AMD "Thuban" Core (Phenom II X6) XS Overclocking Charts

[FlanK3r]

not bad Bex ...1090T 1015EPMW, stock cooling, 4170 MHz wprime 1024M ,-)

[chrissnake]

CPU Model: AMD Phenom II X6 1075T
CPU Stepping: -
CPU Frequency: 3840 MHz
CPU vCore: 1.375V
CPU Multiplier: 15x
CPU Turbo: Disabled
CPU NB Speed: 2304 MHz
HT Ref Speed: 2304 MHz
RAM Speed: DDR2-800
RAM Timings: 5-5-5-15-20-2t
RAM Configuration: 4 GB (2 x 2 GB)
RAM vDIMM: 1.98V
Motherboard: GIGABYTE GA-MA770-UD3
Chipset/Socket: AM2+, 770 + SB700
Cooling: Water (TT BIgwater SE)
Temps: 23C Idle / 49C Load
Operating System: Windows 7 Ultimate
32/64-Bit: 64
Stable/Suicide/Untested: Stable



Uploaded with ImageShack.us

[I.nfraR.ed]

Found 1hr run of my 1010MPMW Thuban @ 4GHz 1.33V, real one should be around 1.34V, but I can't recall...

http://b.imagehost.org/0650/4GHz.png

According to this post the ambient was 21C.

[Roman2K]

Hi fellow Thuban overclockers! I just wanted to say thank you for this overclocking chart. This is very valuable information for beginners like me. I recently switched from a 980X / 1366 system to fully-watercooled 1090T / AM3 one and having no prior AM3 overclocking experience, I have a lot to catch up on. But so far, I am very pleasantly surprised by how much easier it is to OC and manage stable clocks than on the i7 processors (I have tried an 860, 875K, 920 and 980X). Much more predictable, in that from what I have seen until now, if it doesn't crash within an hour of Prime95 Small FFT / BOINC (Einstein@Home + Milkyway@home) then it won't crash 20 hours later, like it happened on the i7.

I'm sorry I'm not contributing much right now, but I will soon. After a whole afternoon of trial and error:
- Model: 1090T (CCBBE CB 1022BPMW)
- Clock: 4.01 GHz (200 x 20)
- vCore: 1.3375V
- Stress test: 1 hour of Prime95 Small FFT, and then a couple hours of BOINC (still running)
- Temperatures (with the CPU, motherboard and GPUs -- 5970 at stock -- in the same WC loop, all crunching in parallel):
--- CPU socket = 50°C
--- CPU cores = 39°C
--- GPUs = 47°C
--- NB = 41°C
--- SB = 41°C

Northbridge and RAM at their minimum so that they don't interfere (ridiculously low frequencies like 800 MHz and 400 MHz, respectively).

Also, I found that the Load Line Calibration setting has a significant impact on stability at a given vCore. For example, at 4 GHz (200 x 20), with LLC = Auto (off I guess) it still wasn't stable at 1.4125V even though I didn't see it fluctuate in CPU-Z, while the same vCore with LLC on it was stable (or at least lasted a lot longer under stress load) but was a also hotter, CPU socket jumped from about 44°C to 55°C. From there I was able to dial it down to 1.3375V (no less since it crashed within ~15 minutes at 1.325V).

I look forward to overclocking the CPU-northbridge, as from what I gathered it has a greater influence on general performance and snappiness than CPU frequency... And that's a new parameter to play with compared to the i7 .

I'm still not clear on the HT-link though. Should it be clocked the same as the CPU-NB? How does it influence performance?

[Particle]

LLC is sort of a double-edged sword regardless of if it's on an Intel or AMD board. What LLC does is overvolt to attempt to compensate for droop under load. Depending on the board, it might overshoot quite a bit. As a consequence though, your CPU will definitely be more stable. Just be careful that it isn't stable because it's being fed 1.6V when you've set it to 1.525V.

