A64 3000+ Venice Overclocking Help? (DFI Ultra-D)

Well, after much more testing, benching, burn-in, etc last night and this morning - I've got to a conclusion.

300x9 is desktop stable, but not much more. No matter how I tune my ram from there (and a 8hr pass in Memtest came up clean) it's still not Prime stable.

285x9 on the other hand lets me tune my ram just like it's advertised at 2-2-2-8 @ 3.2v plus I can also ramp up the divider to 5:6 166 meaning the ram itself is running at 233mhz Prime stable. For the loss of 140mhz I'm willing to take it. I'll leave it at the current 2.56Ghz for the time being and benchmark the snot out of it - then if/when it'll decide it wants to try for 300 stable - I'll be back with updates :)

Long story short, I think I've about monopolized enough time of the people in these forums - and hope that at least a few people can learn from my trial and error, day by day, blow by blow account of my A64 OC'ing attempts - from this thread :D

Thanks for everyone's help! :)
 
Lukano said:
- Athlon64 3000+ Venice @ 2.7GHz (399x9 @ 1.59v) w/ AC Freezer64 (Desktop/3dmark/Gaming Stable - Tuning for Prime Stable)
- OCZ 3200 VX Gold @ 2-2-2-8 3.2v 5:6 233mhz.
- DFI NF4 Ultra-D Motherboard
- MSI NX6600-GT PCI-e Video
- Fortron/Sparkle AX500 500w

that seems pretty high :p
 
Sweet, why dont you keep it at the speeds in your sig though? (399x9, amazing) :p

What kind of ram timings did you see at a 4:3, or even with no divider with 285? Might be worth Sandra benching it to see if youre actually better off with lower latency+speed, cos higher latency is cancelled out somewhat by increases in speed.

Also what kind of voltage does the DFI go up to on the VDimm? im really considering one of those boards and a san diego for a birthday present, but i think id need some better scaling ram. (Ballistix roxors, but only really at low speed, it really doesnt like 245 or more). What i really fancy doing is getting some VX and seeing what kind of latency i can get it down to at pc3200 speeds, but anyway, im rambling, well done.
 
Herulach said:
Sweet, why dont you keep it at the speeds in your sig though? (399x9, amazing) :p

What kind of ram timings did you see at a 4:3, or even with no divider with 285? Might be worth Sandra benching it to see if youre actually better off with lower latency+speed, cos higher latency is cancelled out somewhat by increases in speed.

Also what kind of voltage does the DFI go up to on the VDimm? im really considering one of those boards and a san diego for a birthday present, but i think id need some better scaling ram. (Ballistix roxors, but only really at low speed, it really doesnt like 245 or more). What i really fancy doing is getting some VX and seeing what kind of latency i can get it down to at pc3200 speeds, but anyway, im rambling, well done.

The DFI board does up to 4V vdimm. This board was actually designed with help from OCZ to make it a memory overclocking board, not necessarily a CPU overclocking board. To get 4V vdimm you have to set a jumper but the board will support it. It is a really nice board with a ton of BIOS ram timing options.
 
MadMikeSS said:
The DFI board does up to 4V vdimm. This board was actually designed with help from OCZ to make it a memory overclocking board, not necessarily a CPU overclocking board. To get 4V vdimm you have to set a jumper but the board will support it. It is a really nice board with a ton of BIOS ram timing options.

Quoted for truth, I love my Ultra-D.....best board I've ever seen. :cool:

Oh, and nice typo there Lukano.....399x9...hehe. :p
 
Wow.. . what a typo indeed. Fixed now to 300x9.

It's rock solid for gaming (at least for the hour or so I put it through this morning on 3dmark05, and another hour catching up on my world of warcraft auctions which I've been neglecting for the past week) - but it errors within 2 mins on Prime95 which is frustrating me to no end.
 
Herulach said:
Sweet, why dont you keep it at the speeds in your sig though? (399x9, amazing) :p

What kind of ram timings did you see at a 4:3, or even with no divider with 285? Might be worth Sandra benching it to see if youre actually better off with lower latency+speed, cos higher latency is cancelled out somewhat by increases in speed.

