Phenom 9500 w/ MSI K9A2 Platinum

[KTE]

Anyway if i enable CnQ the only thing that drop with phenom is the cpu multi and that has no measurable effect on poffer usage.
So i thought once it works as with k8 the cpu voltage will also drop.
I found it runs stable with 5x multi and CPUV 1,025V consuming 73,5W VAC idle.
Same system requires 91,5W VAC with 5x multi and CPU at 1,25V (min required to run stable at stock speeds).
So once CnQ also drops CPU voltage the power usage will drop ~18W.
NB runs at 1800 MHZ and also the memory runs at full speed in CnQ idle.
With an k8 cpu the memory speed drops and i assume also the ram voltages, i wonder if that will ever be the case on that board with a phenom cpu.

The RAM voltage on K8 stays the same, same thing as on Phenom. But the main thing which makes the power difference on Phenom as I formentioned is CPU/NB voltage rather than frequency. Frequency won't give you massive losses/gains in power consumption, but voltage will.

Here it got worse, after playing with BE-2400 for two days now the phenom requires 0,25-0,5V more for overclocking, trying to get it stable at 223 ht atm.
Saw you tried the 5000+ BE and must agree that those G2 cpu's are more fun to play with. Thought max ht ref for my board is ~290MHz, found that it can run up to ~340MHz with BE-2400. Run's stable with 10,5 CPU multi and 305MHz ht ref.

I had the same issues, honestly. Before trying X2 I was running stock volts at 2750MHz. Now I need +0.2V for stability and the max clock stable/instable has fallen down by 100MHz.

[justapost]

The RAM voltage on K8 stays the same, same thing as on Phenom. But the main thing which makes the power difference on Phenom as I formentioned is CPU/NB voltage rather than frequency. Frequency won't give you massive losses/gains in power consumption, but voltage will.

Aha ok, so on k8 systems the mem frequency was reduced to stabilize the imc at lower voltages.
I tried AOD from the gigabyte site, i can change the core multiplieres and ram settings all other tuning options do not work, so i can not modifiy p-states that way.
Did phenom reduce the nb-multiplier and the nb-speed in CnQ mode with any of the bioses u used?

I had the same issues, honestly. Before trying X2 I was running stock volts at 2750MHz. Now I need +0.2V for stability and the max clock stable/instable has fallen down by 100MHz.

Yeah, saw your post on that other thread after writingmy answer.
That is better that here, took me the whole day to get from 222 to 224, the day before i made my way from 200 up to 222 again. It runs prime95 for an hour without probs at 224. If i put 0,025V more on volts on the cpu with those settings, the system eighter freezes or powers off after ~5min prime95 (always a few seconds after the second test starts).

[KTE]

Aha ok, so on k8 systems the mem frequency was reduced to stabilize the imc at lower voltages.
I tried AOD from the gigabyte site, i can change the core multiplieres and ram settings all other tuning options do not work, so i can not modifiy p-states that way.

Achim use this AOD: http://www.techpowerup.com/downloads...rive_2.10.html

Go to this page under Preferences and untick the options you don't want to see. It will speed up AOD loading up time: http://img71.imageshack.us/img71/3712/pagelisths7.jpg

Here's what my system runs stock (look at VCore 1.184V):


Same for 2.42GHz, it's fully stable. If I OC using AOD the voltages automatically crank up or down depending on frequency.
This is something new and very odd. For instance, if I put 2750MHz then the voltages will automatically rise higher to 1.452V for VCore and 1.4V for HT.

Did phenom reduce the nb-multiplier and the nb-speed in CnQ mode with any of the bioses u used?

P-States is a BIOS option but still in the early stages. I set my HT/NB options fully in BIOS. When I test CnQ I leave NB at low frequency at 4x multi from BIOS, like this: http://img71.imageshack.us/img71/9030/91wnocnqzj8.jpg

Look at that screenshot carefully compared to others. Usually 1.2V is lowest you can set in AOD. But now AOD offers voltage all the way to 0.7V.

