ASUS A8V Deluxe Overclocking Guide -- Discussion

[Impulses]

Well at 2.4GHz and 1.65v it ran for at 'least 4 hours before it froze (and I had a video looping in the background as Prime ran and I did some heavy browsing so that test was a bit overboard)... but it was running hotter than I'd like. Basically 48-50C all the way thru even though temps got a degree or lower here tonight than my earlier test (where it ran 6 hrs w/o crashing at 2.3GHz and 1.50v).

Not sure what else to try anymore, think I may just stay at 2.3GHz and be happy... At 'least for now, I don't have the time to keep running so many stress tests when I need to get the machine back to normal use. I like the temp headroom I had at 2.3GHz anyway, 44C gives me a margin of error for the summer season or increased heat from the video card (even though it's got an NV5).

I've got one last question, it's something that just ocurred to me... What have you guys set the AGP/PCI speed to when OC'ing? Should it just be left at Auto or 66.6/33.3? I wonder if maybe it's crashing on me at higher speeds (HT of 267 versus 256) because the PCI lock is not enabling itself...

I've got it set manually to 66.6 so I was just curious if it should work that way. It could very well not be the lock's fault, from the way it managed to be a lot more stable with higher vcore I'd say that supports the fact that I'm just hitting the roof here with my cooling and room temps.

Anyway, 500MHz/28% OC ain't too shabby for silent air-cooling.


Edit: Went back and read your FAQ, wow, either this thing (and the amount of A8V info over here in general) has grown or I missed a lot of it back in Oct.-Nov. when I was doing my research.

Either way from what you've said a setting of 66.6/33.3 should be fine. Guess I've just hit my headroom, hrm. I suppose I could lower the multi and stress test with the 267 HTT (at a speed lower than 2.4GHz) but knowing whether it's the processor that's hitting the wall or the mobo (in it's way of handling the Winchester) doesnt' really solve anything.

Edit #2: So I couldn't let it go... Must get to at 'least 2.4GHz! Anyway, I'm gonna try some runs with memory timings on Auto which will set 'em pretty tight I think, I remember some people with Ballistix actually being more unstable with looser timings. I doubt I'll uncover anything since I'm not running it high either way, but it shouldn't take long to crash (w/o 1.65v+) if it doesn't solve anything.

Also I was reading the AGP aperture comments, I've been having mine set at 256mb due to something I had read before regarding my BFG 6800 OC but I'm wondering if this is a big mistake?

 

[Super Nade]

PCI AGP lock should be at 66.66/33.33 MHz. Do not leave it in AUTO!

The AGP Aperture for your VC should be left at 64Mb, anything more than that is pretty useless.

After you have reached you max OC, try this technique (originally proposed by Sentinal) to lower your VCore. According to him, starving your CPU for Volts would get you to lower the VCore eventually.

The gist of the process:
Download CPU Burn and turn off error correction. Set VCore to the lowest value at which you can boot into XP without erroring out. Run CPU Burn for about 4-5 hours and then Prime. Repeat till you get no errors on Prime. Keep your OC freq originally achieved throughout the process.

Worth a try!

Good Luck!

 

[Impulses]

That's just the thing, I never actually error out, the rig just goes from being perfectly fine at 2.3GHz and 1.5v to locking up at 2.4GHz (without even running Prime, lock-ups somewhat delayed with higher vcore). Haven't managed to see a single error yet, just sudden freeze ups which require cold reboots. If I was seeing errors I'd be happy to tag 2.3GHz as my max OC but it seems kinda odd it goes from fine to sudden lock-ups w/o a single error.

1.5v for 2.3GHz is fine but I doubt I can try and starve it at 2.4GHz since I can't even get it stabled out there.


Edit/Conclusion (for now): Man I give up... I tried changing the AGP aperture just on the off chance that it was that (it sure wasn't), tried tighter timings (sure doesn't seem like it was that), tried with and without 2.8 vdimm...

No matter what at 2.4GHz the rig is just awfully cranky, locking up suddenly without anything more than two web pages open. It's started to cause corruption errors that prompt ScanDisk while rebooting and Firefox files seem to have gotten corrupted, first it was lost cookies and now it lost it's entire configuration and it's back to install defaults.

I'm using the Promise (red) SATA connectors btw so I doubt data corruption is what's causing the lock-ups, it's merely a by-product of them (I think).

So I think I'm gonna stop messing with it, restore my backup image and do a full-on stress test along with the video card at 2.3GHz and stay there for now... The only time 2.4GHz was ever halfway stable and managed to Prime for a couple hours was when vcore was at 1.65v and that raised temps way higher than I want them, not gonna even bother trying to starve it down from there when 2.3GHz at 1.5v is silky smooth.