Yes, overclocking the CPU NB will improve overall system performance. This frequency determines your L3 cache speed and also impacts your memory controller. 2.6~3.0 GHz is a pretty standard overclock range for this parameter. It defaults to 2 GHz on all models.

Don't worry about the HT link. Just keep it as close to 2GHz as you can for stability reasons. This is the link between the northbridge on the chipset and the CPU. Since there isn't exactly a lot of bandwidth needed, you're fine with whatever. Not hardly a benefit to risking stability for the sake of pumping it up high.

A CPU doing 4GHz at 1.34V stable is sort of rare, so I'd very much advise you to check what you're actually getting with a DMM. It might be delivering quite a bit more than you expect due to LLC. Also be mindful that it's definitely still possible that you'd crash after 20 hours if 1 hour of small FFT is alright. No platform is going to be able to save you from extensive testing. In any case though, I'm glad you're having fun. I'm curious what prompted you to jump onto a 1090T from a 980X. By all rights, the 980X is faster.

[Roman2K]

@Particle
Thanks for such a comprehensive response! Greatly appreciated. You have cleared up major points I was wondering about.

LLC is sort of a double-edged sword regardless of if it's on an Intel or AMD board. What LLC does is overvolt to attempt to compensate for droop under load. Depending on the board, it might overshoot quite a bit. As a consequence though, your CPU will definitely be more stable. Just be careful that it isn't stable because it's being fed 1.6V when you've set it to 1.525V.

You're right. With Intel processors I could see the vCore fluctuate in CPU-Z, that's not the case anymore with this AMD but that doesn't mean it's actually constant. Your explanation prompted me to install ASUS's flaky-looking PC Probe II and it seems to show more accurate voltages as I see the vCore fluctuate with utilization.

I was testing 4200 MHz at 200 x 21, 1.4V with LLC on. When running Prime95 Small FFT, CPU-Z was showing a constant 1.4V while Probe II displayed 1.52V! Shutting down Prime95 brings it down to 1.40V with peaks at 1.45V.

So thanks for warning me about this!

Yes, overclocking the CPU NB will improve overall system performance. This frequency determines your L3 cache speed and also impacts your memory controller. 2.6~3.0 GHz is a pretty standard overclock range for this parameter. It defaults to 2 GHz on all models.

Ha, excellent. With the CPU back down to 4.0 GHz, I will be aiming at a 3.0 GHz CPU-NB then .

Don't worry about the HT link. Just keep it as close to 2GHz as you can for stability reasons. This is the link between the northbridge on the chipset and the CPU. Since there isn't exactly a lot of bandwidth needed, you're fine with whatever. Not hardly a benefit to risking stability for the sake of pumping it up high.

Perfect, that's one less thing to worry about.

A CPU doing 4GHz at 1.34V stable is sort of rare, so I'd very much advise you to check what you're actually getting with a DMM. It might be delivering quite a bit more than you expect due to LLC.

Given the behaviour described above, I will definitely pay attention to the actual tension values from now on.

Also be mindful that it's definitely still possible that you'd crash after 20 hours if 1 hour of small FFT is alright. No platform is going to be able to save you from extensive testing.

OK. That was too good to be true, I will see how it goes once I will have settled for "hours stable" clocks, testing those for a whole 24 hours.

In any case though, I'm glad you're having fun. I'm curious what prompted you to jump onto a 1090T from a 980X. By all rights, the 980X is faster.

