Phenom 9500 w/ MSI K9A2 Platinum

[madfaze]

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

For Phenom?

both perhaps...

just wanna be safe and dont wanna fry or overvolt on somethin' inside the mobo...

[KTE]

both perhaps...

just wanna be safe and dont wanna fry or overvolt on somethin' inside the mobo...

NB/HT volts on K8 = none need to be adjusted unless going high on HT for added safety. I've done 2GHz HT with K8 on stock volts.

NB/HT/IMC volts on K10h;

NB volts don't need to be touched when oc'ing.
HT volts don't normally need to be touched till at least 2.5G or high HT ref. One bump up can usually settle 2.6G stable.
IMC volts need to be adjusted depending on the chips capability, requirement, speed and RAM speeds/timings. IME 1.25VID will be fine for <2.16G only, then a bump to 1.3VID will cruise to 2.4G fully stable. Past 2.45G starts demanding higher volts.

[madfaze]

NB/HT volts on K8 = none need to be adjusted unless going high on HT for added safety. I've done 2GHz HT with K8 on stock volts.

NB/HT/IMC volts on K10h;

NB volts don't need to be touched when oc'ing.
HT volts don't normally need to be touched till at least 2.5G or high HT ref. One bump up can usually settle 2.6G stable.
IMC volts need to be adjusted depending on the chips capability, requirement, speed and RAM speeds/timings. IME 1.25VID will be fine for <2.16G only, then a bump to 1.3VID will cruise to 2.4G fully stable. Past 2.45G starts demanding higher volts.

thanks KTE, so for now ill only adjust its vcore when OCing...

[justapost]

Oh, I see.

So you have the mighty Sapphire now?

Check this, smart guardian reports 15° higher temps than the core readings on the sapphire. Had it on ~84° core temp for around a minute an hour later the cpu died.

[LIKMARK]

Died for good?

[KTE]

You can very safely say those temps are reported wrongly Achim; the RD790 chip and motherboard temp sensors is reading minimum 20C too high if your ambient is 20-24C. Those core temps readings are actually at 53C minus 30C offset as shown in the CPU diode label taken from the socket.

So it's 53/4C EVEREST continuous load that made your CPU dead? Mine would not boot 52C EVEREST..

Which means, those 53C readings are most likely ~63-68C IMO. By my calcs, 22C ambient, it shows 20C per core, adding minimum 12C idle at 130W TDP and you would get 50C load readings when its at 60-62C actual (approx).

Although your core temps read close to actual, they didn't do so out of internal feedback accuracy but out of another error: high BIOS offsets for all system temperatures.

Woe, are you sure it's the CPU and not your board that's dead? Tried in another board or another CPU in this board yet (your BE)?

Leave it for a a day or two or I suggest put the chip under some cold if you can. See if it fires back up with nothing connected but one RAM stick.

[justapost]

You can very safely say those temps are reported wrongly Achim; the RD790 chip and motherboard temp sensors is reading minimum 20C too high if your ambient is 20-24C. Those core temps readings are actually at 53C minus 30C offset as shown in the CPU diode label taken from the socket.

So it's 53/4C EVEREST continuous load that made your CPU dead? Mine would not boot 52C EVEREST..

Which means, those 53C readings are most likely ~63-68C IMO. By my calcs, 22C ambient, it shows 20C per core, adding minimum 12C idle at 130W TDP and you would get 50C load readings when its at 60-62C actual (approx).

Although your core temps read close to actual, they didn't do so out of internal feedback accuracy but out of another error: high BIOS offsets for all system temperatures.

Woe, are you sure it's the CPU and not your board that's dead? Tried in another board or another CPU in this board yet (your BE)?

Leave it for a a day or two or I suggest put the chip under some cold if you can. See if it fires back up with nothing connected but one RAM stick.

Smart Guardian reports ~55° in idle with fan at ~2000RPM oced to 2,62GHz.
Assume it are ~30°-40° i might have ran at ~75°-85°.
I dont think that short heat test killed my cpu. I did a reboot and everything worked fine for an hour. I optimized memory timing short before and had it at ~500MHz 4-4-4-12-1T or so with 2,26V. Can be the High Voltage issue caused the cpu to die.
Mobo stops with C1 on the display, that should be a memory error. Tried different bios versions and different types of cmos clearing, as it's a know issue on the dfi mobo, but it did not fix the problem.
Tried the cpu in my other two mobos and it does not work.
Sapphire mobo works with X2 cpu. Had to use an X23800 first as the be-2400 hung at DMI pool stage.
Trying the cold thing now, but I guess i gotta rma that thing. Have no other phenom cpu but i assume the mobo has no problem.

[Oldguy932]

Well, I can tell you that the readings on the Asus M3A32-MVP Deluxe are quite a bit closer to actual than the msi board. 1.3v gets me 48C load at about 2.8ghz, and I know that if it hits 60C everything freezes no matter what, so I'd estimate 10C below actual. Temps were the same on both my 9500 and 9600 BE, haven't done prime at that frequency on my regular 9600. I tried that technique of priming for a long period of time and then moving up the frequency, but I hit a wall at 2.82ghz on the BE. 2.9 needed more than 1.4v so I just quit there.

