known problem. AMD CPUs in that generation just don't play well with Kingston RAM. Are you in the UK? So many of our posters using Kingston live in the UK. It must be very available and economical over there.
3. 16 gb of RAM. Puts more strain on the IMC. What programs are you running that you need that much RAM?
4. Trying to run your RAM at 1600 mhz. The IMC of the Thuban core CPUs is rated for 1333 mhz. Sometimes you can get them to run stable with the RAM at 1600 mhz but often they will not. And even if they will, often you must relax the timings in order to pull it off and so there is as performance hit that offsets the higher frequency.
5. Your CPU cooler is inadequate to support much overclocking on that CPU. But it may be enough since your motherboard will limit the overclock anyway.

 

[PetrolHead]

Thank you for your comment, very interesting. I did some digging and it seems I have loads more to read. However, based on what you said and what I've so far found out, my decision to concentrate on overclocking the Turbo Core using stock voltages was even better than I thought.

1. I am not looking for a very high overclock, so this is probably not a big problem. The Turbo Core mode is pretty easy on temps even when overclocked and from what I've seen in testing, running Prime95 on all six cores with stock frequencies stresses the system a lot more. I'll say more about the socket temp when I comment on the cooler.

2. I'm from Finland. I hadn't heard about problems with Kingston RAM and AMD CPUs so I didn't have any reason to avoid the brand. I can't really say much about them yet except that prior to overclocking I hadn't had any issues with the computer.

3. None at the moment. However, I was planning on possibly doing some scientific computing and playing around with virtual computers, so I wanted to have at least 2 GB of RAM per core available. I might in fact have to upgrade to 32 GB of RAM or more (the motherboard supports 64 GB) if I really get serious about scientific computing.

4. 1600 MHz CL9 hasn't so far proven to be an issue, but my stability testing has been limited. If it causes issues, I'll just go back to 1333 MHz CL 9. When I started overclocking and only bumped the Turbo Core frequency to 4,1 GHz, I experienced a BSOD while running 3DMark 11. I read that Thubans like to have all aspects of the system in balance and that RAM that's running too slowly compared to the CPU might cause BSODs. This is why I decided to OC the RAM (and the NB as well) - and I wanted to try to keep the CL9 since, as you said, 1600 MHz CL10 offers no signifigant gains over 1333 MHz CL9.

5. I know. This is why I don't want to increase the voltages, as that could increase the heat load significantly and the cooler is only meant for CPUs with 130 W TDP (as you know, the Phenom II has a 125 W TDP). However, I also realised a further possible complication after your comment: The stock cooler is of the top-down variety, while the A11 fan is mounted on the side of the heat sink. This means the A11 doesn't cool the motherboard at all like the stock cooler. This means I really shouldn't push the clock frequencies of all six cores past the stock values, unless I want to lower the lifetime of the motherboard. However, although ASRock specifies the use of top-down coolers for certain AMD CPUs, the Phenom II X6 1090T is not one of them. Thus, I should be relatively safe as long as I only OC the Turbo Core mode and make sure the socket temps don't start climbing if I decide to OC the CPU-NB as well. I don't want to resort to spot fans, but I am planning on replacing the case fans at some point. This should make the insides of the case slightly cooler even under heavy load and with luck this will help keep socket temperatures down.

 

[trents]

In my experience with stress testing I have found that 3DMark often ferrets out RAM instability issues that don 't show up with other stress testing tools such as Prime95. I don't know why but it does. Some on the forum pooh pooh that idea when I have proffered it but that is my experience. The standard around here for stability confirmation is passing the Prime95 blend test for 2 hr. I add to that the 3DMark run.

 

[PetrolHead]

Thanks for the tip, I'll definitely take that into consideration if the OC proves to be unstable. In addition to Prime95 I will also run Memtest86 to test the RAM. The thing with Prime95 is that one really should run it for a long time. I'm not convinced an hour or two is enough, except when trying to find the limits where the OC is definitely not stable. Yesterday I found out the combination of 4,1 GHz Turbo Core and 1600 MHz CL9 RAM was slightly unstable after all, which is why I've now taken the system down to 4,0 GHz Turbo Core and 1600 MHz CL9 RAM. The system seems to be stable again, so I'll proceed with the CPU-NB/HT overclock and see what happens.