Also what kind of voltage does the DFI go up to on the VDimm? im really considering one of those boards and a san diego for a birthday present, but i think id need some better scaling ram. (Ballistix roxors, but only really at low speed, it really doesnt like 245 or more). What i really fancy doing is getting some VX and seeing what kind of latency i can get it down to at pc3200 speeds, but anyway, im rambling, well done.

I haven't really benched/tested the ram for speed as of yet. I was more looking for stable timings (or at least hitting the rated 2-3-3-8 2.6v and 2-2-2-8 3.2v) at both stock and 300 HTT. I did play with it a bit, typically keep 2-3-3-8 and ramping the HTT up enough to try to push the ram at 1:1 up to around 245mhz - which it did not like.

I'm assuming that the ram needs to mature a bit before I can start pushing speeds on it - although the more I think about it, the more I wonder if trying to overclock the ram (1:1 or 5:6) to run faster but at tighter timings might give me both an accurate idea of the ram's current threshold as well as act as a stress test / burn-in of sorts that would make it a little more tolerant at 300 HTT.

I'm going to be posting a new thread with some general questions about A64's / motherboards / chipsets / voltage / ram timings & speeds - I'll link to it once it's up. It's not quite relevant enough to warrant extending this thread outside of it's current scope.

edit - here it is ; http://hardforum.com/showthread.php?p=1027712045#post1027712045
 
Am I correct in my math when I factor 300HTT, 2:3 (133) memory divider should be 200mhz speed for the ram - and not the 192mhz I'm currently getting?

If so, then I'm wondering if the memory tables in the 414 and 510 DFI bios's that I've tried might be a little screwy for this ram, or am I just doing bad math?
 
I am so very very close here, I can almost taste it.

300HTT x 9 x 2-2-2-8 192mhz running 12 hrs on memtest without a single hiccup. That proves to me that it is NOT my ram, and all my cpu or motherboard limiting me.

So I decided to drop the voltage on chipset and ldt to see if perhaps it -was- the motherboard, but noticed there was no difference at all. That leads me to beleive it's the CPU that's got a wierd memory controller, or it's doing funky math.

It's only the torture test that's giving me errors - which I assume is the correct test for me to be running. Right? In-place FTT And Small FTT both error (blended too, being a combination of both).
 
i use in-place FTT to check if cpu is stable. i wouldnt consider it stable if it runs for more than 24 hours without errors.
 
peterhoang2002 said:
i use in-place FTT to check if cpu is stable. i wouldnt consider it stable if it runs for more than 24 hours without errors.

And it's not. 300x9 2-2-2-8 runs through memtest without errors, but the second I fire up Prime95 in-place ftt I get errors. Drop to 290 - still errors. 280 - still errors, 260 - still errors.

What in the hell.... I feel like I'm forgetting a vital step here but for the life of me I can't figure out what.
 
Have you tried reinstalling prime, i know you mentioned you reinstalled windows after it went corrupt, but if you only recovered that could be issue there. How weirdly pissed off an elated would you be then eh?
 
Herulach said:
Have you tried reinstalling prime, i know you mentioned you reinstalled windows after it went corrupt, but if you only recovered that could be issue there. How weirdly pissed off an elated would you be then eh?

That's a really good idea. I had 23.xx installed, then found 24.11 and just extracted it overtop of 23.xx in the /program files/prime95/ folder. I'll either uninstall/reinstall or try running prime95 24.11 from it's own folder after uninstalling 23.xx .
 
Stupid newb request of the day.....anyone got a link to 24.11?

I'm gonna start working on the stability side of things with my setup, doing it right....and I'm obviously gonna need a "working" version of Prime. ;)
 
cornelious0_0 said:
Stupid newb request of the day.....anyone got a link to 24.11?

I'm gonna start working on the stability side of things with my setup, doing it right....and I'm obviously gonna need a "working" version of Prime. ;)

I think it's http:/mersenne.org/p95v2411.zip or something like that. It's in my 3000+ venice thread somewhere, I'll find it after I get some coffee.

edit - found it - http://www.mersenne.org/gimps/p95v2411.zip
 
Wait, Lukano- are you running the board without the floppy power connector? I just received the same power supply, mobo, and CPU combo from Fedex, and the Fortron AX500 doesn't have any floppy connectors on it! It's recommended on DFI-street that you plug in all 4 power connectors on the board, and I'm surprised that you've o/ced that high without the floppy connector. Maybe it's not that important and I'm worrying over nothing...

