Phenom 9500 w/ MSI K9A2 Platinum

[justapost]

Well well.. we're making good progress.

See the above 2.6G 20hr stable?
I run that setting daily... replicated it just now at 30C, 35C, 40C, 45C, 50C and 55C BIOS (70% load)/EVEREST -> results:

30C idle/load -> stable
35C idle/load -> stable
40C idle/load -> stable
45C idle -> shaky
45C load -> stable
50C idle > no boot!
50C load -> shaky
52C idle -> no boot!
52C load -> mostly errors
55C idle -> no boot!
55C load -> full of errors!

Gack I was certainly surprised.

Yuk, watch this

I assume the 70° limit is for core temp.
Running at stock voltages my system ran stable for an hour at ~80° cpu-temp reading. Everything (mosfet's, nb, mem) was nearly too hot to touch.
With lower temps I always had temp spikes ~+7-10°C. With temps above those spikes in general there where no such spikes anymore.
Temperatures only become an issue once vcore is increased.

[KTE]

Achim, those temps look OK to me.

Can you try to run stability testing at 70C for each core, +10C what you have there and see if it holds?
(core diode readings [DTS] are the individual ones that we're concerned with, socket diode reading is the first one labeled CPU - ignore this when you have the individual core temp readings from the DTS functioning)

And yes, EVEREST devs can't read separate Phenom core temps because they only know how to access one, that's the reason you have only one temp. feedback for all the cores synchronously.

Also, you're at much lower CPUV/NBV and speeds than me, you have better cooling and lower ambients and you have higher load temps... quite impossible. It means either mine are minimum Tcore+15=Treal or yours are at least >5C too high. Cores aren't individually calibrated at a separate high Tcase, it's fixed at 70C and monitored by a separate TCC diode and feedback to those which initiates core shutdown by pulling the Vdd pin supply.

Can you try booting with a fan on the heatsink keeping a temp probe on the IHS in between them, getting into the BIOS monitoring very quickly and then disconnecting the fan to see at what probe/BIOS temperatures it shuts down? Don't let it get past 85C BIOS...

[justapost]

Achim, those temps look OK to me.

Can you try to run stability testing at 70C for each core, +10C what you have there and see if it holds?
(core diode readings [DTS] are the individual ones that we're concerned with, socket diode reading is the first one labeled CPU - ignore this when you have the individual core temp readings from the DTS functioning)

And yes, EVEREST devs can't read separate Phenom core temps because they only know how to access one, that's the reason you have only one temp. feedback for all the cores synchronously.

Also, you're at much lower CPUV/NBV and speeds than me, you have better cooling and lower ambients and you have higher load temps... quite impossible. It means either mine are minimum Tcore+15=Treal or yours are at least >5C too high. Cores aren't individually calibrated at a separate high Tcase, it's fixed at 70C and monitored by a separate TCC diode and feedback to those which initiates core shutdown by pulling the Vdd pin supply.

Can you try booting with a fan on the heatsink keeping a temp probe on the IHS in between them, getting into the BIOS monitoring very quickly and then disconnecting the fan to see at what probe/BIOS temperatures it shuts down? Don't let it get past 85C BIOS...

Just got the new mobo, gimme a few hours to get familar with it.
I tested stability with a config whom was stable for at least 12 hours priming. I removed the extra 12" fan and ran the cpu fan at lowest possible speed, that's why my temps are so hot.
DMM with temp probe should arrive in a few days.

[KTE]

Oh, I see.

So you have the mighty Sapphire now?

[tictac]

tictac: what was that about left shift+tab F1 right shift+tab F2 or something you were saying... ?

KTE... press left shift+F2 with left alt+ F3 at DigiCell to unhidden setting

[jesteronyer6]

I took my CPU to 2.6 using KTe's settings except I ran the htt @ 2000 and my memory benchmarks have dropped and lantency is poor. This bench is running back at 2.5 which was before the overclock getting much lower latency any ideas. I ran stress testing for over an hpur at 2.6 and it was pretty stable then I shut it down as was just messing around and started running the benches and noticed the poor mem performance. I also got a BSOD during the testing so I backed it and even went to stock settings and got the same poor memory performance am i missing something? Is is possible the memory controller is damaged?

[KTE]

jesteronyer6: latency is totally dependent on your CPU speed, NB speed and RAM speed/timings/mode + TLB boost.

So depends what you chose. I don't even get those high latencies at 2.2G CPU, 1.8G HT/NB, 400MHz 5-5-5-15 RAM with AOD button=Green, they are too high. I get those only with the TLB fix enabled BIOS's.

Hence what is happening there, what are your total settings?
What latencies were you getting before?
Switch all off, close EVEREST and start AOD then click the caching button twice until it turns red. Then close it and run EVEREST again... what do you now get?

