Maximus Extreme + Zalman GT1000 tri-loop setup - picture heavy

[polar bear]

That's it baby I'm back in business!

http://service.futuremark.com/compare?3dm06=6023743

[polar bear]

Catalyst 8.2: 20025
Catalyst 8.3: 20016

http://service.futuremark.com/compare?3dm06=6023743

[polar bear]

great work man, reminds me of this http://67.90.82.13/forums/showthread...34877&page=312
I don’t know which gave the milky look the tubes or the coolant, but I wish you used transparent tubes with some colored coolant (green or red)

I'm using transparent tubes and only anticorro+H20. And it's only in CPU loop the water is whitish. Dunno why and for me it does not matter that much. I was considering using colour but so far have decided against it as I wanted a "clean" look and also have read stories about the dye forming residues etc degrading performance.

[Kilyin]

I thought one of your ATI cards were bad? Did you manage to fix it or just buy a new one?

[polar bear]

No no, the 3DMark score is using just one 3870X2. Still deciding whether I should RMA the faulty one.

EDIT: when I get a chance I will post a picture to show the broken card - guess I have to remove the EK waterblock at some point anyway (but phew mounting it was quite a bit of work hehe). Perhaps someone here has a suggestion as to how I can fix it myself.

[Kilyin]

Oh, 20k with one card... I really need to get rid of my 8800 GTX.

[polar bear]

Well seems my overclocking feast is coming to an end. Dont think this chip is capable of much more than 4050, but it's okay, I'm happy Didnt think I would break the 20k 3DMark06 barrier either so quickly but I did, once I resolved that catalyst AI issue.

One day I might just get a 45nm quad (QX9650 ) let's see been dreaming of this for months but it's got one hell of a price tag.

Tried to step up a little bit further to 466*9=4200 but got a BSOD during vista boot. Dont really wanna go higher than Vcore=1.6 and V_CPU_PLL=1.7 lol.

Well, to conclude, it seems the equipment upgrade has brought an extra 400MHz on the quad.

As for temperatures:

[At 3.6GHz using everest stress test]
CPU idle 56 to 48 = 14% cooler
CPU load 85 to 68 = 20% cooler
NB idle 41 to 36 = 12% cooler
NB load 47 to 37 = 21% cooler
SB idle 41 to 34 = 17% cooler
SB load 44 to 35 = 20% cooler

[GPU temps using 3DMark06 and reporting hottest core]
idle 60 to 34 = 43% cooler
load 83 to 45 = 46% cooler

[polar bear]

Oh, 20k with one card... I really need to get rid of my 8800 GTX.

If you are planning to replace your 8800GTX, seems a 9800GX2 is the way to go, at least judging by this thread.

On average, the GX2 is currently 40% faster than the 3870x2 at 1920 x 1200 res.

[polar bear]

Auch, seems I'm not sailing completely smooth yet as I have discovered YET another issue with the 3870X2 :

Been stability testing with 3DMark06 loops (prefer this over everest/OCCT/orthos as it tests the system as a whole including 3D performance/stability). Anyway what happens is into the 2nd or 3rd loop graphical glitches start to appear, e.g. horizontal lines and similar artifacts. A little while after this the screen goes black for a few secs, then 3DMark06 stops (at least system doesnt freeze this time). I can start another run if I want. It gives the following error messages (different message on 2 different occasions).

On the 2nd run (no reboot in between) the desktop resolution jumped back to 800x600 then the system froze as I tried to restore to 1280x1024 via CCC...

Also in both screenies one can see the vista notification area "display driver stopped responding" message just before it fades away.

Will post a thread about this in the video card section - I have not experienced it before. And btw it seems to occur with 8.2 as well as 8.3 drivers.

[polar bear]

I always thought those cases were really sharp. I like the lines of it, and thick AL construction. Really nice job, and great chip you have there.

congrats on a great build, very sharp.

andyc

Hey, thanx! Glad you like it. It's a thoroughly made case yes, very sturdy, very sleek. Nice clean lines all around, no gimmick, no BS. I've been very happy with it since the day I got it - it replaced a Tt Armor.

[polar bear]

Decided to try to solve my stability issues and first go for overall system stability - as this might be the root of the 3D instability and ATI driver error messages.