I loved overclocking Intel processors, AMD ones are equally as fun, I have no regrets regarding the switch from a 980X. I switched because I wanted to sell the 980X while it was still worth the same amount of money I paid for it (second hand already), before it was too late. It's a beast that was worth trying out but its value is becoming volatile with Sandy Bridge and the upcomnig 2011 socket high-end processors, so it was risky to keep it. The money leftover was to go into fully watercooling my system (at the time I had a GPU-only loop and a leftover 560 to fit in a FT-02), so buying a used 920-950 would have been the obious option, only there existed no full-cover waterblock for my motherboard (X58A-UD3R), so I would have had to buy a new one too. Since there was an EK waterblock for both the Crosshair III and IV Formula, and the CH3F fitted my requirements (no glued-on USB3, cheap SATA6, unused crappy onboard audio, same amount of PCI-E lanes as 890FX) and I could get the corresponding EK waterblock for a bargain price, I was dead set on moving to a "cheap" CH3F + 1090T + waterblocks . (Sorry for the long response .)

[DexNfX]

Is it worth going to an 1100T from a 1090T? Does anyone know what's the major difference other than a higher stock multi?

[Roman2K]

Is it worth going to an 1100T from a 1090T? Does anyone know what's the major difference other than a higher stock multi?

Given the price difference for a +0.5 default multiplier increase, one would think the 1100T is binned higher than the 1090T but I have yet to see this confirmed.

[Particle]

If your 1090T is already doing alright, an 1100T probably isn't going to be a notable improvement (if any at all).

[DexNfX]

Guess I'll just wait to see what Bulldozer has to offer then.

[Roman2K]

After a time consuming trial and error, I finally achieved 4 Ghz stable for a night:

• CPU Model: AMD Phenom II X6 1090T BE
• CPU Stepping: CCBBE CB 1022BPMW
• CPU Frequency: 4000 MHz
• CPU vCore: 1.425V (LLC off)
• CPU Multiplier: 20x
• CPU Turbo: Disabled
• CPU NB Speed: 2800 MHz
• HT Ref Speed: 200 MHz
• RAM Speed: DDR3-1333
• RAM Timings: 5-5-5-15-1T
• RAM Configuration: 4 GB (2 x 2 GB)
• RAM vDIMM: 1.5635V
• Motherboard: ASUS Crosshair III Formula
• Chipset/Socket: AM3, 790FX + SB750
• Cooling: Watercooling (Black Ice SR-1 560 with CPU, GPU and MB in the loop)
• Temps: 36°C Load (cores as reported by Core Temp, haven't checked the socket temperature yet)
• Operating System: Windows 7
• 32/64-Bit: 64
• Stable/Suicide/Untested: Stable

Stable = Tested stable for almost 9 hours of BOINC (Einstein@Home) with Prime95 Small FFT running the the background, picking up BOINC's leftovers.



With the exact same stress test, vCore one notch lower (1.4125V), it crashed (BOINC's processes stopped and some Prime95 workers stopped) in the middle of the night yesterday. At 1.425V, it has survived the whole night. We'll see tonight if it's still running, and I will take the opportunity to report the CPU socket temperature.

As much as I like this constant 4 GHz overclock, I would have preferred to take advantage of the Turbo CORE feature, at something like 3.8 .. 4.2 GHz but even at 3.7 .. 4.1 GHz, it was unstable on a single active core at 1.45V so I have little hope.

I'm beginning to wonder if the mysterious CPU VDDA setting could have an inpact on high multipliers? Isn't it like the CPU PLL on Intel? If I remember correctly, increasing it helped with 30+ multipliers, maybe it would be the case on Thubans with 20+ multipliers. Has anyone experimented with it?

Also, I tried to reach the golden 3000 MHz CPU-NB clock with several HT Ref. speeds, without success, but in a weird way. After increasing the CPU-NB voltage several times, I was able to have it stable for about 1 hour of Prime95 Blend, but 1 time out of 4, it wouldn't post after rebooting (on purpose), telling me the overclock has failed. So 3000 MHz prevents it from booting 1 time out of 4 but once in Windows, it appears to be stable!? I would have thought if it can pass this long of Prime95, posting should never be a problem. Is 3000 a limit for the Crosshair III Formula?

[Particle]

Updated

[Roman2K]

Updated

Thanks .

[Roman2K]

We'll see tonight if it's still running, and I will take the opportunity to report the CPU socket temperature.