2.3GHz is halfway between what anyway, a 3500+ and a 3700/3800+? Not a bad jump I guess even if I couldn't make it to my target. /me consoles himself

 

[Cisco Kid]

Looking for suggestions since I started to tinker,wonder what you guys think about my results so far and if you can suggest some tips. I have yet to test via prime for stability but as it stands at least I am up n running as I just got formatted. I also relaize that obtaingin further speed is useless unless it is stable so priming will be next

Read through the A8V stickies and tips got a few questions.

Here are my mem timings for my 2x512 PC3500EL on my week 44 CBBFD 3000 winnie. Currently at 250 fsb. On air using a Zalman 7000 copper in a very cool office what do you think I should realistically push for? 260 fsb then prime for stability.

Am I understanding the divider correct that now running 250fsb x 9 = 2250/11 divider based on a 5:6 memclk ratio that my ram is now running 204? So realistically I would want to continue upping fsb to reach the rams stock speed of 217 is that correct and test for stability? I think I am on the right track correct?

I still have to test as I said and I will have to confirm temps. Since I just formatted,for now only have Asus Probe (no clue how reliable this is, LOL).

Currently running A8V deluxe v 2.0 on 1007 bios chip 2x36GB Raptors in raid 0 with an 80 GB Seagate 7200.7 slave powered by Enermax Cooler Giant 480w PSU in a Sonata case. Current board bios is latest 1009 from Asus. I am using the Promise controller as well with the VIA 4 in 1 455 drivers

Asus Probe shows as I type this:
CPU 27
Mobo 22

Mem CLK mode 5:3
DRAM over 4G - Disabled
Node - Disabled
Burst Length - 8
Bank interleaving - Disabled

Cas 2.5
TRC 9
TRFC Auto
TRCD 3
TWR Auto
TRWT Auto
Tras 7
TRP 3
TWCL Auto
AsyncLat - Auto
2T disabled

Vcore at 1.55
Vdimm 2.8
Vlink 2.6
HTT 1000
AGP/PCI set to 75/ whatever or should I be at 66/33

All feedback welcomed

CK

 

[Impulses]

You got the mem clock divider info correct, works exactly like that. Of 'course, you don't necessarily -have- to get the RAM up to stock speed but it -should- be able to go all the way to 217 with it's rated timings in a stable fashion, yea.

ASUS Probe seems pretty reliable on my end so far, temp-wise and reliability-wise (no crashes or weird behaviors from it). On the other hand... I just noticed the temps you've listed and they seem pretty low though plausible. It's 72F inside your office?

Should probably set your AGP/PCI lock to 66/33 and turn down the HTT to 800 or 600 if you're going to raise the FSB beyond 250. I'm running it at 600 with a FSB of 256 and no problems or performande degradation so far.

Something about the first few RAM settings you listed seems off but I can't put my finger on it, isn't Bank Interleaving enabled by default? Maybe someone that remembers them better off the top of their head will comment, I'll check mine otherwise, I may just be confused.

 

[DaButcha76]

nice post there impulses - saved me a lot of write-up time

 

[Super Nade]

Wow! Great detailed posts you guys.

Impulses:
Try keeping HTF to 1000MHz.

Looks like you have hit a wall just like me and falling short of my 2.6GHz goal
I faced the same data corruption problem and the exact same Firefox problem when I tried to push for for 2:1 with my now RMA'd PDP (Actually the modules were bad. Gave memtest errors at stock settings).

Remember, you are running a Winny and Winny's don't really like high VCore. This is whats happening with your system:-

Lockup errors you described are due to low VCore and possibly a XP data corruption as a consequence (try a system restore and see if it goes away), but remember,
the finiky on die memory controller doesn't like high VCore and you will see Memory related problems as you increase VCore. Hence, you really can't do too much at this point.

BTW*2.3 is pretty decent!

C.K:
-Your CPU Temp looks too good to be true! Could you double check with speedfan?

-Please update to the 1009 BIOS

-Enable Bank Interleaving (If you don't I think its done automatically but I'd say do it).

-If you face problems increase TRC to 10 (TRC=TRAS+TRP)

-What Video card? If you face problems i.e No Video, set AGP Voltage to 1.6V

-PCI/AGP lock at 66.66/33.33 MHz

-Keep HTF at 1000MHz. In my experience anything lower, surprisingly, causes instability!

Good Luck!

 

[Impulses]

Oh the data corruption and related errors go away as soon as I stop pushing it that hard, for sure. Maybe after extended burn-in it'll take 2.4GHz someday... In any case, you think it's worth it to test with memtest at 2.3GHz or should the RAM probably be ok if it doesn't fail Prime95 and others? It's not like I'm pushing it too hard since I've got the 533 (11) divider on and only recently did I tighten the timings to 2.5-2-2-8. It's doing 210MHz atm.

