Gigabyte X58A-UD3R Overclocking/Review (Part2)

ColonelCage

ColonelCage

ColonelCage

Associate
Joined
7 Apr 2010
Posts
89
How to Overclock with Gigabyte X58A-UD3R in an afternoon!

So this is going to be a run through of everything I did to overclock a new X58A-UD3R, with a D0 stepping i7 920 and 6GB of OCZ1600 Ram. It's really part 2 of the quick review I put up of the motherboard here http://forums.overclockers.co.uk/showthread.php?t=18140710 , both of which I might continue to add to. It should serve as a guide to overclocking particularly this board, but should be applicable to any Gigabyte X58, and hopefully be useful to overclocking i7's in general.

Memory Overclocking
The first task to take care of is to get the memory stable, at its rated speed and timings. This is important. When the memory is working and we start to overclock the CPU, we can adjust the CPU clocks and then adjust the memory multiplier up and down to suite the CPU overclock, example...

BCLK at 180 which will give us 3.6GHz on the CPU we use a memory divider that sets the memory to 1440Mhz if we know the memory is stable at 1600MHz then if there's any problems its most likely going to be with the CPU. Essentially we try to just focus on one thing at a time.

In this case the memory is OCZ3P1600LV2G which does not have XMP profiles which set the timings for us, so they were manually set at...

CAS 7
tRCD 7
tRP 7
tRAS 7
command Rate (CMD) 1
tRFC 88
everything else was left at Auto,
Vdimm was 1.64V
Performance Mode was set to Turbo.

The Memory multiplier was set to get 1600, and Uncore automatically adjusted to be twice this at 3600 which was the x24 multiplier. Remember Uncore must always be x2 memory.

This Passed Memtest and a blend test in Prime, without any increase to QPI/Vtt voltage, when I began to OC the CPU, QPI/Vtt did need an increase to 1.28V

Software
I have always used Prime95 for primary stability testing, but when I started working on overclocking the i7 I have decided LinX is very useful. LinX stresses the CPU more and increases the Temperature more than Prime95. You can run Prime95 for 4-6 hours before getting an Error that would show up in LinX in under 10 mins, so using LinX will greatly reduce the amount of time needed to Overclock, I did all this in one afternoon using this method.

Also used.
EasyTune6 for Voltage Readings, to check clocks etc.
RealTemp for temperatures which were kept under 75c.

Begin CPU Overclocking

Combined with the above memory settings,

1. BCLK 133 20x 2.66GHz,
CPU Vcore set at Auto in Bios sets Vcore to 1.225V actual reading was 1.152V Idle 1.168V Load, these are shown in EasyTune6 software.
Dram Voltage set to 1.64V actual reading was 1.616V, so this board under-volts CPU and Ram.
Uncore always 2x the Memory, and QPI x36 at 4.8
Everything else at Auto
EIST and C1E enabled
Turbo Disabled.

LinX, 3 runs, all memory in 13m 7s. PASS

2. Vcore was then set to Normal in Bios this sets Vcore to 1.1375V actual (0.928V Idle, 1.072 Load) Board is still showing some Vdroop, repeated above test Pass

3. BCLK 150 3.0GHz, RAM 1500MHz LinX, 3 runs all memory 11m 41s. PASS

4. BCLK 160 3.2GHz Ram 1600MHz, LinX, 3 runs all memory, stopped after 7m 19s upon Error. Fail

Turned on Load Line Calibration LLC to Level 2. actual Vcore now showing as (1.120V Idle, 1.136V Load)
LinX, 3 runs all memory in 10m 57s. Pass

5. BCLK 170 3.4GHz, Ram 8x (1360Mhz), Uncore 16x (2720)
LinX, 3 runs all memory in 10m 27s. Pass

6. BCLK 180 3.6GHz, Ram 1440MHz, LinX, 3 runs all memory in 9m 53s. Pass

7. BCLK 190 3.8GHz, Ram 1520MHz, LinX, 3 runs all memory in 9m 22s. Pass

8. BCLK 200 4.00GHz, Ram 1600MHz, Fail Vcore ++ (incremented by 2 steps in Bios), 1.150V Fail
QPI/Vtt increase to 1.22 Fail

9. BCLK back to 195 3.9GHz, Ram 1560Mhz, LinX, 3 runs all memory in 9m 8s. Pass

10. BCLK back at 200 Vcore++(1.1625V), Fail. Vcore++(1.175V), Fail. QPI/Vtt(1.24V),Fail
Vcore++(1.1875V), Fail. Vcore++(1.20V), QPI/Vtt(1.26V), LinX, 3 runs all memory in 8m 54s. Pass

Overclocking Conclusion
I can tell you from here the processor continues to scale like this. you need 8-10 incremental increases in Vcore and 1-2 increments to QPI/Vtt for every extra 10 on the BCLK.
I brought the CPU up to 4.2GHz. You might need to increase the CPU PLL voltage towards 1.88V and might need to increase IOH Core to 1.2V , CPU Clock Scew 300ps for stability.