EDIT: Nevermind, it does. I just didn't see it at first, and it wasn't listed in the manual. :eek:
 
Ozymandias said:
Wait, Lukano- are you running the board without the floppy power connector? I just received the same power supply, mobo, and CPU combo from Fedex, and the Fortron AX500 doesn't have any floppy connectors on it! It's recommended on DFI-street that you plug in all 4 power connectors on the board, and I'm surprised that you've o/ced that high without the floppy connector. Maybe it's not that important and I'm worrying over nothing...

EDIT: Nevermind, it does. I just didn't see it at first, and it wasn't listed in the manual. :eek:

Either there's multiple revision AX500's - or you got one that's missing one that SHOULD have one. My AX500 has one and only one FDD power plug, chained on a line with a single molex (which infact helps, as I can use that one line for both the HD and FD power plugs on the board).

I guess perhaps the FD connector is somehow drawing all the juice on that line, which means I can try an HD/molex plug from an un-used line to see if it helps... not holding my breath though.

edit - heh, guess you edited just as I hit quote :)
 
up the voltage if it isnt stable.

also, i like to try superpi 32m to test for stability/errors too. if you get errors while running 32M, then its spitting out bad math so u can either lower the HTT or up the voltage. Id up the voltage if it isnt running too hot.
 
Ok, I had superpi running for a little bit last night, but I didn't get as high as the 32m test. I'll try that - maybe it'll help me decide if it's a Prime software issue or a clock/cpu issue.

Also - I've been as high as 1.81v (and as low at 1.42+4%) with no noticeable improvement. I'm really concerned about overvolting the chip as I don't want to have to shell out another $300+CDN for a new one (I'd be going 3200+ venice or 3700+ SD if I were to, but that's money I don't have).

I've got a few idea's to try tonight. These include;

- Try higher and lower HTT multipliers just for shits and giggles.
- Try SuperPi 32m test to watch for errors.
- Try using seperate lines from the power supply for the FD and HD (Molex) power plugs.
- Try uninstalling Prime95 and then running Prime95 24.11 from it's own new folder (no installer for it as it's still beta).
- Try reflashing to the 510-1 and/or 510-2 bios and reset EVERYTHING back to default again and try clocking back up. There's a chance (albeit slim) that I changed something somewhere and forgot about it - and/or another bad bios flash (unlikely).
- Try dropping ram divider down to 100 - again for shits and giggles.
 
This thread is getting encyclopedic, Its incredibly useful though. Im so tempted by a 3800 and that DFI board its untrue, you bastard :D
 
Herulach said:
This thread is getting encyclopedic, Its incredibly useful though. Im so tempted by a 3800 and that DFI board its untrue, you bastard :D

That's pretty much the only reason I keep posting interim updates and all/most of my questions in this thread. If it can be a tutorial or learning experience for someone starting in the same position I was in two weeks ago - fantastic. :)
 
So I've tried the following thus far with no fix to my woes in sigh;


Lukano's Bright Idea #1 said:
- Try higher and lower HTT multipliers just for shits and giggles.

No luck, 4x and 5x freaked out and BSOD'd and 2x was about the same as 3x (albeit of course slower).


Lukano's Bright Idea #2 said:
- Try SuperPi 32m test to watch for errors.

So far on Loop 9 (11m00s) with no errors. My question is if I'm able to due a 32m run of SuperPi with no errors, but Prime95 cacks out on me - what should that be telling us? I don't quite understand what's going on.


Lukano's Bright Idea #3 said:
- Try using seperate lines from the power supply for the FD and HD (Molex) power plugs.

Haven't tried this yet, but I wasn't holding out much hope as I haven't been given indication that there's no enough juice.


Lukano's Bright Idea #4 said:
- Try uninstalling Prime95 and then running Prime95 24.11 from it's own new folder (no installer for it as it's still beta).

No dice. No change. Damnit all anyways.


