Phenom 9500 w/ MSI K9A2 Platinum

[batmang]

So, is AOD the method used for OC'ing or are you guys using the BIOS strictly?

[KTE]

I use BIOS strictly. Although I do use and tinker with AOD (it's needed to test a few things)
You can use BIOS P0H if you want to try AOD better because it works much better with it although with that BIOS you have hardly any BIOS oc options.

BTW, my system has totally gone naft. Won't boot at any MHz without BSOD. It happened straight after the system crashed on opening AOD. It's not a hardware problem though (ran LiveCD/UBCD4Win etc), the crash ruined my OS and my profiles so all I get is a BSOD each time even in Safe Mode because the files on the HDD have become corrupted and many, lost.

[Oldguy932]

Man that really sucks KTE. Last time I corrupted windows I was able to reinstall and I didn't format so I could get to my stuff.

Update on Vapo: 250x11.5 made it 2 hours on small fft's hitting a max of -9C cpu and -35C evaporator head.The cpu temp isn't altered, speedfan is reading directly from the bios. I'm gonna try a few other things now.

[tictac]

nice.. its quite hard to keep higher htt than 266mhz stable on phenom. would be handy if you have unlocked cpu multiplier on that.

[Oldguy932]

My BE can use volts above 1.325 at least. Anything more than that on this phenom causes it to not boot. I think I just got it stable on 255htt though, gonna make it run a few benches first though. If it even runs 270, if it even boots, then there will be an improvement over running it on air. I just hope the B3's are a bit more tolerant than these B2's I'm testing with.

[KTE]

Any 8hr Prime95 runs captured yet with temps?

How much vCPU NB are you feeding it plus 245HT? What multi is your NB at?

Your 1.325V issue may very well be your BIOS. I've not heard or seen that happen with CPUs yet, mine have all booted high volts on air and subzero, just they make Phenom unstable IME.

[Oldguy932]

Its not bios, the BE will take it. That voltage I put to 1.3v and then leave it alone, moving it does me no good. 260htt just will not stabilize no matter what I do. I just dropped it back to 255 and I'm gonna prime it for a while. I might just put the BE back in if I have to deal with this much crap from the htt.

Specs to try and get 8 hours stable

HTT=255
CPU=11.5x
NB=9x
HT-LINK=1.8ghz

vCore=1.275 set, 1.26 actual
vCore-NB=1.3 set
vNB=1.3 set

I'll take another look at the clamshell later, I'm busy for the next 6 hours or so.

[justapost]

@oldguy932: Do you get that 1.3V stable wall chilled or with "normal" aircooling?
Also what does cpu-z report for the voltage in idle and under load?

[Oldguy932]

Whatever is set in bios goes down one setting under load. 252 wasn't stable, gonna drop the multi and see how everything else scales right now.

[justapost]

Whatever is set in bios goes down one setting under load.

On the 690G chipset there was a different behaviour, that drop increased and also the difference between what i set in the bios and what cpuz reported. High drops where a sign for too high vcore there.
I have a 790fx mobo since a few days and get an behaviour like yours.

252 wasn't stable, gonna drop the multi and see how everything else scales right now.

On the M3A ref HT's above 260 where posiible with 533 mode for the memory. Modest watercooling only helped in power consumption but i could not gain higher frequencies.

[Oldguy932]

I do have memory set to 533. Right now cpu multi is at 8x, htt at 260, and cinebench 10 seems to be running pretty good right now. Still doesn't explain the instability at high clocks though. I might just be going for the superfast nb and have to live with a slightly lower cpu speed than I would like.

[jonspd]

I'm thinking more along the lines that my 9600 was just a really good pick. Only thing it won't do is handle voltage over 1.325 to make it more stable once I got to much past 3ghz, but 3ghz is game stable for me at 1.3v.

Edit: Yeah, the snow was a bit on the unexpected side for me.

I want to see some prime 95 12-24 hour test that was run at that setting...

[Oldguy932]

You'll have to wait till its back on air, the vapo is not making these chips do any better whatsoever yet. I have the BE in now, not sure what its gonna do. At least mine isn't coldbugged like the one in the other thread.