So just to demonstrate that system is not stable at all with some screenies.

First I roll back to BIOS 0904 - the "safe" bios if there's any such thing lol...

Then I change to 1:1 ratio for mem at auto timings ie FSB450 => mem 900. At the same time NB goes from 1,61 to 1,71 lol (been fine temp wise so feel I can do this - heck I'm desparate so if I fry the board at least I'll be done with it...) FSB termination goes up to 1,60 (last "yellow" setting before red lol). Leave DRAM at 2.06. SB on auto, same with SB 1.5V. FSB450*8. Vcore still on 1.6. CPU PLL 1.70.

Idle temps:
CPU 56
NB 33
SB 32

Now for some OCCT...

Test Completed - Stable ! but cannot access the system as it is completely unresponsive - this happened from around 51 mins (1 hr test)



Notice artifacts on taskbar as well as missing values in everest sidebar...



Then back to "stock" settings with FSB267*9=2400 but with voltages from above ie totally maxed out. Cannot really run 1:1 here (534 lol) so decide to go for 2:1 ie 1069.



Then volts back to stock. System is unresponsive towards the end of OCCT 1hr again (no screenie). At least system is dead stable at 2400 but some voltage tweaking compared to stock is necessary - oh well I dont think the board is faulty at least then...

Now for some FSB testing with multiplier at 6x. Set following voltages for this:

Vcore = 1.43125
CPU PLL = 1.60
NB = 1.73 (last yellow before red)
DRAM = 2.06
FSB term = 1.60 (last yellow before red)
SB and SB 1.5 = auto

First out is FSB 475 with DRAM 951 (1:1)

[polar bear]

Nope. Same thing, fails after 50 mins or so. Well, looks like I have narrowed this down to a FSB stability issue, guess we can agree on that. Problem is, I'm not comfortable with upping NB/FSB volts any higher than this.

One crazy thought - could it be that running everest in the background is interfering with the OCCT stress test? The only stable run so far (9h 41m) has been with only OCCT running - no other apps incl AVG suite, no internet connection. Will try another run with only OCCT. 1 hr is good enough initially before I shoot for 12 hrs finally when things are looking brighter.

[polar bear]

Well, seems that might have done it - 1hr stable, no BS this time around. Hence in the future I'll ditch that everest monitoring in the background - it's only complicating matters further.



As for FSB 500, system freezes during vista load. Tried with different combinations of mem (1002MHz, 1201) and maxing out on the volts (CPU PLL 1.70, NB 1.73, FSB 1.60) but no luck so far. Well, that's okay, as I probably wont be needing that high FSB with my Q6600 anyway as I dont think it's capable of much more than my existing 4050MHz OC.

So now I'm all set to increase multiplier to see if I can reach stability with higher CPU clocks...

First one out is FSB 333*9=3000MHz with DRAM=1066. Leaving Vcore at 1.60 for these tests initially, other volts as stated above.

[polar bear]

So far so good.



Next up is FSB 400*9=3600 with mem @ 961.

[polar bear]

Next up is FSB 450*9=4050 @ mem 900MHz

[polar bear]

That run was fairly short lived, this is what happened after maybe 5 secs lol:



Increasing Vcore from 1.60 to 1.65 and during post i get this: "CPU Over Voltage Error! Press F1 to Resume"

Pressing F1 and vista load starts (guess it was some sort of "warning" msg) and a new OCCT run crashes in same way.

Think I will stay below 1.6 on the Vcore, dont really have balls to go any higher, temp wise its not good either with more than 60C on idle... So let's see how high I can get using 1.6V in terms of OCCT 1hr stability, then 3DMark06 stability. Finally I will increase DRAM, hopefully to 1800, tighten the timings, reduce the voltages... perhaps after all I might end up with a decent system, who knows!

Next up FSB 425*9=3825 with mem @ 850.



Next shooting for FSB 435*9 = 3915 with mem @ 871

Same story here. Guess I could try out FSB 430 but feeling too lazy lol anyway I actually dont care about the extra 45MHz on the CPU.

So that's for max OCCT 1hr stable: FSB 425*9=3825. Will have to change mem settings though.