Lukano's Bright Idea #5 said:
- Try reflashing to the 510-1 and/or 510-2 bios and reset EVERYTHING back to default again and try clocking back up. There's a chance (albeit slim) that I changed something somewhere and forgot about it - and/or another bad bios flash (unlikely).

This is my "To Do" big one for tonight. I'm bloody well going to start with 510-3 (doesn't make sense, but then again neither does my current situation - at least not to me) and then 510-1, and finally try reverting to 510-2 (currently running) if all else fails. I will not load anything from CMOS reloaded, and I will custom tune every frigging option in the book if I have to.


Lukano's Bright Idea #6 said:
- Try dropping ram divider down to 100 - again for shits and giggles.

Believe it or not, my machine REFUSES to get past the windows splash screen unless I'm at 2:3 (133) or 5:6 (166). Everything else doesn't even get to DMI verification (or can't find blah.sys before windows loads) or various other PFI / IRQ related BSOD's if it does get into the splash screen.

I'm running out of idea's here folks. L'il help?!
 
got a question for you, you running at 1T or 2T?

if its at 1T, this maybe why you can run at other dividers. i had the same problem with my old ram; it wouldnt run on certain dividers at 1T.

I received my ocz ram today. ill test it out and let you know if i have the same probs as you.
 
peterhoang2002 said:
got a question for you, you running at 1T or 2T?

if its at 1T, this maybe why you can run at other dividers. i had the same problem with my old ram; it wouldnt run on certain dividers at 1T.

I received my ocz ram today. ill test it out and let you know if i have the same probs as you.

Yes, infact it is 1T.

And might I add;


YYYAAAAAAARRRRRRGGGGGGGGGHHHHHHHHHHHH!!!!!!!!!!!!!!!!

not meant for you peter, sorry!

Just venting my frustration on the fact that I've reflashed and re-setup the bios 5 frickin times now, tried everything but the power plugs on my "To Do" list - and still no frippin frappin luck.
 
go for 2T, see if that changes anything. my new sticks arent going into windows at 1T either right now, but with 2T at least i got it running 215mhz 2-2-2-5 so far passing memtest.

yea i know its a bit slower latency wise too, but i guess im gonna have to accept that its not gonna do 1T.
 
peterhoang2002 said:
go for 2T, see if that changes anything. my new sticks arent going into windows at 1T either right now, but with 2T at least i got it running 215mhz 2-2-2-5 so far passing memtest.

yea i know its a bit slower latency wise too, but i guess im gonna have to accept that its not gonna do 1T.

Sry to hear that, but it is something that's worth trying Lukano...
 
wow look at this "value ram". i got it clocked at 245mhz 2-2-2-5 @ 3.6v and running memtest with no errors so far:

RSCN0766.jpg


edit: ok, photobucket was being dumb and it resized my picture for no reason...
 
2T did nothing.

This is really driving me bonkers. All indications point to the fact that it's NOT the ram. Plus the fact that the symptoms persist no matter if I use 260 HTT, 280, 290 or 300 - so it has me thinking it's something else, and I can't for the life of me figure out what.

Going to sleep on it and take a fresh stab at it tomorrow. Any and all tips are welcome :)

edit - you know what's really wierd. With the exception of stock speed 200 HTT, things appear MORE stable at 2-2-2-8 or 2-3-3-8 - and less stable if I try for 2.5-3-3-8 or 3.0-3-3-8 (or 4-4-10, etc).
 
Quick question bro.....what's the stepping on your chip it's not my cpu itself but the mem controller on this thing that's bunk i think...cus with the ram on the 133MHz divider the chip is 32M stable at 290x9...but 1:1 it fails instantly.....whereas the ram is capable of 320+.....so i dunno, I'm doing what I can with this thing, but ah well.
 
Quick question bro.....what's the stepping on your chip it's not my cpu itself but the mem controller on this thing that's bunk i think...cus with the ram on the 133MHz divider the chip is 32M stable at 290x9...but 1:1 it fails instantly.....whereas the ram is capable of 320+.....so i dunno, I'm doing what I can with this thing, but ah well.