The Stock Voltage Overclock
I have decided to stick with 3.8GHz as my 24/7 overclock and here is my argument for this. If you followed the above you will see that on Normal voltage with nothing other than LLC enabled and an increase to QPI/Vtt this CPU went right to 3.8GHz which is a huge jump from 2.66GHz.

Using the DES2 Dynamic Energy Saver software, you can get a reading of the power the CPU is using in Watts and here they are.

2.66GHz 68-70W

3.8GHz 98-101W

4.0GHz 130W

Yes to get that extra 200MHz requires 30Watts of extra power and if you push towards 4.3GHz you will soon be seeing 150W+, that's just the power the CPU is using! You also start to see a huge temperature increase, my CPU was already above 75c in RealTemp at 4.2GHz. So other than bragging rights is there a point to pushing the CPU this far and potentially shorting the lifespan of your components? The difference is almost negligible between 3.8 and 4.0GHz and in games things are completely GPU bound.

The X58A-UD3R is a very capable board but with only an 8 phase power circuit, I think pushing to far beyond 4.00GHz may be asking too much. I don't believe it has the cooling capability for this either, and at 4.2GHz there was a slight whining noise staring to come from I expect the power circuit.

LinX and Prime95 not Infallible
LinX and Prime95 both passed, but while doing some multitasking, uTorrent, Music playing in Media Player, A DVD ripping and also browsing using Explorer, the system Froze.
I expect this was the extra information passing along the QPI Links and the X58 chipset etc. This happened again when similarly multitasking and transferring files across USB.
Neither Pime95 or LinX runs really test this scenario, so are not infallible.

After incresing QPI/Vtt to 1.28V and IOH Core to 1.14V the system has been rock solid through many hours of gaming, encoding 20+ hours of 1080p video using 264 codec, and other multitasking.


Final Settings

BCLK 200 CPU Clock Ratio x19 3.8GHz
Memory Multiplier (SPD) x8 / 1600MHz
QPI Clock Ratio x36 / 7.2 Link Speed
Uncore Clock Ratio x16 / 3200
Load-Line Calibration (LLC) Level 2
Turbo Disabled
CPU Vcore 1.13750V (1.120V Idle, 1.136V Load actual reading in Easytune6)
QPI/Vtt 1.280V
Dram Voltage 1.640V
IOH Core 1.14V
Everything else Auto

BCLK was set to 200 with x19 clock ratio to allow the Memory to run at 1600. At this stage LinX was set for 20 runs and this was done several times without error. Prime95 was then run overnight and through the next morning without errors for 16 hours.


I hope this is helpfull to anyone attempting to overclock with the X58A-UD3R, it's a new enough board and also about the cheapest X58 around, so its probably going be a popular choice and so, I thought I'd share my results.

****Edit**** Pushing on to 4.0GHz, Load-Line Calibration Investigated.
I think most people want the 4.0GHz OC, nice round number and with 200BCLK you can run the Memory at 1600MHz. After Realising I actually needed QPI/Vtt at 1.28V for stability at 3.8GHz and 1.30V for 4.00GHz. I went back and redid the overclocks, and also investigated the effects of Load-line Calibration Level 1 and Level 2.

The original route I took was, using LLC at Level 2,

BCLK 200, x20, 4.0GHz, Ram 1600MHz.
Vcore 1.20V (1.184V Idle, 1.20V Load), LLC at Level 2, QPI/Vtt 1.30V, 127W PASS

BCLK 200, x20, 4.0GHz, Ram 1600MHz, Using LLC at Level 1, started with Vcore at...

Vcore(1.875V), Fail. Vcore++(1.20V), Fail. Vcore++(1.20V), Fail. Vcore++(1.2125V), Fail. Vcore++(1.225V), Fail. QPI/Vtt (1.30V), Fail. Vcore++(1.2375V), PASS.

Vcore 1.2375V (1.20V Idle, 1.20V Load), LLC at Level 1, QPI/Vtt 1.30V, 127W PASS

You can use LLC level 1, or2, but you still need to see 1.20V actual Vcore reading in Easytune6 under load to get it stable. There was no difference in Power consumption either. I also did some quick testing at the 3.8GHz overclock and had the same results.
 
Last edited:
Yes I probably should have mentioned that my D0 seems to be a good chip, 1.184V is still good, I'd imagine it keeps you under the 130W the CPU is rated at. Above 3.8GHz mine starts to need some voltage though. It needed 1.20V actual Vcore to reach 4.0GHz. Out of interest what did yours require?
 
Sounds like you both have good chips, or maybe mine just isn't great. Mine needs 1.21v for 3.8ghz, although I haven't had time to see if I can get that down to 1.2v. Also, I need 1.39v on the QPI/vtt (memory controller) for stability.

I found that LLC level 2 overvolts under load so I stick to level 1, which appears to be favoured elsewhere too.