Coming back from work, my PC was still crunching for Einstein@Home and Prime95 had no stopped workers. So that's 23h stable. The CPU socket is at 52°C as reported by HWMonitor (left) while Core Temp indicates 36°C for the CPU cores (right):

[Manicdan]

Coming back from work, my PC was still crunching for Einstein@Home and Prime95 had no stopped workers. So that's 23h stable. The CPU socket is at 52°C as reported by HWMonitor (left) while Core Temp indicates 36°C for the CPU cores (right):

where did you get that widget to show the cpu usage? i have one like it but, just says intel, lol

[Roman2K]

where did you get that widget to show the cpu usage? i have one like it but, just says intel, lol

It's called Core Temp Gadget, available in the add-on section of the Core Temp website. It requires Core Temp to be running in the background, as it's where it pulls its data off of. Better looking, pastel colors (cores 1 though 6):

• FEFF99
• FECD66
• FF9966
• FF99FD
• 9ACCFF
• CCFE67

[felix_w]

It's called Core Temp Gadget, available in the add-on section of the Core Temp website. It requires Core Temp to be running in the background, as it's where it pulls its pulls data off of. Better looking, pastel colors (cores 1 though 6):

• FEFF99
• FECD66
• FF9966
• FF99FD
• 9ACCFF
• CCFE67

Excellent..

What was your CPU-NB frequency for this 24hr run ?

[Roman2K]

Excellent..

What was your CPU-NB frequency for this 24hr run ?

The CPU-NB was clocked at 2800 MHz (1.1375V). I can't get it to 3000 because of the behaviour described above :

Also, I tried to reach the golden 3000 MHz CPU-NB clock with several HT Ref. speeds, without success, but in a weird way. After increasing the CPU-NB voltage several times, I was able to have it stable for about 1 hour of Prime95 Blend, but 1 time out of 4, it wouldn't post after rebooting (on purpose), telling me the overclock has failed. So 3000 MHz prevents it from booting 1 time out of 4 but once in Windows, it appears to be stable!? I would have thought if it can pass this long of Prime95, posting should never be a problem. Is 3000 a limit for the Crosshair III Formula?

[felix_w]

How high did you go with CPU-NB Voltage ? 2800MHz for (stock i believe) 1.1375V is very good i guess...

[Roman2K]

How high did you go with CPU-NB Voltage ? 2800MHz for (stock i believe) 1.1375V is very good i guess...

At stock it was a little above 1V. Here's what I managed, each time for 1 hour of Prime95 Blend:

• 2400 (12 x 200): 1V
• 2600 (13 x 200): 1.0625V
• 2800 (14 x 200): 1.1375V
• 3000 (15 x 200): failure, even at 1.3V
• 3000 (12 x 250): 1.2125V*
• 2996 (14 x 214): 1.2V*

* random posting "OC failed" error, even after pushing to 1.3V

[Particle]

Once you find the wall, NB isn't going to want to budge much. It gets kind of picky, and sort of like you noticed more voltage doesn't seem to help much. If you sacrifice some core clock, you can usually get an extra notch out of NB. The same is true for maximum core clock by sacrificing a touch of NB. Most people try to walk the line and split the difference. Both my 1090T and 1100T don't like going above the 2800 notch.

[Daveburt714]

*** BUMP ***

Just because it deserves it, and I needed some info from this thread...

[felix_w]

Just got my 1100T .... 1105MPM is the batch...hope i take it stable with minimal effort to 4G....though only on M4A79 board...

[moorhen2]

Just got my 1100T .... 1105MPM is the batch...hope i take it stable with minimal effort to 4G....though only on M4A79 board...

Be interesting to see what the VID is for that chip.

[felix_w]

Auto is 1.4V...Coretemp reports 1.4V also...BIOS reports 1.392V at Auto...now tryin' to set the best i can for 30 min of experimenting on it....