IMC doesn't get damaged if it's stable, temperatures good and voltages within a safe range. You are running 600-400MHz lower IMC than I do daily, that can't damage nothing.

tictac: it doesn't do anything.


I'm currently checking out the AutoXpress BIOS setting.. it has CPU Boost, RAM Boost, GPU Boost options underneath it. If you choose all 3 your system idles 32W AC above idling with them disabled (all extra's kept constant) yet the cores have 0% usage and TLB caching button is green and performance is also the green state... however the cores are idling 10-12C higher too. Max load wattage is also 5W AC higher.

[KTE]

Dave: I hope you weren't living in the four States beaten up badly tonight - wishing safety.

[jesteronyer6]

KTE

I am using the 1.13 beta bios I didnt think it had the TLB fix?

NB = 2000
CPU = 2500
Timings = 4-4-4-15 2T

[KTE]

Nope, it doesn't.
As I mentioned, click the AOD button to red and then re-run and let me know what you get.

[Hosser]

No PCIe choices in the BIOS's so far madfaze.

You can overclock PCI-e bus from AOD though?

[jesteronyer6]

Nope, it doesn't.
As I mentioned, click the AOD button to red and then re-run and let me know what you get.

I will do that when I get home tonight, something is really weird with my setup as teh cores show #f00001 or somehting like that instead of a temp. And now the memory has gone south since overclocking the CPU, I wonder if the BIOS is b orked maybe a reflash will fix it because I was getting high meme scores and good latencies right up untill I took it to 2.6 and primed for an hour with the higher nb voltages of 1.3.

[Manicdan]

am i the only one to see his name from D412K5t412 as DARKSTAR

[KTE]

You can overclock PCI-e bus from AOD though?

Yep. I've had max AOD MHz earlier in this thread.

I will do that when I get home tonight, something is really weird with my setup as teh cores show #f00001 or somehting like that instead of a temp.

Since when though? And in what app exactly? Have you tried cross checking multiple apps? Please explain fully with all details and testings any problem or I am clueless as to what you mean exactly.

And now the memory has gone south since overclocking the CPU, I wonder if the BIOS is b orked maybe a reflash will fix it because I was getting high meme scores and good latencies right up untill I took it to 2.6 and primed for an hour with the higher nb voltages of 1.3.

As I asked earlier, why makes you say your mem is degraded? What scores are you comparing those to? You need to troubleshoot in a fixed clear way or it's just going to confuse you. Mem has nothing to do with the multiplier overclocking you did as you didn't touch it. The only other problem that can happen is if you degrade your IMC running high volts, not 1.3VID though, which is around 0.02V higher than stock. The only MEM degradation I've heard is when you run higher than 2.2VDIMM on RAM with low VCore and even that was only found by a select number of MB MFG employers, no one else to date. But in the end, this is oc, not warrantied and anything can happen as you are always warned - only you are liable yourself for what happens.

am i the only one to see his name from D412K5t412 as DARKSTAR

Are you deciphering it yourself or actually seeing that?

I see it as it's written: d412k5t412

[jesteronyer6]

That is what I mean by the temps missing.

As far as performance I hit the red button and all the performance is back which brings the question of why is TLB fix enabled when I am using 1.13 Bios is it possible that SP1 RTM brings an automatic enabled TLB fix?
http://www.xtremesystems.org/forums/...d.php?t=176165

[KTE]

I think that's temporary; the core feedback looks slightly offbase. Check using EVEREST, AOD and Coretemps to see what the temps they show. TMPIN0 is the core temp too for your reference.

The performance of that button has been shown in the last pages jesteronyer6. Without it performance is low on any board even without the TLB fix and even worse in Vista. With the TLB fix you would be getting far higher latencies when even using the red button in XP. Typically green is low and yellow/red are both high performance. Keeping it yellow for general running is best IMO and that's what I do.

You can check your MEM perf. by just running WinRAR multi-threaded benchmark. If it's plus 1300KB/s there is no TLB fix and if it's sub 800KB/s then their is 'the TLB fix'.

I've been over the SP1 changelog before and there is no TLB fix mentioned nor any specific update only for Phenom processors. The biggest chnages are all imporvement of performance but some will be implemented in the next few updates, such as:

Improves performance over Windows Vista’s current performance across the following scenarios:
25% faster when copying files locally on the same disk on the same machine
45% faster when copying files from a remote non-Windows Vista system to a SP1 system
50% faster when copying files from a remote SP1 system to a local SP1 system
Improves responsiveness when doing many kinds of file or media manipulations. For example, with Windows Vista today, copying files after deleting a different set of files can make the copy operation take longer than needed. In SP1, the file copy time is the same as if no files were initially deleted.
Improves the copy progress estimation when copying files within Windows Explorer to about two seconds.
Improves the time to read large images by approximately 50%.

And so on.