I've tried it but didn't measure the power consumption/temps at that voltage.

Yeah, saw your post on that other thread after writingmy answer.
That is better that here, took me the whole day to get from 222 to 224, the day before i made my way from 200 up to 222 again. It runs prime95 for an hour without probs at 224. If i put 0,025V more on volts on the cpu with those settings, the system eighter freezes or powers off after ~5min prime95 (always a few seconds after the second test starts).

It's your NB mainly. You'll need to get it below 1900MHz and then try this same procedure again. You should be able to get at least 2600MHz stable IMO with a low NB multi.

Which BIOS are you running?

[justapost]

Achim use this AOD: http://www.techpowerup.com/downloads...rive_2.10.html

Go to this page under Preferences and untick the options you don't want to see. It will speed up AOD loading up time: http://img71.imageshack.us/img71/3712/pagelisths7.jpg

Thank you for the tip, that version does not work with AM2 chipset, but M3A (770 chipset) should arrive this week finaly.

Here's what my system runs stock (look at VCore 1.184V):


Same for 2.42GHz, it's fully stable. If I OC using AOD the voltages automatically crank up or down depending on frequency.
This is something new and very odd. For instance, if I put 2750MHz then the voltages will automatically rise higher to 1.452V for VCore and 1.4V for HT.

Same behaviour here, after startup CPU-Z shows the voltage i configured in the bios (+-0,005V). Once i start prime95 it drops ~0,01V.

P-States is a BIOS option but still in the early stages. I set my HT/NB options fully in BIOS. When I test CnQ I leave NB at low frequency at 4x multi from BIOS, like this: http://img71.imageshack.us/img71/9030/91wnocnqzj8.jpg

Look at that screenshot carefully compared to others. Usually 1.2V is lowest you can set in AOD. But now AOD offers voltage all the way to 0.7V.

I've tried it but didn't measure the power consumption/temps at that voltage.

Looking at our results in voltage optimisations i think once CnQ works as it should the systems power usage will drop about ~10-15W.

It's your NB mainly. You'll need to get it below 1900MHz and then try this same procedure again. You should be able to get at least 2600MHz stable IMO with a low NB multi.

I do not have a M9A2, posted my M2A-VM results in conjunction with power consumtion comparison.
Tried to modify the NB multiplier via wpcredit but that does not work.

Which BIOS are you running?

Used 1501 on the first days updated to 1603 last weekend (only available from the Asus FTP site, allows to change the cpu-multi now but not for different cores). I'm not sure if my problems are related to the new bios, i think i did not run a stability test with the new bios and the phenom after upgrading the bios. The bad thin with the 1603 bios is, it has a downgrade protection included, so i can not go back to 1501 without an eeprom programmer.
OT: Guess it's time to get such a programmer together with an AM2 board supporting the linux bios, so i give that a try too.

[Blacklash]

My K9A2 CF arrived... dead right out of the box.

/sigh

[madfaze]

My K9A2 CF arrived... dead right out of the box.

/sigh

huh!? what happened dude?

[KTE]

Sorry to hear that Blacklash.

[KTE]

I wrote a full post addressing some Phenom directed FUD out there repeated every day lately as well as some updates and advice. Unfortunately a USB lead fell on the board and shorted it so it all switched off giving some sexy fireworks and I lost all the info.



It's more appropriate.

Quickly then, one word of advice:

If you oc Phenom (on MSI esp.), then I would advise you do this setup you see in this pic:



-Have things aligned similar to that. Then go up 1MHz HT ref. at a time. When you reach 5MHz above where you started, run P95 4-cores with error checking and keep an eye on your temps and which core screws up quick. There will be one weak one usually.
-If it gets past 5 minutes, stop it. Do the same oc again and again until you reach a point where it errors after the first 1 minute (not less).
-Then start giving it voltage and repeat the P95 test. Find what you can get plus 30 minutes stable/temps/volts. Play a game or bench it.
-If you've maxed out voltage and it is still erroring below 30 minutes while temps are well controlled, then you have to back down a few MHz and test again.
-Once you've found something plus 30 mins stable/good temps, stress that for a few hours (leave it overnight is best).
-When you get back and hopefully it's still running then pack all other tools away leaving AOD running, max volts and now try oc'ing it for max MHz/valids/highest benches.
-You should be able to reach far more this way than other methods, definitely on my board anyway.

