Some tips on taming the FX60 under cold

[mike]

I have played around with a FX60 in the last week trying to figure those buggers out and have the following findings that should hopefully help some FX60 owners - these findings were mainly under subzero cooling and on an Expert board and UTT based memory.

Here in very brief some observations:

- you will not be able to get the same clocks on UTT/BH5 type memory as you are able to get on a single core AMD

While on an FX57 some of my ram did ~270mhz on the FX60 it did more like 260mhz. Don't try to get your ram to clock the same as it was on your single core - or you'll run into a wall.

- You need to run might softer Data and Drive strengths - especially for Phase Change.

With UTT/BH I found 5/2 - 5/3 to be the best combo for me. it is mandantory to play alot with both these settings in order to fully unleash you FX60 under cold.


- Most FX60 will find a Vcore barrier - in my case I got the best results @ ~ 1.67V. While I have run FX60s as far as 1.75V - lower vcore appeared to be better. VCORE played a mayor role for me - having too much VCore killed many clocks

- You can run FX60 much colder with a higher multi and lower FSB. Slightly below subzero - the B2 controler seems to crap out @ ~ 258HTT, -20C @ ~ 230HTT and -35C @ ~ @218HTT.
Mayor has already posted about the updividers and I can only recommend using them. Higher multis will give you 216/233/250 while lower multis sometimes only give you 233/250 iirc




I think most B2 FX60 should easily be able to handle almost 3.5GHZ. I was able to test three 60s and was able to push them all to that speed.

I am still have not managed to have the time to properly bench the FX60 - but if you are able to control your temps - 3.6ghz+ benches are very well possible. The key thing is to either lower your temps during load - and raise them again after you bench is finished - or to have something put a load on your system immediately after the bench is finished. MPS Multi could help - but I haven't personally tried it yet - when I updated to the MPS muli driver from ACPI - my XP install on one drive wouldn't boot anymore - and I haven't bothered to try to get it to work yet ( I assume I must install MPS from the XP setup...)

My fastest 3Dmark05 CPU speed was 3.63GHZ - unfortunately I was running my 7800GTX 256MBs on high quality - otherwise I would have been easily faster than my previous personal best on 256MBs SLI cards and easily broken 18K

As soon as I remember some more things to add I will post them.




Some fun OCs:

While I was able to push my FX60 past 3.9GHZ with a single stick of memory - I only managed to get 3.868 GHZ Validator file saved:

http://valid.x86-secret.com/show_oc.php?id=75052




Dual SP2004 screeny @ 3.71 - I actually let it run for another min and raised the clocks to 3.74 - which it lasted for about 20 secs - but crashed and froze my system before I could get a screenshot this was with 2X512MB ram with an updivider for the ram ( believe 216 or 233)

[Major]

Good work Mike

[formyfaith]

Those are very nice tips for FX-60 owners.

I really hate the cold bug with FX-60...

Had a chance to test 3 of them also, and none of them liked the cold.

Gotta wait forever for a non-cold buged stepping. lol.

[mike]

Good work Mike

Thanks.

At first I was dissapointed when I saw you post about the updividers -hoping that I too had something that would potentially give me an edge over others for some benches - but decided this is better shared. Especially with my second DFI Expert mysteriously disintegrating ( PCB underneath a PCI slot is peeling and traces are hanging out)

Here's the 256MB sli bench @ 3.62GHZ
http://service.futuremark.com/compare?3dm05=1840258

What confuses me still is that I can dual prime @ 3.7 but not not 3Dmarks @ 3.7GHZ which actually tasks the core even less. Must be mostly be related to memory I assume.

Those are very nice tips for FX-60 owners.

I really hate the cold bug with FX-60...

Had a chance to test 3 of them also, and none of them liked the cold.

Gotta wait forever for a non-cold buged stepping. lol.

We all hate the coldbug - but I have to say these B2 FX60 are great cpus nevertheless. 3.15 - 3.2 on good air is better than most FX57 were able to do. Just testy under cold :-)

Non coldbugged FX60 - every benchers dream!

[Gautam]

I've heard that loosening up your Trc and Trfc help a bit in getting your memory running quicker in the cold. It might be worth a shot as it is pretty strange that you're priming nearly 100MHz above where you 3DMark...should normally be the other way round, lol.

Once you change an installation with an ACPI HAL to an MPS HAL, there's no telling what will happen. Usually it does fail to boot. :-/ Except personally I'd avoid the MPS Multiproc since it hurts performance. More of a last resort IMHO.

Keep on clocking bro, those are some insane speeds you got on your hands. It looks like with sufficient tweaking, the cold bug can be fought off pretty well.

[mike]

I've heard that loosening up your Trc and Trfc help a bit in getting your memory running quicker in the cold. It might be worth a shot as it is pretty strange that you're priming nearly 100MHz above where you 3DMark...should normally be the other way round, lol.

Once you change an installation with an ACPI HAL to an MPS HAL, there's no telling what will happen. Usually it does fail to boot. :-/ Except personally I'd avoid the MPS Multiproc since it hurts performance. More of a last resort IMHO.

Keep on clocking bro, those are some insane speeds you got on your hands. It looks like with sufficient tweaking, the cold bug can be fought off pretty well.