Now onto the highest benching clock. I'll go right for the "jugular" with 4150=461*9 with mem @ 1845 and 8-8-8-26-2T.

Right. Well that results in BSOD during vista load lol. Changing to DDR3 @ 1476 and crossing fingers. Nope. Increasing Vcore to 1.65. Nope. Running 1:1 mem timings ie 923. Nope. Vcore at 1.70 - seems to get a bit further during vista boot then reebots - probably due to excessive heat lol.

I know that I've been successful in getting to FSB 475 with NB 1.73 and FSB term 1.60 so no point changing these for a lower FSB.

So I will have to face it - 4050MHz is max for benching, could probably get little higher for vista boot and perhaps a 1M super-pi run though. For OCCT stability 3825MHz semms to be the absolute max for my Q6600 as I can't really raise the Vcore beyond 1.60 due to temperature limitations.

To optimize for benching then - set FSB 450*9=4050 with mem @ 1801 at 7-7-7-20-1T. Booting into vista with idle temps as follows:

CPU = 62 for hottest core
NB = 38
SB = 36

Now for 3DMark06.

[polar bear]

Well, due to tightened timings at 7-7-7-20-1T compared to 8-8-8-26-2T i was able to get a whoopping 91 extra 3DMarks /sarcasm off.

http://service.futuremark.com/compare?3dm06=6068636



Well, I am happy with this. Don't see any reason to tweak the timings any more. Now for some 3DMark loop testing. I will aim for 3 loops on all tests. That should do it.

...and it passes the 3-loop test. Now all that is left is to clean up the voltages. My system is finally stable!

[polar bear]

An extra 219 points from tightening the mem timings in Vantage

Not sure, but as I am a flight sim addict, I would guess PCMark reflects flight sim stability better than 3DMark as it is so CPU/mem/HDD intensive. 3D rendering plays a minor role, that's why I am quite happy with "only" 1x 3870X2 at the moment... Anyway stability is absolutely crucial as you dont want to spend 15-20 mins preparing for a flight, then around 30 mins doing pre-flight setup (boeing 767-300) then take off for a 5 hr flight say from Hawaii to LA and just 2 mins before touchdown, on final approach, your system crashes Anyway flight simming seems to be a bit of a niche in gaming but I'm loving it - also I hold a real world private pilot licence. Guess what I'm saying is that I'd much rather be soaring over photo-real Hawaiian island of O'ahu than running around photo-realistic jungle shooting down vietnamese soldiers lol

Oh, and even at 4050MHz quad-core with DDR3-1800 and a quad-raptor 10,000rpm raid array with the latest and greatest dual-core GFX card (save for the latest 9800GX2 from nvidia), it is STILL not possible to run FSX at highest settings at 1920x1200 and get decent performance Perhaps with a 5GHz phase-cooled QX9650 lol it's possible to get smooth fps lol.

So in that respect I guess FSX is sort of like the CPU system hog analogue to Crysis being a 3D hungry beast.

http://service.futuremark.com/compare?pcmv=64541

[polar bear]

[polar bear]

[polar bear]

Well, as I said I have been adjusting the volts - and below is what I will keep for 24/7 usage:

Ai Overclock Tuner : Manual
OC From CPU Level Up: Auto
OC From Memory Level Up: Auto
CPU Ratio Setting : 9.0
FSB Frequency : 450MHz
FSB Strap to North Bridge : 330
PCI-E Frequency: 110

DRAM Frequency : DDR3-1801MHz
DRAM Command Rate : 1T
DRAM Timing Control : Manual
Primary Info : 7-7-7-20-5-60-9-6
CAS# Latency : 7
RAS# to CAS# Delay : 7
RAS# Precharge : 7
RAS# Active Time: 20
RAS# to RAS# Delay: 5
Row Refresh Cycle Time: 60
Write Recovery Time: 9
Read to Precharge Time: 6
Secondary Info : 8-5-6-4-7-4-7
Read to Write Delay (S/D) : Auto
Write to Read Delay (S) : Auto
Write to Read Delay (D) : Auto
Read to Read Delay (S) : Auto
Read to Read Delay (D) : Auto
Write to Write Delay (S) : Auto
Write to Write Delay (D) : Auto
DRAM Static Read Control: Disabled
DRAM Dynamic Write Control: Disabled
Ai Clock Twister : Strong
AI Clock Skew for Channel A: Auto
AI Clock Skew for Channel B: Auto
Transaction Booster : Enabled
Boost level: 2