Edit: Here is my only plus so far from using the vapo, 12x nb on the BE is actually stable but is not on air. Its the only one so far.

[jonspd]

What do you consider stable if your not running prime95 for at least 12 hours?

[Oldguy932]

I played Command and Conquer 3 on it for about 3 hours, no hitches whatsoever.

Oh and I kinda started hating my BE again, I have a feeling I should probably sell the sucker cause I hate it so much. Wouldn't even start windows on 1.3v at 2.8, which worked perfectly on air. I have a feeling that the bios is reading at least 10C above what my temps are, evap at -39C and cpu at -7C...that's way off...When I first got it, I ran my E6600 on it. I could run that at 3.9, 1.55v or so, and load would hit in the negative teens because I turned the fans down to 25% cause I wanted it to be quiet. At 100% I think it was -25C or -22C, I don't remember cause the mobo died and I didn't ever get a new one. Hooray for my cheap 120mm case fan that I mounted over the chipset, no fan on it got it up to 40C, and now its at 35C so I'm happy about that. Oh and current priming settings are...

HTT=250
CPU=11.5x
NB=9x
HT-LINK=1.8ghz

vCore=1.2625, 1.25 actual
vCore-NB=1.3
vNB=1.3

And make that 34C on the chipset, its very slowly cooling down. If this gets through the night, I will take a screenshot in the morning and show you guys.

Edit: Going for 1:45 now, right about where I was earlier before I started tweaking again. Turning off the heat pads keeps the evap head at a much more constant temp, otherwise there's a slight but constant rise over time for priming.

[tictac]

ok leave it at 1.45v . let it cold burn-in for the rest of night. time for sleep.

[KTE]

Yeah, try some >8hr P95 smallFFT stability testing. We'd finally like to see some screenshots now of what you can get 24/7 stable on phase.

Or if nothing on phase then air.

BTW do you have any pics of your phase setup with Phenom? Especially the mounts and head as that's what I'm interested in.

I have a feeling that the bios is reading at least 10C above what my temps are, evap at -39C and cpu at -7C...that's way off...

Evap is always lower than IHS and IHS is always lower than core temps. Original core temps should be about 15-20C higher than IHS. Maybe you have a contact problem because in my testing -15C IHS was 5C BIOS, which is good.

And for everyone try this software 144M test to see if anything is initially stable at any MHz. It works to catch errors excellent on Intel CPUs but seems to have problems with AMD CPUs. Please monitor your temps, volts and power draw: http://www.superprime.org/

Someone post a screenshot back of 144M completed with Phenom because I'd like to see if it runs on AMD CPUs. It's a cracking piece of software if it does run, better than Linpack 32b IMO.

My Phenom isn't booting yet, destroyed HDD has thousands of errors now. Using Q6600.

[Oldguy932]

Man that just really, really sucks KTE. And that's the last time I turn the monitor off during prime, something happened between 3AM and when I actually woke up this morning and I couldn't see what because it wouldn't go back to windows when I turned it on. Just a random thought, but would AS5 do anything to give me lower temps. I was just thinking of maybe using it but I think I read something about it not really being worth it at such cold temps.

Edit: Either the "bad" core on this cpu is different than almost every other phenom, or core 1 just doesn't like cold. I hate windows update, just randomly decided to start installing updates and now its going to take forever to reboot.

[tictac]

afaik.. as5 only good for air cooling. dielectric grease would be better on sub-zero. do you really need it to be orthos stable? benchstable would be better for hwbot rank attempt..

[KTE]

Don't use AS5 for cold, use AS Céramique.

Can't still boot mine.

[Oldguy932]

The only records I don't have on hwbot for the phenoms are cpu-z max overclock, the super-pi runs on the 9600, and pcmark2k5 on the 9500. All the others are golds, although I'm waiting for someone to beat the pifast on the 9500 cause that was a crap run compared to what I can get it to do now, I can get it into the 39 sec range.