Also, some other owners have said that they knocked 10C off of the NB temps by taking the blue Gigabyte plate off the heatsink. I haven't had chance to check this myself yet, but intend to.............once the decorating is finished! :rolleyes:
 
Last edited:
Hey, thanks for this guide I had been finding it difficult to keep the temps below 80c on IBT. I had my vcore at 1.225. This is what I got following your instructions.. :)

38stable-1.jpg


.. as you can see the highest temp reached was 72 and that was on just one core..

hmm I just realised the threads were on auto I normally pick 4, going to run again now :D

edit: same result with 4 threads selected.
 
Last edited:
No way, I was looking at that little blue plate thinking I should rip it off! It really looks like its obstructing the airflow, but then I was thinking maybe having the extra surface area helps or something, anyway its coming off Now for sure!
 
Davola said:
hmm I just realised the threads were on auto I normally pick 4, going to run again now :D

edit: same result with 4 threads selected.
Click to expand...

You should be running IBT on 8 threads. Although, have you got Hyperthreading disabled as your Task Manager page is only showing 4 threads?

Could you reduce the size of your screen shot a bit please?

ColonelCage said:
No way, I was looking at that little blue plate thinking I should rip it off! It really looks like its obstructing the airflow, but then I was thinking maybe having the extra surface area helps or something, anyway its coming off Now for sure!
Click to expand...

I've been told that it's held on with an adhesive strip so be careful not to damage the mounting clips or the heat sink itself when removing the plate. Post the difference in temps when you've done it please. :)
 
Yeah, I have HT disabled, core temp says 4 cores 4 threads that should have give it away :p

And I shrunk my image, you did say please after all :)
 
ColonelCage said:
Yes I probably should have mentioned that my D0 seems to be a good chip, 1.184V is still good, I'd imagine it keeps you under the 130W the CPU is rated at. Above 3.8GHz mine starts to need some voltage though. It needed 1.20V actual Vcore to reach 4.0GHz. Out of interest what did yours require?
Click to expand...

mine needs a big jump for 4ghz. need to put 1.28 through to get it stable. temps max out around 75ish running linx with a megahalems and 1 push fan.

when you say keeps me under 130w how do u mean? the tdp? core temp currently shows 147.3w running 3.8 not sure if its accurate tho.
 
Last edited:
Yeah sorry TDP is 130W which is the CPU Thermal Spec. and corresponds to Tcase_max of 67.9c according to Intel. This I beleive is why most people will say keep it under 70c, 75c max. You say 75ish so you're probably good, that's about all I'll push mine to. Your reading of 147.3W Electrical Power is same as mine(around 150W). My whole point was it took 50W extra to go form 3.8-4.0GHz which seemed a lot for very little gain.
 
I have found HT produces worse frames in some games than without it on and it makes the CPU run a little bit hotter so I figure why bother :p
 
ive just been messing with my 3.8gig oc and lowered the multi from 21 to 19 and increased bcllk to 200 as well as changing from llc level 2 to level 1.

anyway my oc is now showing a tdp of 130w as opposed to 147.3. wonder why? obviously im gonna need to stress test it again so im not sure if its 100% stable.
 
Last edited:
I'm keen to run a BCLK of 200 as well, so that the RAM can run at it's rated speed.

The tdp is probably lower as level 1 will allow the vcore to drop more than level 2.

Edit; just tried 200BCLK and tdp went to 130W. Probably the multiplier is how Core Temp calculates this.

Also, got my vcore down to 1.2v (@3.8ghz) and tested OK so far. Seems my chip needs a QPI/vtt voltage of 1.49v for 200 BLCK, but it didn't increase cpu temps though.

Edit again; IBT stable at 1.18v load (1.2v bios). Running Prime now.
 
Last edited:
OK no flames!

Just want to ask why no overclock on the memory ?

Also changing the multiplier and BLCK may give more options WRT to memory speed
I seem to be able to hit a better speed on the memory using 21/181
What are other advantages/disadvantages
 
Ok so we could argue about LLC all day, and I've seen many forums elsewhere that do. I found it helpful using Level 2 as it got me to 3.8GHz without an increase to Vcore. The Vdroop this board was showing was quite large under load and that is why I used it. My Vcore was set well below recommended max and should be safe. I would however caution against using it if you are starting to push the Vcore higher, if you set Vcore to 1.3V in Bios with Level 2 setting you would no doubt be putting way more voltage through it than you realise. I think I'll edit above to say this, thanks.
 
Using 21x181 keeps your QPI Link speed lower at 6.4GT/s and QPI/Vtt can be lower, where as at 200BCLK the link runs at 7.2GT/s, so this could allow you more room on the memory. The performance difference in QPI Link Speed won't be noticeable, but 200BCLK seems to be what to aim for if you want your memory to run at 1600, as you only have x6, x8, or x10 Memory Multipliers and these leave you way to low or way to high, x8 is just right. The higher Memory Multipliers seem to require more QPI/Vtt too.
 
Made some Edits above, Newer version of DES software, revised some of the power figures. Also pushed on to 4.00GHz stable LinX and Prime, and tried to investigate effects of LLC Levels.
 
out of interest colonal. what bios version are you using? ive just reverted back from f6b to f4 as it seemed my ocs required additional voltage.
 
You must log in or register to reply here.
Back
Top Bottom