 

[caddi daddi]

get over the spot fan issue and install one, if you plan on real computing you have no choice with that setup.
I have 9 rigs, all the rigs that I run on water have a spot fan on the vrm section with enough airflow to cover the north bridge also, so that's 9 rigs with spot fans.
if you push a rig over a few days you find out real quick that more airflow is always needed.

 

[PetrolHead]

The computing I might do will likely keep all six cores busy, in which case I'm practically running a stock system as the small CPU-NB bump and RAM OC do not seem to affect the socket temperature much - assuming of course the system is stable with those settings. Since I've no free fan connectors on my MB, I think I'll rather go and buy a new CPU cooler when I can afford it. One that's a) more efficient than the current one and b) of the top-down variety. This way I'll get better cooling for the CPU as well as for the parts around it. I'm also planning to replace the two 120 mm BitFenix Spectre case fans with one 120 mm Noctua NF-S12A FLX and one 140 mm NF-A14 FLX. That should increase the air flow through the case and, depending on which rpm I'll run them on, possibly also lower the noise levels.

 

[PetrolHead]

Okay, I've now run several benchmarks for each of the settings I've tried. The highest scores for each test are in bold italic. I'll move on to stability testing and post CPU-Z data etc. after I'm satisfied I've found a stable combination of settings. If I decide to try any new OC settings, I'll benchmark them as well.

NOTE1: I have only run each benchmark once! Since running a benchmark 3 times will likely give you 3 slightly different results, small differences should largely be ignored and the displayed values should be considered qualitative at best.

NOTE2: Some test results vary more than others! The first time I ran Cinebench on my stock system, I got an OpenGL result of a bit over 69 fps (I had to run it again, since I didn't include the single core test in the first run). Even though there seems to be a nice progression in the numbers shown below, they're essentially the same result. Also, the consistency of Disk Mark in Performance Test 8.0 is terrible! The score can easily vary more than 20% between different runs and the differences in Disk Mark points thus do not reflect the effects of the OC. Since the Disk Mark affects PassMark Rating, it should also be considered a bit iffy. Luckily the rest of the scores seem to be less prone to random variation.

NOTE3: I am not a 100% of the stability of any of the settings, but I do know that "3" is unstable. I decided to include it in the table regardless, since I had already run all the benchmarks.

… Stock 1 2 3 (***) 4 5

Cinebench R15
OpenGL 63,93* 67,72 69,15 71,11 70,77 71,62
CPU 472 483 484 478 483 486
CPU (Single core) 92 92 93 104 101 102

CPU-Z
Single thread 532 533 532 605 590 590
Multi thread 2228 2260 2259 2261 2265 2261

Geekbench 3.3
Single-core 1875 1895 1901 2093 2083 2091
Multi-core 8543 8654 8712 8718 8660 8726 PerformanceTest 8.0
PassMark Rating 2588 2571 2566 2693 2669 2632
CPU Mark 5740 5809 5866 5900 5928 5945
CPU (Single threaded) 1283 1286 1289 1465 1428 1430
2D Mark 493,1 472 477,9 506 503 506
3D Mark 1857 1860 1868 1857 1859 1867
Memory Mark 1198 1252 1291 1292 1284 1329
Disk Mark 3595 3399 2890/3457* 3904 3594 2807


1: RAM 1600 MHz CL9
2: RAM 1600 MHz CL9, FSB/HT 2200 MHz
3: RAM 1600 MHz CL9, CPU 3,2 GHz / 4,1 GHz ***
4: RAM 1600 MHz CL9, CPU 3,2 GHz / 4,0 GHz
5: RAM 1600 MHz CL9, CPU 3,2 GHz / 4,0 GHz, FSB/HT 2200 MHz

*First test run 69,xx.
**Ran the Disk Mark again as a standalone test. The PassMark Rating is based on the lower score.
***Unstable

 

[Johan45]

Best way to test stability/heat as you know is P95.