I restored my backup image to start a fresh test yesterday. I've had it running overnight now at 2.3GHz / 1.5v with Prime95, rhtdribl (video stesser), a looped MP3, mIRC, ASUS Probe, CPU-Z, and the vid card temp monitoring tab of the drivers open and it's holding strong so far... 7 hours so far, gonna leave it all the way to 'least 12. Proc temps haven't reached 50C and video card's stayed under 66C I think. Prime95 isn't making as much progress as when it's running alone, should I maybe close the MP3 and/or rthdribl for the last leg of the test?

P.S. On the topic of the HT, I think it was here I read some people had actually seen increased performance in certain tests with a lower HT (counter-intuitive but very documented)? OTOTH I've read plenty of cases where HTs above 1,000 were also perfectly stable although most people commonly say otherwise (of A64 OC'ing)...

Basically I've heard it both ways and everyone who's more stable at over 1,000 HT usually seems surprised. I guess I'm kinda confused, you think I should bump it back to the 800 (4) divider with my 256 FSB? That was the only thing I didn't try when pushing for 2.4, would've made for an HT of 1068, think it would've made any difference?

 

[Super Nade]

I would recommend you stick to the max HTF of 1000MHz. This board is weird, trust me . When you run all thoses test together, you may not be as efficient as running a component specific test.

I'd recommend you start with the CPU, stress test it for > 14 hrs; move to the RAM, test that for >14 hrs; move over to the VC and stress that out for >14 hrs. If individual components are stable then you can stress them out together. In most cases if temps are steady, you won't have a failure.

 

[Cisco Kid]

Well I had the time to prime

SN: I enabled bank interleaving as you mentioned above, currently I am on 1009 bios and I also left TRC at 9 and set the agp/pci lock to 66/33.

Prime failed on me after 3hrs 1 minute running blend test. I guess I should now prime again individual components. I lowered the fsb I had to 245 at

1000 HTT with the same memoru settings as I noted above
Vcore 1.55
Vdimm 2.8
Vlink 2.6
agp left at 1.5 I think for 6600GT

So in essence back to the beginning to find out where or what is the issue.
temps by speedfan 2hrs into the run were 40 for the cpu and 27 for the board.

Any thoughts

CK

 

[Super Nade]

CK:

You are on the right track. If you still have speedfan on, can you tell me if your VCore or your 12V line spiked?

Two things about P95:-

1.Not a very reliable Test as it has problems with Winnys. You might want to try super Pi or RightMark Memory Analyser as well to be very sure.

2.A failure after say 3 hours points to a sudden voltage/VCore spike pushing your OC over the edge.

Also, instead of lowering FSB, could you try loosening the timings in this order:-

2.5-3-3-10-1T
3-3-3-9-1T
3-3-3-10-1T
3-4-4-10-1T

Good Luck!

PS* I hope you isolated your CPU first by running a high CPU OC at a very low divider say < DDR333.

 

[Impulses]

Only problem with testing individually is the temps don't go up as high as they do when I test it all together...

The CPU idles anywhere from 42-44C and will go up to 45C at worst if I'm just running Prime but if I've got the video card cooking as well then it might go up to 47-48C at times (particulary in the middle of the day).

I'm gonna have to DL 3dmark or just install HL2 to finish testing the video card though since rthdribl doesn't force it into the performance 3D OC settings (so it's been stressed at 350/800 instead of 400/800).

The test that I left running for close to 14 hrs. was pretty succesful though. A looped MP3, rthdribl, Prime95, and mIRC @ 2.3GHz / 1.5v (rest in sig)... No crash or errors reported after 13 hours and 40-some minutes.

 

[CunningLinguist]

I have a question about dual channel memory with the A8V. I just put together a system with 1GB dual channel. I thought about maybe adding another GB in the future. I've read that if you run dual channel, you can only use 2 modules. Is this true, or would I be able to add another pair?

 

[Impulses]

To run in dual channel you must add matched pairs, i.e. if you wanted 2GB you'd need either four 512mb sticks (which can hamper your OC potential, some boards default to 2T with 4 sticks for instance) or two 1gb sticks (which aren't always available with the best timings but in time will be). This is the reason most people go with 1GB total right now (512mb x2) since the benefits of more aren't really evident with the vast majority of apps right now anyway.

 

[Super Nade]

If you add another gig of RAM this will happen:

1.You cannot run 1:1 with the HTT no matter what. It is a limitation imposed by the current crop of A64 CPU's.

2.You may have to enable DRAM over 4G remapping (I know the BIOS says it is useful only if the total RAM > 4Gb , but still..)