Now I know you guys have been against bumping up the NB multi on the regular phenoms cause it gives wierd readings. Well bump it up to 10x and tell me how fast your L3 cache is compared to at 9x or 8x. On my board it makes the L3 Alot faster. 13x cuts it down from 48 to 34ns according to the latency batch program with cpu-z. The 11x multi gives me 6.9ns in everest and 40ns in cpu-z. The only problem is how the cpu speed gets calculated...

HTT, NB multi is NB, CPU multi is CPU

NB Speed=9*HTT as long as NB multi is 9x or above
CPU Speed=((9*HTT)/NB)*CPU

In short it means that your cpu is going slower, meaning you need less vcore. But at the same time, your NB is working much better than at stock. Here's a simple pic of the efficiency...



I kinda surprised myself with that run to be honest. Not quite comparable to a Core 2, but its on the road to being closer. 13x seemed to be comparable, but since the NB multi is higher than the CPU multi the cpu can barely overclock because it just keeps knocking the HTT limit down, from about 300 on this cpu.

I think that's good enough for a first post.

[KTE]

Smart Guardian reports ~55° in idle with fan at ~2000RPM oced to 2,62GHz.
Assume it are ~30°-40° i might have ran at ~75°-85°.

Yeah that's 35°C idling to 45°C idling maximum (-20 to -10). I would definitely think that's 35°C idling though based on what we've seen. So the load temps were 64°C-74°C there (-20 to -10). Again, knowing 60-65°C actual usually reboots/freezes Phenoms or doesn't POST, this would be inline even with my own findings (52C plus 10C offset) and those of Oldguy932 above.

I dont think that short heat test killed my cpu. I did a reboot and everything worked fine for an hour. I optimized memory timing short before and had it at ~500MHz 4-4-4-12-1T or so with 2,26V. Can be the High Voltage issue caused the cpu to die.

Have you ever ran above 2.2V (real) on the RAM before with this Phenom? Was your CPU voltage 1.4V?
I can run all volt ranges pretty casually.

Mobo stops with C1 on the display, that should be a memory error. Tried different bios versions and different types of cmos clearing, as it's a know issue on the dfi mobo, but it did not fix the problem.
Tried the cpu in my other two mobos and it does not work.
Sapphire mobo works with X2 cpu. Had to use an X23800 first as the be-2400 hung at DMI pool stage.
Trying the cold thing now, but I guess i gotta rma that thing. Have no other phenom cpu but i assume the mobo has no problem.

Hmm... so many days of priming and high VCore and you now have a dead CPU?

Well, I can tell you that the readings on the Asus M3A32-MVP Deluxe are quite a bit closer to actual than the msi board. 1.3v gets me 48C load at about 2.8ghz, and I know that if it hits 60C everything freezes no matter what, so I'd estimate 10C below actual. Temps were the same on both my 9500 and 9600 BE, haven't done prime at that frequency on my regular 9600. I tried that technique of priming for a long period of time and then moving up the frequency, but I hit a wall at 2.82ghz on the BE. 2.9 needed more than 1.4v so I just quit there.

Now I know you guys have been against bumping up the NB multi on the regular phenoms cause it gives wierd readings. Well bump it up to 10x and tell me how fast your L3 cache is compared to at 9x or 8x. On my board it makes the L3 Alot faster. 13x cuts it down from 48 to 34ns according to the latency batch program with cpu-z. The 11x multi gives me 6.9ns in everest and 40ns in cpu-z. The only problem is how the cpu speed gets calculated...

Welcome Oldguy932

We did many bench tests around 2 months ago, real applications rather than synthetic, a few of us with 9500/9600's and the truth was, that NB speed did not increase at all past 9x multi with them, CPU speed stayed at what it would be at 9x NB multi and so did HT speed. The readings were just being reported downlocked since you went past the setting NB can be placed at, hence why CPUZ won't validate those clocks and neither could you raise HT any higher than 9x.
HT speed can be moved up to the maximum NB speed generally, up until 2.6G.

The only software which showed any gain was synthetic which bases its values on the speeds it detects, which picked up the wrong high values (maybe 2 apps). Remember, I could hit 3.1GHz NB using those methods right from the 1st week and those we soon found out to be wrong (false - even Sami realized this).

The 9600 BE run NB multi/speed high, it has unlocked NB multi (locked downwards AFAIK). I've had upto 2.596GHz on the NB with it (actual). So yeah, on those it's possible, all values stay perfectly as you set them and it really does affect perf. very well.

Looking at the above screenshot, you only mention having a 9500 and a 9600BE, so if that's a 9600 BE, I can't reconcile what AOD shows and what CPUZ shows?

Why is AOD showing 2.5G when CPUZ shows 3G? Have you stopped SMBUS and PCI detection in CPUZ so it's now picking up wrong values? Have you clocked one core to 14x 3G while the rest are at 11.5x 2.5G?