So the best paste/grease would be the the Arctic Ceramique? You think I would even notice a difference from the stuff that comes with the vapo?

Oh and 11.5x251 is currently 2 hours stable, load on speedfan says -5C.

[KTE]

Stability depends on the need.

If someone is chasing records, then they don;t need or wnat stability and to be frank, that doesn't matter to me. Anyone with some H20, Phase, LN2 or DICE and a cherry chip would get them in a hunch, since Phenom competition is low. This is air inside a case mostly. Most after a title can and have got 3-3.1GHz out of their chips but to get better scores they need to work on efficiency issues. I know Sami has higher MHz and scores than what we've seen publically already.

In this thread, we're only really interested in what you can get Phenom to run every day stable and MSI RD790, since making suicide bench runs isn't what I'm looking for with a 100£ chip. Anyone chasing MHz/benches will just go Penryn now. HWBot isn't my aim nor concern when buying a Phenom and won't be for the majority of Phenom buyers. Only reason I have a few low MHz runs on there recently is for a HWBot mod for Phenom comparison tests since it's data set was really low for analysis, i.e. the efficiency at those MHz, and I saw people posting very low efficiency results requiring high MHz for low scores. They could all do much better yet.

Give me a 2.9GHz benchable chip and you'll see what's possible with Phenom.

[Oldguy932]

The main reason your phenom does better at lower clocks is because you keep your memory that high. My memory either isn't stable at that speed, or the memory controler just can't handle it at the speeds I set it for. I compete on hwbot cause I find it fun to compare how well you can clock a chip, I tend to stay away from the graphics card benchmarking now because all the benchmarks are heavily cpu dependent. Phenoms are very good at doing multicard though, I did see that before I put the x1650xt's in my bros computer.

I hate waiting for prime to finish... 3 hours down...

[Manicdan]

KTE: i did the superprime on my x2 5000be @ 2.8ghz and got 89ms for the 144M test then did it again and got 24ms, and does 23-25ms for any of the tests i do

[KTE]

The main reason your phenom does better at lower clocks is because you keep your memory that high. My memory either isn't stable at that speed, or the memory controler just can't handle it at the speeds I set it for.

My few runs are at quite low RAM clocks, some at 820 4-4-4-4 which is very low. What RAM speeds can you bench 1M tops?
High RAM clocks Phenom can't handle in my testing, it gives them too high latencies. Good RAM clocks are like 1260 4-4-4-4 10-10 PL4, MaxAsyncLatency 42, that indeed would get high scores and perf.
You don't need high RAM clocks, you only need 820-980 4-4-4-4 or 740-800 3-3-3-3/3-4-4-3. You need a high NB clock, nothing else, as long as your chip can do high CPU. But bear in mind; actual high IMC clocks means lower/loser RAM clocks at the same volts. It's like Intel system MCH strap performance.

So say at the 1:2 divider, you need something like 218x14 (3052), 872 4-4-4-4 218x12 (2616) on NB and every single one of the Phenom results I've seen to date will be bettered without doubt and some by quite a big margin. Only wPrime 1.55 may not because wPrime runs at least 1s faster on Vista than on XP, unless the person changes to Vista as guys chasing ranks do. I know 30h6 and some others are playing and learning with a Phenom build to bench, so you might very well see the high MHz+efficient results soon.

1M at those settings will be sub 24s quite easily if the person knows what they're doing with Pi (i.e. can get sub 14s with a C2D/C2Q at 3600MHz on DDR2). A Phenom 3600MHz can get 20s 1M and even break into the 19s barrier.

Just normal knowledge of how Phenom works needed.

I think you missed something I asked earlier Oldguy932

BTW do you have any pics of your phase setup with Phenom? Especially the mounts and head as that's what I'm interested in.

KTE: i did the superprime on my x2 5000be @ 2.8ghz and got 89ms for the 144M test then did it again and got 24ms, and does 23-25ms for any of the tests i do

Yeah it means it's not working.

Thanks, but do you have any screenshots when it finishes (press show and take ss). I'll let the developer know. If it runs properly, you can expect some grand surpirses.