This is why even when my system fails bootup at 2.42GHz 1.48V (like today) as shown here:



I can still pull 2.75GHz stable benchable at 1.39V out of it following the above method, like so (today):



And 2.85GHz short benched at 1.42V :

[informal]

Thanks for that post KTE.
What are your thoughts on this Phenom review by bit-tech?Something is VERY fishy with their game scores..According to their charts ALL Phenoms get some weird bottleneck when it comes to graphics related tasks(no matter the clocks).It looks like a board/HTT problem,or possibly PCI-E problem.Can you take a look and write your thoughts when you get time?

[linhvndiy]

Hi KTE: Thanks a lot for the post.
Please give more info:"run P95 4-cores with error checking and keep an eye on your temps and which core screws up quick. There will be one weak one usually". How to run P95, in Blend test or large FFTs?
And how the temp in well control? Under what temp?

[blargman]

So erm, ya
reading thru all this. I had a dumb question. I'm running 1066 @ 800mhz (geil 2x2gb). The OS here (vista64) is only report 3GB, appears to go down when I add my second 3870. :s

Anyone know whats up with that? I have no idea what voltage my ram is running at. I'm guessing 1.9. I tried raising it and it just made the system unbootable. Same went for trying to set the divider to allow for 1066 operation.

Is 1066 ram not supported by this board? does anyone know? And/or does anyone know why its not reporting the full 4gb in vista 64, or even in the DMI splash screen at boot. :\

I really think it's causing stability issues with my setup here
Phenom 9500
MSI K9A2 Platinum (bios rev 1.0)
2x2GB Geil PC2-8500
2x Sapphire 3870


I'm near sending this board and cpu back and just buying an intel chip and board.

[KTE]

Thanks for that post KTE.
What are your thoughts on this Phenom review by bit-tech?Something is VERY fishy with their game scores..According to their charts ALL Phenoms get some weird bottleneck when it comes to graphics related tasks(no matter the clocks).It looks like a board/HTT problem,or possibly PCI-E problem.Can you take a look and write your thoughts when you get time?

Can't really comment on games because I haven't tested them. I've not gamed properly since I started studies and gradually I've completely left it (only play Urban Terror & HL2:LC at times). I would've tested gaming, I still can with pretty much any card out now but I'm really not into it, it's a headache with too many sub-issues and variables that take up too much time. My monitor is also not big, there's 3 here and they're only 22.5", 19" and 16".

Some things I noted:

a) Their memory bandwidth is twice mine.

b) Their memory latency is 20ns slower than mine.

c) They have higher Core VID (higher amps fed), so naturally that will have higher power draw/temps than retail Phenom 9500 and 9600.

d) Their test system has 122W higher power draw idling than mine at stock.

e) They need to make the files they use for benchmarks standardized so everyone can re-run them and all the sites can test the same benchmark each time. Right now I cannot comment on those times because of so many variables involved.

f) I usually use SUPER for my media file conversions, every day. It runs faster on Phenom 2.4GHz than on E6750/E6850 at 2.8GHz as I've tested on the same day converting a 150MB 640:480 528kbps 25FPS Sorenson H263 encoded MP4 file with ACC VBR 1 channel 22k 128kbps audio into 320:240 384Kbps 15FPS H.264/AVC encoded MP4 file with ACC VBR 2 channel 48k 128kbps audio.

g) WinRAR looks a little low to me.

h) The NB clocks are too high for retail Phenom (messes up oc quick).