I spent quiet some time loosening and tightening memory - and honestly hadn't found abything that really helped me greatly. It truly is a mystery how I can prime higher than 3Dmark05 - it has to be in the memory - but even when I run super loose I still bork in the same spot in the 3Dmark05. I am pretty sure that I would be able to run PCMarks closer to 3.7GHZ - but 3Dmarks

My top clocks in PCMark04 were 3.6GHZ a while back - and that was on a small single stage phase change - and under same conditions I was only able to 3Dmark @ 3.53 .

Won't be spending much more time with the FX60 unfortunately - will have to move to the FS section. I will try to give it one more run for a short 512MB SLI bench session nad then call it a day.

Memorywise I actually tightened the crap out of my memory - since it isn't possible to run high clocks on it anyways at least try to be as tight as possible - I went running 2.2.2.3.12 and the ultra tight everywhere else.

MPS update - thanks - I assumed that this was going to happen - I need to get some slipstreamed XPs going for myself - at the rate I am dorking OSs...lol

[stone_cold_Jimi]

I've heard that loosening up your Trc and Trfc help a bit in getting your memory running quicker in the cold. It might be worth a shot as it is pretty strange that you're priming nearly 100MHz above where you 3DMark...should normally be the other way round, lol.

Disclaimer: my chip is a FX-60 wannabe, a runt of the litter (CCBWE 0546UPMW). Yet it shares many of the characteristics of its big brother; I recognise most of what has been said here in my chip.

The point about trc and trfc is valid. My CPU is running 2700 @ 1.49v, mem at 270 3-4-4-8 @ 2.59v, CPU idles at -33C lowest, load runs as high as -15C. I can not get good stability with a low trc; 7 was too low, auto sets it to 15 which is fine. 3DMark of any flavour will not complete with trfc at 3684, but it's fine with 3120. With trfc at 3120, it completes SPi32M and primes for an hour; with 3684, either of these will crash at some random stage.

So I think memory does come into it and, certainly, the "wrong" secondary memory settings will invoke coldbug behaviour at higher temperatures. Trfc at 3120 and a load of secondary settings on auto lets me boot as high as 285 HTT @ -27C. There's some mileage to be had in getting this right.

[mike]

Disclaimer: my chip is a FX-60 wannabe, a runt of the litter (CCBWE 0546UPMW). Yet it shares many of the characteristics of its big brother; I recognise most of what has been said here in my chip.

The point about trc and trfc is valid. My CPU is running 2700 @ 1.49v, mem at 270 3-4-4-8 @ 2.59v, CPU idles at -33C lowest, load runs as high as -15C. I can not get good stability with a low trc; 7 was too low, auto sets it to 15 which is fine. 3DMark of any flavour will not complete with trfc 3684, but it's fine with 3120. With trfc at 3120, it completes SPi32M and primes for an hour; with 3684, either of these will crash at some random stage.

So I think memory does come into it and, certainly, the "wrong" secondary memory settings will invoke coldbug behaviour at higher temperatures. Trfc at 3120 and a load of secondary settings on auto lets me boot as high as 285 HTT @ -27C. There's some mileage to be had in getting this right.

Very interesting input - well I will give it another go in a couple of days after both you and Gautam stressed this point - I mostly ran 3684 and 3120 as tref - but will look into the TRFC and TRC some deeper. Sad thing is that I have been mostly working of my memory (brain..) and not kept screenies to be able to make a more profound analysis on mem timings.

I had found tref 3684 to be the most stable in the beginning - but after constantly tweaking it might have turned into a bad choice afterall. On the other hand this could be different with your memory controler and the B2 controler - but let's hope there is a perfect setting that cures the 3D problems.

EDIT:
TRC I actually had never played with always set it to 7 - but TRFC whether it was on 12, 14 or 16 hadn't made a real difference for 3DMark05 stability - which I was using as my benchmark to test the 60.

[dinos22]

I am still have not managed to have the time to properly bench the FX60 - but if you are able to control your temps - 3.6ghz+ benches are very well possible. The key thing is to either lower your temps during load - and raise them again after you bench is finished - or to have something put a load on your system immediately after the bench is finished.

in terms of this comment i have a possible solution

why don't you visit this page: http://i4memory.com/showthread.php?t=1289

the F@H client is a low priprity task hence it uses left over CPU power......so during benching this proggie will stay at 0% CPU usage and jump to 100% or whatever percentage is left not doing anything hence keeping the CPU busy ............NOW another good thing about this proggie is if you are not 100% stable at benching settings you want to use you can set the program to use 50/60/70% or whatever %-age of CPU power rather than full 100% which means that it won't fully stress the CPU....you can fine tune it the way you like .......it can also be setup with multiple cores and auto startup....it's very flexible

another good thing is you'll be folding for a kick arse i4memory.com team and help save lives you like my sales pitch there

[CometMan]

there is something that i am not understanding please help out, i can use the fx60 whit low multi, high HTT but whit mem divider (233, 250)under cold is this correct?

[Vapor]

You want to keep HTT as low as possible when under the cold. The 233 and 250 up-dividers help you do this (esp since FX60s have freely adjustable multis)