[jesteronyer6]

I am trying V.POF Bios and it has a disable TLB option but it doesnt seem to be working, the good is that my DDR 800 is running at 1066 with a lil voltage bumps abd relaxed timings.

[Mathos]

This is what your WinRAR bench should look like with no TLB fix.



I know my temps may seem low, but I'm running 2x 120mm case fans, the Zerotherm nirvana blowing through towards the rear exaust fan, and 3 80mm case fans, 2 of those are assisting with gpu cooling on the Smilodon case. Not to mention it may be around 65f at the most right now in my apartment, it's cold down here right now in east texas.

Notice the crappy 12v rail output from the Raidmax PSU I have biting me in the rear. Has been the cause of some instability since installing the 3870.

And your memory should be similar to mine or faster depending on timings without the tlb patch at 2.6/2.4.

[madfaze]

when is the time you have to adjust the NB and HT voltages?what are safe number to adjust it?

[Mathos]

when is the time you have to adjust the NB and HT voltages?what are safe number to adjust it?

nb has about the same limit on voltage as the cores as far as I've read. You can adjust the NB with the k9a2 plat bios through the custom p-states menu. I believe for 2.4ghz I've got mine set to nb fid=8, vid=24 (1.25v vid), DID=By 1.

I haven't been able to adjust the actual HT multiplier in bios yet, the setting that is supposed to for what ever reason changes the soft nb multi instead causing the ht ref to under clock.. Though technically any system based of HT3.0 should be able to safely set ht speed to 2.6ghz(ddr5.2ghz) without issue. Only thing limiting that being the nb speed, since htt can't be any higher than the nb/imc. Raising that up would probably give the cpu some of the extra bandwidth that it's starving for.

[KTE]

I checked out the AutoXpress BIOS options:

AutoXpress Boost Off - AutoXpress Boost On

-

-

Improvement with AutoXpress On:

2k1 = +2%
CB10 = -0.2%
WPrime32M = 0%
Pifast = -8.2%

I've no idea what they're doing deep under but these settings are giving better GPU perf. but lower memory perf. overall and no driver/hardware settings we can see are being changed.

Also, funnily I checked Hexus Pifast. Oc'd my GPU using AtiTool, yes GPU, and leaving everything else constant. I tried it 3 times for each run. Once the GPU was oc'd from 800/700 to 843/963 the time was consistently going from 48.7s to 46.5s.

Below score at same settings was with just GPU oc'd:



Now with no GPU oc a few minutes after:




Heck this is fully consistent too!

BTW, a few personal updates.

My Phenom is now 1.24V 2.6G stable due to lower temps.
My RAM is running 1.98V real 5-5-5-15 tRC 11 tRFC 75ns 1066 Memtest 8hrs stable without extra cooling.


Mathos: Your PSU is poor but those volt readings are wrong anyway. It's the Fintek chip, it is always reading around 8V on the +12V rail, even when they're actually at 12V. Mine are 11.98-12.00 consistent measured with a DMM and software still shows 8.4V. Actual +12V line sub 11.4V and you'd have dead/damaged components pretty fast.

What kind of RAM do you have that requires 2.3V at 800 5-5-5-15? That's very high for 800 CAS5, usually you only need 2.0V for that at most.

madfaze: are you asking with Phenom or with a K8?
For K8, no NB/HT volts needs to be touched even at +2.0GHz HT.
Neither for Phenom, maybe HT volts are required when it's past 2.4G, I've not needed it yet at up-to 2.596G.

[madfaze]

nb has about the same limit on voltage as the cores as far as I've read. You can adjust the NB with the k9a2 plat bios through the custom p-states menu. I believe for 2.4ghz I've got mine set to nb fid=8, vid=24 (1.25v vid), DID=By 1.

I haven't been able to adjust the actual HT multiplier in bios yet, the setting that is supposed to for what ever reason changes the soft nb multi instead causing the ht ref to under clock.. Though technically any system based of HT3.0 should be able to safely set ht speed to 2.6ghz(ddr5.2ghz) without issue. Only thing limiting that being the nb speed, since htt can't be any higher than the nb/imc. Raising that up would probably give the cpu some of the extra bandwidth that it's starving for.

hmmm i got amd 5200+ x2's system as you can @ my sig...does custom p-state menu available on my system?cant see any @ bios...

[KTE]

madfaze: your IMC/NB is locked at CPU speed and you don't have P-State options to configure.

[madfaze]

madfaze: your IMC/NB is locked at CPU speed and you don't have P-State options to configure.

yup and who would have thought that this will be enable yet am just using x2's ...for now (still waiting for the b3's)

anyway regarding my inquiry KTE, when will be the time you have to adjust this NB and HT volt?...

[KTE]

anyway regarding my inquiry KTE, when will be the time you have to adjust this NB and HT volt?...

For Phenom?