i) Their sample looks like an early batch. It has power problems and overclock problems worse than retail. My core gets less Amps fed to it than theirs and yet is still fully stable at 1.296V at 2.55GHz (as of todays renewed testing).

j) Their power consumption rise is simply because of high volts/amps to the core. That is quite obvious.

k) How did they overclock properly if they have the same issues as me of AOD random lockups? Most overclocking is best done within Windows. We're talking 500MHz higher possible this way.

l) Their Phenom core is multiplier unlocked but far worse than any retail yet we've seen. It can only go to 205MHz HT ref. Even the worst around can do 220MHz HT ref and average is 240MHz HT ref.

m) Their power consumption figures are strange for me. I'll have to checkup on those before I comment. Specifically, QX9650 being lower than C2D while its double the TDP and Phenom/Q6600 consumption.

n) Their gaming shows Phenom 2.2GHz is no better than 2.6GHz and all are far behind a Q6600, even more so than QX9650. If any of the multiple QX9650 people on this board could just test some apps/gaming on it then we can compare properly. I cannot find a spare Yorkfield to test, so I would have to buy one and since they are all non-existent yet, they don't count to me. I would consider myself the most stupid person if I bought one of them... not even when I'm drunk. It's like Gates buying Isle of Man just because he needs to build an acre large house on it. Waiting for feasible Yorkies like Q9550.

o) If they could've tried to overclock NB/HT/RAM/CPU separately and compare the results, we could see more clearly of any bottlenecks. I did some quick HT and PCIe testing and the results are quite surprising to me.

Hi KTE: Thanks a lot for the post.
Please give more info:"run P95 4-cores with error checking and keep an eye on your temps and which core screws up quick. There will be one weak one usually". How to run P95, in Blend test or large FFTs?

Small FFT for just cores while oc (large in the end for hours test). Make sure your RAM is stable though.

And how the temp in well control? Under what temp?

This is personal opinion really.
*For just benching, I tolerate upto 68C load.
*For 24/7 stability, I tolerate upto 55C load but try hard for sub 50C load.

I never usually run a CPU which tops 53C 8 hour P95 in-place large FFT load.


blargman: Can you give us a CPU-Z screenshot of the SPD tab?

Also update your BIOS to 112, 113, P0C, P0D or even the 1.1.

64-bit only sees 3GB? That is usually the case in 32-bit only since you have limited address space and it's used by hardware I/O mapping. Not sure what is going on here. Whhich slots have you installed the RAM in?

You're all welcome.

[informal]

Thanks for your reply KTE .Pretty much in-line with my own thoughts about this "test" by bit-tech.

Did your board recover from that weird no-OC syndrome?You said renewed testing @2.55Ghz so i recon it's back in OC business .

[linhvndiy]

Thank a lot KTE.
But even i can P95 in one hour, but suddenly systems hang without log. I dont know why. I dont thing becos of cpu or mem. Hic hic.

[KTE]

informal: Nope not yet. But I might have DDR-1066 working now, working on it.

linhvndiy: Then stress test it at stock and see if it does the same

[Blacklash]

[H]ardOcp did benches of Phenom a while back that includes games-

http://www.hardocp.com/article.html?...50aHVzaWFzdA==

[KTE]

That paints a much more different picture.

It speaks in gaming:
Phenom clock for clock = Q6600 G0.
But a little better power consumption.

[mongoled]

So erm, ya
reading thru all this. I had a dumb question. I'm running 1066 @ 800mhz (geil 2x2gb). The OS here (vista64) is only report 3GB, appears to go down when I add my second 3870. :s

Anyone know whats up with that?

Dont know if this will be of help but I think i read somewhere that there is a quirk with the memory allocation when a second graphics card is added. Are both the cards 512mb? Whats the amount of memory Vista is showing in mb? Is it 3072mb? I will try to see if i can drag some information up

I have no idea what voltage my ram is running at. I'm guessing 1.9. I tried raising it and it just made the system unbootable. Same went for trying to set the divider to allow for 1066 operation.