 

[PetrolHead]

I ran Prime95 blend test with 6 workers for 8 hours with no issues, so Prime95 seems to be relatively happy with the current setup. I would still need to run Turbo Core mode separately. However, I set RealBench 2.41 Stress Test to run for 2 hours with 16 GB of RAM earlier today and within 90 minutes it had encountered several result mismatches. Whether that's because not all 16 GB of RAM was available or something else, I've yet to figure out. The testing continues...

Edit: Btw, as you can see from the screen capture, the socket maxed out at 70 C. This was probably just a short spike, as the temp mostly seemed to hover around 67-68 C.

 

[PetrolHead]

I ran the blend test on just one worker today to see if Turbo Core mode was stable. No warnings or errors in four hours, so I guess it's at least reasonably stable in Prime95. RealBench continues to not be happy. Interestingly, it produces a result mismatch in about 10 minutes even after I returned the UEFI to stock settings. Unless the software's to blame, my stock setup might thus not be stable.

 

[Johan45]

Most likely it's the 16Gb of kingston ram as Trents alluded to previously. Usually the IMC in the thuban is fairly robust and "should" run 1600 fairly easily. Mine will go up to 2000 without a lot of fuss TBH but it requires more CPU_NB voltage. Keeping your CPU at stock try upping the CPU_NB volts and testing the ram again with realbench. I'm certain it'll use as much ram as you have available and P95 blend on standard mode only uses 2Gb I think. Not very much out of 16Gb any way. You can do a custom test with P95 and set the ram amount if you prefer that. Either way test it and when it fails add a bit to the CPU_NB.
BTW urbo only works if half the cores are idle. You're further ahead IMO just shutting it off.

 

[PetrolHead]

The fact that Turbo Core only works when (roughly, it's a soft limit as I have stated before) half of the cores are idle is exactly why I concentrated on overclocking it instead of all six cores. Most of everyday use will employ the Turbo Core mode and overclocking it does not produce any heat issues. The six cores are running at full speed outside benchmarking only occasionally and for games the Phenom II is not a bottleneck at all even at stock speeds - at least as long as I only use 1 GPU - so in the hunt for FPS it's in practice useless for me to OC anything except the GPU. Even OC'ing my GPU is of questionable use, but that's another story.

As I mentioned above, RealBench produces an error at stock settings as well. Thus, the problem shouldn't be the IMC's ability to handle 1600 MHz, since RAM is currently at 1333 MHz CL9. I'll have to test whether the amount of RAM I use has any effect.

 

[Johan45]

Kingston has a tendency to set the read/ write timings too tight for AMD setups from what I have seen anyway. So you either increaase CPU_NB volts or loosen those timings. They'll be further down the chain of timings and usually set at 4-6 for AMD

 

[trents]

Make sure the CR is set to 2T.

 

[PetrolHead]

Kingston has a tendency to set the read/ write timings too tight for AMD setups from what I have seen anyway. So you either increaase CPU_NB volts or loosen those timings. They'll be further down the chain of timings and usually set at 4-6 for AMD

Make sure the CR is set to 2T.

Thank you for the tips, Johan45 and trents! The CR has been in Auto mode so far, which has meant that it's been 1T for 1333 MHz CL9 and 2T for 1600 MHz CL10. I was going to look for an SPD table of some other RAM module that's supposed to be compatible to see where I should start. Turns out I've taken an unnecessary risk when buying RAM. The Kingston HyperX Fury series doesn't officially support AMD 9-series chipsets at all, which I probably hadn't even noticed before for some reason. Also, ASRock lists two HyperX 2 x 8 GB 1600 MHz kits and zero HyperX Fury kits from Kingston in their supported RAM module list for my motherboard. This I remember noticing back when I was choosing which parts to buy. I think in the end I made the decision to buy HyperX Fury based on someone having no problems with them and then just hoped for the best. The specs of my HyperX Fury modules and the supported HyperX modules are reasonably similar based on their spec sheets, but I know that when it comes to RAM it isn't always enough to even have modules of the same type.