Voltages:
CPU Voltage : 1.5625
CPU PLL Voltage : 1.64
North Bridge Voltage : 1.61
DRAM Voltage : 2.06
FSB Termination Voltage : 1.54
South Bridge Voltage : Auto
Loadline Calibration : Enabled
CPU GTL Reference : 0.63x
North Bridge GTL Reference : 0.67x
DDR3 Channel A REF Voltage : Auto
DDR3 Channel B REF Voltage : Auto
DDR3 Controller REF Voltage : DDR3_REF
SB 1.5V Voltage : Auto

Advanced CPU Configuration
CPU Ratio Control : Manual
Ratio CMOS Setting : 9.0
C1E Suppport : Disabled
CPU TM Function : Disabled
Vanderpool Technology : Disabled
Execute Disable Bit : Disabled
Max CPUID Value Limit : Disabled

Temps: (see graphs below)
[Using Everest system stability test with all components selected for 10 mins]
CPU min 58 max 87 average 73.5
NB min 37 max 39 average 38.4
SB min 36 max 37 average 36.2

[GPU temps using 3DMark06 and reporting hottest core]
GPU min 37 max 47 average 42.5

Everest system stability test 10 min run: (4050MHz)



3DMark06 run with Everest monitoring in background: (4050MHz)

[polar bear]

Probably this is about as high as I can go for a Vista boot and 1M:

http://valid.x86-secret.com/show_oc.php?id=339690

[polar bear]

Well, I have reached another milestone here with this... Seems like I'm already worlds away from yesterdays woes when I wanted to give up. However, until I raised the Vcore a bit, I could not pass more than a couple of "return to proxicon" loops. Hence, new Vcore is: 1.575.

http://service.futuremark.com/compare?3dm06=6072311

[polar bear]

Well, folks, sorry for my rants hihi (dunno if anyone's even reading this thread anymore lol) but IMO I've come a long way. I never thought I would get this beast stable, BUT with a logical, systematic approach, I can finally say that I have achieved it.

EDIT: OCCT has just passed 12 hr stability at 425*9 This with nice voltages too. I am VERY happy with this result as it means I can finally start to USE this machine and not worry about the next BSOD lol. Here's the voltages I used to achieve this:

CPU Voltage : 1.575
CPU PLL Voltage : 1.64
North Bridge Voltage : 1.61
DRAM Voltage : 2.06
FSB Termination Voltage : 1.54
South Bridge Voltage : Auto
Loadline Calibration : Enabled
CPU GTL Reference : 0.63x
North Bridge GTL Reference : 0.67x
DDR3 Channel A REF Voltage : Auto
DDR3 Channel B REF Voltage : Auto
DDR3 Controller REF Voltage : DDR3_REF
SB 1.5V Voltage : Auto

It seems there is probably some more headroom for OCCT stability upwards towards 4.0 but from there it's a battlefield haha! It requires a massive voltage increase anyway - even at 1.65V it cannot tolerate more than a few secs of OCCT at those clocks. At 1.70 the system is overheating and auto shutdown occurs. So my interpretation is that the chip is just not capable of those levels. Which is completely acceptable of course, considering how it has performed so far. To reach OCCT / prime95 stability above 4.0GHz with a quad, I'm reasonably sure I will have to go for 45nm (although I have seen some lucky bastards bring a kentsfield to those heights, I guess those folks have had a very fine revision). Also of course is the cooling solution - mine's a pretty good one I think - EK-Supreme, DDC Ultra and PA120.3 is not anything to scoff at and will probably bring OC potential with a 45nm up towards 4.5 if not more judging by the result of others around here.

Well, a last screenie showing my efforts, then I'm off for some serious flight simming... or maybe *just*one*more*BIOS*adjustment*then*another*3DMark *run haha /polar bear out

[Anemic]

Really kewl build! I like how much time you put into documentation of the clocking process. Really sweet!