Is 1066 ram not supported by this board? does anyone know? And/or does anyone know why its not reporting the full 4gb in vista 64, or even in the DMI splash screen at boot. :\

Try running the board with 1 stick to see if the ram will runn at its rated speed. Then check some of the stuff KTE has wrote abt the board he has, very good info there

[beerking]

msi+phenom9600


http://valid.x86-secret.com/show_oc.php?id=281206

[KTE]

msi+phenom9600

http://www.vmodtech.com/files/3_________998.jpg

http://valid.x86-secret.com/show_oc.php?id=281206

Nice.

Details?
Also can you run 1.2.66 RAM divider from BIOS?

[Blacklash]

msi+phenom9600


http://valid.x86-secret.com/show_oc.php?id=281206

Now this is what we want to see.

[KTE]

I did some quick PCIe test on MSI Platinum using AquaMark3.

Phenom 9500 @ 2.2GHz, 1.8GHz NB/HT
DDR2-800 4-4-4-4-18 1T
Sapphire HD 2600 XT
XP Pro SP2 32-bit

Stock CPU/RAM + GPU @ 800/700 + PCIe 100MHz:



Stock CPU/RAM + GPU @ 843/981 + PCIe 100MHz:



Stock CPU/RAM + GPU @ 843/981 + PCIe 116MHz:



Very funny, as CPU score is different each time but overall 300 points were added from 100MHz PCIe to 116MHz PCIe.

[blargman]

blargman: Can you give us a CPU-Z screenshot of the SPD tab?

Also update your BIOS to 112, 113, P0C, P0D or even the 1.1.

64-bit only sees 3GB? That is usually the case in 32-bit only since you have limited address space and it's used by hardware I/O mapping. Not sure what is going on here. Whhich slots have you installed the RAM in?

You're all welcome.

Updated to 113. Windows sees 3GB, CPUZ just detect each slot as 2gb but I dunno if that really means anything.

Still 800mhz and single channel though even after the bios update changing it to 1066 makes the machine unbootable.

I installed the dimms in slots 3&4 ( or 2 and 3 whichever you wanna call em).

I'll throw up some SPD shots tomorrow, I changed the memory to 1066 and my machine won't boot anymore and I was too lazy last night to reset the CMOS again! so my machine is off at the moment

[KTE]

Updated to 113. Windows sees 3GB, CPUZ just detect each slot as 2gb but I dunno if that really means anything.

Still 800mhz and single channel though even after the bios update changing it to 1066 makes the machine unbootable.

I installed the dimms in slots 3&4 ( or 2 and 3 whichever you wanna call em).

I'll throw up some SPD shots tomorrow, I changed the memory to 1066 and my machine won't boot anymore and I was too lazy last night to reset the CMOS again! so my machine is off at the moment

Yep, same case here with 1066.

Put your RAM sinto slot 1&2 (side by side). What do you now get?

Run Unganged mode and it should be dual channel.

[blargman]

Yep, same case here with 1066.

Put your RAM sinto slot 1&2 (side by side). What do you now get?

Run Unganged mode and it should be dual channel.

I'll give it a go when I get home this morning. I think I tried 1&2 before, no difference really. When I run 1 3870's it detects erm 3278 I think? , then 3072 or so is when I run two 3870s. :s I really hope this isn't affecting my 3d scores. I have yet to run 3dmark yet with my 2 3870s, I think I may do that this morning too.

Do you know if I should have this Memory hole thing enabled? it is by default. I've considered modding the Timings manually, but I have no idea where to begin. The ram I've got is 5-5-5-15 pc2 8500 I believe.

GX24GB8500ESPDC

*edit* umm are the PSU readouts at all reliable with this board? My 12 is around 12.5V which is kind of concerning. I think the 5v was a bit off as well. I've got a Corsair 1000W psu.