You should check to verify this. I just got the board and CPU yesterday, stuck HTT at 300x3, put a 2:3 divider on because my RAM sucks, bumped up voltage to 1.55v, and voila! 2.7GHz that's 8 hours Prime stable. So maybe it's your memory controller. Have you tried the 3/16/05 BIOS? It's what RGone and Angry_Games run in all of their boards over at DFI. That's what I'm running right now, so I could get RAM slots 1&3 working, because I've got 3 sticks of RAM- 2x 256, 1x 512. Too bad the board doesn't support mixed RAM sizes, so I've got to run with half the RAM that I had before.
 
cornelious0_0 said:
Quick question bro.....what's the stepping on your chip it's not my cpu itself but the mem controller on this thing that's bunk i think...cus with the ram on the 133MHz divider the chip is 32M stable at 290x9...but 1:1 it fails instantly.....whereas the ram is capable of 320+.....so i dunno, I'm doing what I can with this thing, but ah well.

CBBLE 0516 - which yes, I have heard rumors that the memory controller is junk on. But on the same note, I've heard the same rumors about 0512's, 0514's and LBBLE's. So there's really no defenite proof that's kicking around to really make me worry.
 
Ozymandias said:
You should check to verify this. I just got the board and CPU yesterday, stuck HTT at 300x3, put a 2:3 divider on because my RAM sucks, bumped up voltage to 1.55v, and voila! 2.7GHz that's 8 hours Prime stable. So maybe it's your memory controller. Have you tried the 3/16/05 BIOS? It's what RGone and Angry_Games run in all of their boards over at DFI. That's what I'm running right now, so I could get RAM slots 1&3 working, because I've got 3 sticks of RAM- 2x 256, 1x 512. Too bad the board doesn't support mixed RAM sizes, so I've got to run with half the RAM that I had before.

I've got nothing better to do so I'll give 316 a try. I had thought all -2's (414-2 and 510-2) were spinpoffs of the same memory table, but it can't hurt to try.
 
Lukano said:
I've got nothing better to do so I'll give 316 a try. I had thought all -2's (414-2 and 510-2) were spinpoffs of the same memory table, but it can't hurt to try.

I'm starting some VERY thourough testing this morning......starting from 205htt, going as high as I can.....5htt at a time, passing SuperPi 32M each time. I started with 510-1, and each time I can't pass SuperPi I'll go through any/all BIOS' I haven't yet tried and in both memory slot configurations, to make sure that I'm hitting the max clocks I can without raising any voltages. After I go through this lengthy process at 1:1 I'm going to give 'er a go again, but with the 9/10 memory divider. The reason I want to is for comparison...I have had a hell of a time getting 32M to finish at 290x9 and up....but as soon as I set in the 9/10 divider it flew right through.....

I wanna see what my max clocks are (with tweaked ram settings/timings) and do a bunch of benchies...and then find my max at 9/10 on the ram...cus even though I'm taking a bandwidth hit (not a HUGE one)...if I can clock considerably higher with the (slightly) slower ram and the results are in favor of that score, then I'll settle with that.

If you have time I suggest doing as much of the same thing as you can...but it's up to you.

The next major purchase I make is going to be 2x512MB sticks of Mushkin PC4000 Redline or 2x512MB of GSkill PC4800LA....and after that will be the r520....so I'll be "stuck" with this CPU for awhile while I'm saving/waiting to pickup an FX-57.....so I figured I might as well use that time and get the most I possibly can out of this setup for the time being. ;)
 
cornelious0_0 said:
I'm starting some VERY thourough testing this morning......starting from 205htt, going as high as I can.....5htt at a time, passing SuperPi 32M each time. I started with 510-1, and each time I can't pass SuperPi I'll go through any/all BIOS' I haven't yet tried and in both memory slot configurations, to make sure that I'm hitting the max clocks I can without raising any voltages. After I go through this lengthy process at 1:1 I'm going to give 'er a go again, but with the 9/10 memory divider. The reason I want to is for comparison...I have had a hell of a time getting 32M to finish at 290x9 and up....but as soon as I set in the 9/10 divider it flew right through.....

I wanna see what my max clocks are (with tweaked ram settings/timings) and do a bunch of benchies...and then find my max at 9/10 on the ram...cus even though I'm taking a bandwidth hit (not a HUGE one)...if I can clock considerably higher with the (slightly) slower ram and the results are in favor of that score, then I'll settle with that.