3.You may face stability issuses because there is additional overhead on the memory controller as there are 2 matched pairs to address instead of one.

4.Unless you are running a server or doing intensive multi-media reated work, 2G is pretty much useless. Heck even with 1G, the page file is hardly used!

 

[Cisco Kid]

CK:

You are on the right track. If you still have speedfan on, can you tell me if your VCore or your 12V line spiked?

Two things about P95:-

1.Not a very reliable Test as it has problems with Winnys. You might want to try super Pi or RightMark Memory Analyser as well to be very sure.

2.A failure after say 3 hours points to a sudden voltage/VCore spike pushing your OC over the edge.

Also, instead of lowering FSB, could you try loosening the timings in this order:-

2.5-3-3-10-1T
3-3-3-9-1T
3-3-3-10-1T
3-4-4-10-1T

Good Luck!

PS* I hope you isolated your CPU first by running a high CPU OC at a very low divider say < DDR333.

Thanks for reply, I have not had time to tinker but planning to Friday eve, I will try super Pi . How am I able to tell if the 12V rail spiked, unfortunately I closed P95 after it stopped when I returned. I have not also isolated the highest CPU overclock I can attain at DDR333 yet something I overlooked.

I am gonna run p95 again and also super pi, should p95 lock up again after 3 hrs as I have same settings still I will check via speedfan. So I am assuming speedfan should be left open while prime is running?

Next question how is 1T enabled or is it run automatically if 2T is disabled?
I want to achieve somewhere in the range of 250-260 FSB

CK

CK

 

[Super Nade]

CK:
1t is enabled if 2T is disabled. This is another unecessary cause for confusion with this motherboard.

You need to leave speedfan on as you Prime.
250/260 FSB is quite possible!

 

[CunningLinguist]

If you add another gig of RAM this will happen:

1.You cannot run 1:1 with the HTT no matter what. It is a limitation imposed by the current crop of A64 CPU's.

2.You may have to enable DRAM over 4G remapping (I know the BIOS says it is useful only if the total RAM > 4Gb , but still..)

3.You may face stability issuses because there is additional overhead on the memory controller as there are 2 matched pairs to address instead of one.

4.Unless you are running a server or doing intensive multi-media reated work, 2G is pretty much useless. Heck even with 1G, the page file is hardly used!

So, if I did buy another Gig, I would be better off running in single channel? I do plan on doing video editing/dvd authoring, etc...

 

[Super Nade]

You'd be better off running dual channel. If you plan on doing any intensive work, I recommend some high speed memory with tight timings. You may want to invest in some TCCD RAM.

 

[Impulses]

I've got an interesting update on my rig that I'm gonna go ahead and describe since it might be of use to others and it's something I had questioned from the start... My system had actually passed quite a number of stressful Prime tests (+rthdribl, etc.) so I was moving on to test the vid card OC, try out my recently DL'd (thru Steam) HL2, etc.

Then I started having sudden lock-ups in HL2 as well as CS:S, the sound kept looping in the back (though all distorted like a scratched record) usually. Had to force a cold reboot every time since the system was unresponsive... At first I blamed the vid card and I tried everything (re-locked the pipes, lowered the OC, etc.) but it kept doing it.

I figured it might be HL2 itself so I installed KOTOR and Splinter Cell 2... KOTOR locked in much the same way and SC gave me a BSOD upon quitting after the first time I played it (IRQD_LESS_THAN_EXPECTED or something to that effect). At this point I figured it might be something else so I went back to my original OC.

I looked up the BSOD error and one of the things that it could be attributed to was bad memory. I had actually never ran Memtest on my system so I DL'd, burned it to a CD, and off I went to test my memory... So far it looks like either the A64 or my board just really really don't like TRAS set at 8.

Don't ask me why... The Ballistix is rated for it, the internal SPD on it reads 8 in CPU-Z, Auto settings actually set it to 7... But 8 gave a consistent Memtest error on Test 6 even at stock speeds. I can post a SS of the exact error if someone's interested. So I've turned it up to 10 and I'm now running Memtest again, no error on Test 6 this time even though I've even got it OC'd to 218 (says 222 for some reason though).

Anyone got any insight into this? TRAS as far as I can tell is a pretty obscure timing, I've heard all kinds of stuff regarding it (certain chipsets liking it at different clks or performing better with it higher/lower). I know all of Anand's memory tests on the A64 platform (they used an NF3 though, not my K8T800pro) were done with it at 10:

http://www.anandtech.com/memory/showdoc.aspx?i=2226&p=9

I'm wondering now whether this may have been tied to my original failed attempts to get to 2.4GHz, I'm not quite sure how much leeway I lent to TRAS at the time. I'll update tomorrow after Memtest is done (and I re-check this was the lock-up culprit by playing some games).