The safest move would probably be to go out and buy RAM that's officially supported. The fact that I didn't notice any issues (at least none that I remember) prior to this OC project and benchmarking gives me hope, however. Loosening the timings might very well be all that's required to pass RealBench with no errors. Fingers crossed...

 

[PetrolHead]

I managed to retrace my steps concerning the decision to buy HyperX Fury modules. Kingston's own memory search lists the modules I have as compatible with ASRock 970 Pro3 (R2.0), which is basically the ATX-sized version of my mATX motherboard. This is why I chose to risk it even though ASRock hadn't tested those modules on the ASRock 970M Pro3.

 

[PetrolHead]

I checked the memory timings. The timings listed by CPU-Z are 9-9-9-25-33-1T, and the timings that CPU-Z doesn't list, but can be found in UEFI, are:

tWR: 10
tRFC: 110 (On Auto)
tRRD: 4
tWTR: 5
tRTP: 5
tFAW: 20

I've changed both tWTR and tRTP to 6 and increased CR to 2T, but RealBench still announces a result mismatch at 8 GB and 16 GB of RAM about 650 - 700 seconds into the stress test. I guess that's what it takes to finish one cycle. Should I also try changing tRRD?

I also checked the voltage. The official voltage for the RAM is 1.5 V and BIOS had something like (Edit 1.585 V. However, there's no voltage that corresponds to that in HWMonitor. All I have are:

CPU VCORE: 1.200 - 1.440 V (latter when in Turbo Core mode, otherwise less than 1.3)
VIN1: 0.192 - 0.200 V
AVCC: 3.296 - 3.328 V
3VCC: 3.296 - 3.312 V
VIN4: 0.488 - 0.512 V
VIN5: 1.672 - 1.680 V
VIN6: 1.808 - 1.824 V

I don't really know what these refer to, except for the Vcore. Is there anything that looks worrying? I haven't touched any of the voltages in UEFI.

Edit: I just tried running the RAM @ 1066 MHz CL7, which had no more luck in clearing RealBench stress test. I then also tried running the RAM using the timings listed for 1600 MHz in the SPD, but with the clock frequency set to 1333 MHz and 1066 MHz. Both failed to boot.

 

[Johan45]

I know that's not an ROG board but this BIOS guide may help you with setting up that ram. https://rog.asus.com/forum/showthread.php?2585-ASUS-Crosshair-V-Formula-BIOS-Guide-Overclocking

 

[PetrolHead]

I know that's not an ROG board but this BIOS guide may help you with setting up that ram. https://rog.asus.com/forum/showthread.php?2585-ASUS-Crosshair-V-Formula-BIOS-Guide-Overclocking

Thanks for the link. I haven't done more than browsed that guide yet, since I wanted to first test my RAM with default values. I had some issues with MemTest86 v6.2.0 and I'm now trying to see if the people at PassMark's forum could help. After those issues I tried running Windows Memory Diagnostic Tool and cleared 1 pass of the standard test with no issues. I then tried the extended test and it freezed at 21% - or so I thought. I then manually set the memory voltage to the correct value of 1.5 V (the motherboard used 1.585 V as default), since this trick had helped someone else with RAM issues. I also set the clock frequency to 1600 MHz and changed the timing values to the ones listed in the SPD table so that the modules would be running at "true" stock values. I then ran the WMDT's extended test again and then just left the computer running. As it turned out, it just takes several hours (more than two) for the counter to progress from 21% and after 26 hours I had cleared four passes with no reported errors. This would likely have been the outcome even with the old values. However, RealBench stress test is still not happy, so I'll wait to see what the PassMark people have to say and then I'll see if the ROG guide can help. I've also presented the issue at the ROG forum, but so far nobody's said anything.