If you have time I suggest doing as much of the same thing as you can...but it's up to you.

The next major purchase I make is going to be 2x512MB sticks of Mushkin PC4000 Redline or 2x512MB of GSkill PC4800LA....and after that will be the r520....so I'll be "stuck" with this CPU for awhile while I'm saving/waiting to pickup an FX-57.....so I figured I might as well use that time and get the most I possibly can out of this setup for the time being. ;)

I doubt I'll have the time today, but if not then I've got a handful of days that I can run through some really detailed stuff next week.

I really need to go and bench out my ram to see what it's max is, because for the life of me nothing but a divider resulting in 200mhz or less will work, no matter my HTT - which is just stupid. Doesn't make sense to me.

BUT - on that note, I decided to start using ClockGen from windows and dropping HTT 1-2mhz at a time until I saw something that looked stable in Prime.

I only had to go down 6 notches to 294mhz to get there. I've been playing WoW (catching up on auctions) while I left Prime running in the background and managed to run about 20minutes now of assorted tests without an error. I know it's not officially prime stable, but it's sure as hell better than what I had.

My only beef is the @*#%&!!! ram divider. 2:3 is all it'll work at with the current settings, and that results in 189mhz ram at 2-2-2-8.

I'm going to let this run nonstop today while I do some errands. Part of me hopes the memory controller just needs a bit of burn-in time at its peak to open up a bit. I really hope so anyways.
 
Whoa now! Hold the phone!

294x9 2646.3Mhz
5:6 (166) ram divider for 240.6mhz @ 3.4v 2-5-5-11
1.66v cpu voltage
1.5 LDT voltage
1.8 Chipset voltage

1 1/4 hrs prime stable. Going out for a few hours, leaving it running.

It was an absolute fluke that I stumbled on those ram timings while running it through 5:6 divider in memtest. Had everything but the RAS/CAS Delay and CL timings set to auto and it picked the remainder of the values. I then ran memtest stable for a few passes, set manually in the bios and ran it into windows for Prime.

Be back in a few hours for an update.

edit - I'm fricking cursed. No sooner did I post this than I got a BSOD (not a prime error though).

edit 2 - gave it a bit more juice (vcore) but left the rest the same and made it 3 1/2 hrs prime in-place ftt before it failed. Going to ramp up the voltage again and give it another go.
 
Ok, I've updated my screenshot (same URL as before) to show as much detail as possible. Pushed to 2-2-2-8 240mhz without a hitch on the ram, still 294x9 on the cpu but with some voltage fine-tuning.

The ram's a little more volted than I care for, but for the time being it's fine (assuming the 0.02 is margin of error, that stuff still under warranty but it's a wee bit warmer than I had expected to the touch). If I can't get it to run at those timings at 3.2 or lower (after some burn-in and furher stability testing time) I'm going to have to look at multiple side-fans on my case. One for the video, one for additional for the cpu and/or one blowing directly and actively on the ram.

I've got a suspicion that the memory controller was simply not behaving well at the well rounded HTT numbers (280, 260, 275, 300, 305, 310, 315) and was directly proportionate to those well rounded numbers vs. the odd numbered ram dividers. I'm thinking I may be able to push this to an odd number above 300 - if my guess or feeling is right. Otherwise this isn't exactly too shabby. 60mhz shy of my goal.

It's been a long road (and a long thread) but I'm about done I think. Just fine tuning from here.

Next step is to re-arrange my office, pretty up my case (while adding side fans, I might actually add a bit of rice in the form of a window and a light... or at least a couple of fans and a light).

From there, who knows. Maybe water. Maybe phase. I've caught the bug though :)
 
hey lukano, can you post your 32M and 1M superPi results after you're done tweaking? i just want to see how mine compares to yours.
 
peterhoang2002 said:
hey lukano, can you post your 32M and 1M superPi results after you're done tweaking? i just want to see how mine compares to yours.

Sure, I'm out of town at the in-laws right now but I'd be happy to when I get back (2-3 days max, maybe tomorrow if I'm lucky).

What times should I be looking at for good, bad and ugly on both 1M and 32M?
 
Back
Top