Phenom II X6 1090T + ASRock 970M Pro3

I thought I'd start a small thread concerning my rig and what I've managed to do to it. Here's what I'm running:

CPU: AMD Phenom II X6 1090T BE
GPU: Asus GeForce GTX 650 DirectCU 1 GB GDDR5
MB: ASRock 970M Pro3
RAM: 2 x 8 GB Kinston HyperX Fury Black (official numbers 1600 MHz CL10)
PSU: Cooler Master V650S
HDD: Samsung 850 EVO SSD 250 GB
Case: BitFenix Prodigy M
CPU Cooler: Noctua NH-C14S
Fans: 1 x BitFenix Spectre 120 mm, 1 x Noctua NF-A14 FLX 140 mm
OS: Windows 8.1 64-bit & Linux Mint 17.2 Cinnamon 64-bit

I've tried overclocking using only stock voltages and only now I'm starting to see what I can get out the CPU by increasing voltage. Using stock voltages is a good way to keep temperatures in check, but it naturally limits what speeds you can achieve.

I mostly use Prime95 for stability testing (custom runs, basically blend with more RAM in use), with the occasional RealBench stress test, various benchmarks, Unigine Heaven etc. to keep it interesting. Here's what is proven to be stable so far:

1. Stock voltages

CPU: 3500 MHz / 4000 MHz (Turbo Core) (I think this was Prime95 stable for 4 hours at least)
CPU-NB: 2400 MHz
HT: 2400 MHz
RAM: 1600 MHz CL9 (9-9-9-25-2T)

Edit: The 4000 MHz Turbo Core speed was tested only on the first core. However, based on other stability tests, it seems core number five is the weakest link in my CPUs chain. It may thus be that core number five is not stable at this speed on stock voltages, but I never experienced any issues running TC at that speed.

2. Non-stock voltages

Turbo Core mode disabled.

CPU: 3600 MHz @ 1.3375 V
+CPU-NB/HT: 2200 MHz @ stock voltage (not tested)
+CPU-NB/HT: 2400 MHz @ stock voltage (3 hours Prime95 stable), Geekbench 3.3.2: 1967 / 9765

CPU: 3700 MHz @ 1.375 V
+CPU-NB/HT: 2200 MHz @ stock voltage (2 hours Prime95 stable)
+CPU-NB/HT: 2400 MHz @ stock voltage (5 hours Prime95 stable), Geekbench 3.3.2: 1975 / 9940

CPU: 3800 MHz @ 1.4125 V
+CPU-NB/HT: 2200 MHz @ stock voltage (8 hours Prime95 stable)
+CPU-NB/HT: 2400 MHz @ stock voltage (4 hours Prime95 stable), Geekbench 3.3.2: 2043 / 10323

3. Notes

The thing about the higher voltages is that I'm reaching socket temperatures that seem relatively high considering the CPU stays below 50 C. With Vcore = 1.14125 V the socket temperature peaked at 68 C. I am not sure if a high socket temp is the sort of problem some people say it is, nor am I sure the reading is trustworthy. In any case, 70 C is often quoted as the safe limit and I'm only staying below it at the moment due to the fact that it's winter and the room temperature is lower than what it'll be most of the year.

Another thing I'm slightly worried about is the VRM. I'm not sure how much stress it's under and how much more it can take. After all, the 970M Pro3 only has a 4+1 power phase design. At the moment I mostly use HWMonitor to monitor the temperatures and even if there is a temperature sensor monitoring the VRM section, I'm probably not seeing the correct values. I'll see if any of the other available programs could give me more info.


Edited by PetrolHead - 23 Jan 2016 at 10:05am

wardog wrote: Heating of the substrate below the socket, and in turn more importantly the traces leading to and from laced throughout in the substrate below the socket, will change the resistance of said traces.

True, and it's easy to see how this could affect the stability of the system, since less current would be reaching the CPU. Thus, the user would need to increase voltage, which would lead to increased heat and also increased resistance... Apart from possible stability issues, this increase in resistance also affects the power draw, as you probably already know: A CPU can use several tens of watts more power at a higher temperature, even if clock speed and voltage are kept constant. But if one doesn't care about anything else than stability, does the temperature really matter?

I wish (AMD) motherboard manufacturers would specify a Tmax for the socket temperature, since they bother to measure it. Or maybe there's just no sense. Maybe the VRM or the CPU will put out smoke or start throttling before the socket components are in any actual danger.

I'd be better of running this like I couldn't afford replacing it. ;D

I bumped CPU-NB and HT to 2400 MHz on stock voltages. Didn't seem to have much impact on heatload, had a very slight impact on a few benchmarks (for the better) and seems to be stable as well (4 hours on P95).

I updated my BIOS and lost all my settings, so before setting everything up again, I decided to see what the "Load Optimized CPU OC" settings had to offer. Compared to my settings, the ~3.8 GHz load optimized setting had less Vcore, Vcpu-nb was something like 1.3 V, the bus speed was higher (250 MHz) and RAM speed was set at roughly 1000 MHz with CL 5, I think. I ran Geekbench (only once though) and it didn't seem to do any better than my current setup. Otherwise it was pretty even, but the memory performance was notably worse.

I then decided to try the 4000 MHz load optimized setting. Vcore was 1.45 V and everything else was pretty much the same as with 3.8 GHz as far as I can remember. Geekbench scores were marginally better than with my manual OC, partially because memory performance was again notably worse. I then decided to run Prime95 to see if the settings were stable. Were they? Not even close. It didn't take many seconds for several of the workers to produce an error, which would imply the Vcore was way too low. When I tried 4.0 GHz with a Vcore of 1.45 V I lost only one worker and it took slighlty longer, so it would seem the load optimized settings are so well optimized for the piece of silicon I own.

Of course this is not hugely surprising. This sort of "cookie cutter OC" rarely works quite as it should, especially if the OC is at all demanding. Still, it was interesting to see what sort of settings ASRock had chosen. Now my rig is back on my manual settings and running at 3.8 GHz. I might drop back to 3.5 GHz at stock voltages later on purely out of pity for the VRM section (more on this in another topic), but we'll see...

I didn't check what voltages the CPU reached using those load balanced settings, but using my manual settings, Vcore never seems to even reach 1.4 V even though the value set in BIOS is 1.4125 V. During a four hour P95 run, the max Vcore value was 1.384 V. The corresponding value was 1.344 V with 3.7 GHz and Vcore at 1.375 V in the BIOS, although I only tested this for two hours. In any case there seems to be a 0.041 V difference between the set value and the value detected by HWMonitor. Safe or not? Who knows. I might drop down to 3.7 GHz for 24/7 use just to be "safe".