That's a 3.059GHz SPi 1M run there, AOD is well known to report downclocked values when they're not actually downlocked but the CPU speed will be as CPUZ reports it, NB speed will be what AOD reports it and so will HT. I can't make out the multi but switch on the PCI/SMBUS detection in CPUZ and validate it and we could tell what the various speeds are pretty easily.

[Daveburt714]

Just... I'm not trying to be a man, but it sounds like you pretty much abused that chip.

Won't you feel bad about RMA'ing it? I know I would.....

Maybe it's just my nature...

[Daveburt714]

I'm starting to think the BE's might actually be "Cherry Picked" for their NB speeds rather than Core Speeds....

I've tried all sorts of combinations and the NB seems to get squirley around 2100Mhz. My chip is not a BE though, so I'm limited to 1.25v...

I did manage to squeeze a little bit more out of it, It hasn't crashed for a couple days... Seems stable..

[KTE]

For the dead chip; well any oc is totally out of spec and anything can happen after that, MFG didn't guarantee it, we take risks and it's our fault what happens. Hardly any user wants to accept their liability though and the easiest way is to always just get a new chip free so that's what they do. Problem is, if AMD inspect that chip, they can tell if it was oc'd just looking at the voltage paths.

Dave, as I said much earlier, IMO the 9600BE are the 9700s and 9900s unlocked. The NB speed on these were originally supposed to be 2000MHz. Thus stock volts you can still get 2GHz NB. However most Phenoms I've seen incl. 9500/9600 can do 2.14G NB at 1.25VID. With a voltage bump, they do 2.4G and more voltage gets them 2.6G. Haven't been past that to date and neither have I seen more. I expect all of them to do 2.4G with voltage but without voltage the limits are the same on all I've seen yet.

However, 100% stable, 2G seems to be the limit for 1.25VID NB.

[Daveburt714]

Yeah KTE, I really think I have this chip pretty well tweaked out, hey, it was 6 weeks of fun!

The sad thing is, I'm seriously thinking about spending some of my Tax Refund on a BE, and saving this 9600 for the HTPC and a 780G board... I'm a computer Junky man...

I do worry about this mobo's PWM though. I'm going to MicroCenter this weekend and buying some heatsinks for those Inductors. It's kinda hard to figure out where the blame lays with Phenom performance...

BTW, thanks for your concern about the tornado's... I live in Cincinnati, it was the same line of storms, but the devistating stuff was well south of here...

[justapost]

Just... I'm not trying to be a man, but it sounds like you pretty much abused that chip.

Won't you feel bad about RMA'ing it? I know I would.....

Maybe it's just my nature...

For the dead chip; well any oc is totally out of spec and anything can happen after that, MFG didn't guarantee it, we take risks and it's our fault what happens. Hardly any user wants to accept their liability though and the easiest way is to always just get a new chip free so that's what they do. Problem is, if AMD inspect that chip, they can tell if it was oc'd just looking at the voltage paths.

Must say I still feel been tricked by them with the tlb-fix impact. As I bought him the problem was known and public but the scale of speed decrease was made public later.
Yeah i abused that chip but at the time the chips died i did not run so out of spec parameters. 2.62GHz/2GHz at 1,375V/1,3V did not make a chip die from all i have read.
It was the first day I could apply more than 2,25V to the ram and ram timings /voltages was what i was playing with as the chip died.
To me it looks like the chips died from too much voltage for the ram, whom ran only 0,05V above what they are rated for.
So i don't feel bad trying to rma that proc, they still can inspect the chip and decline my request. If they find out it was that memory issue killed my chip i guess they'l rma him.

[KTE]

Yeah Achim, I'd also like to see if and how it was the DRAM volts that killed the chip priming. I mean 2.25V is not a lot at all.

[justapost]

Yeah Achim, I'd also like to see if and how it was the DRAM volts that killed the chip priming. I mean 2.25V is not a lot at all.

Three things where different that day.

1. I did that temp torture.
2. Applied more than 2,2V to the ram
3. Ran at 2,6GHz (max before 2,55GHz on the M2A-VM)

AMD will indeed detect that i ran at high voltages. I don't know how close they inspect those chips they get for rma and if they can detect those memory caused failures. I assume they'l decline the request just because of the voltages and don't look any further, so just in case i ordered an 9600be.

EDIT: these memory issues seem to have gone now, have not heard of chips dying in the last time. But mine was from an very early batch can be only thoose early ones where affected.

[KTE]

Which memory issues have gone now Achim?

[justapost]

Which memory issues have gone now Achim?

I mean phenoms dying cuz of too high memory voltage. Last thing i heard has that an microcode update should fix this issue.
You remember bingo13, he said he had a few dying cpu's whom had an early batch version. The ones he got back from rma where from an later batch and worked flawless.

[KTE]

Yeah his 0743 died but 0744 didn't but I had a 0743 at the same time too and ran 2.2V-2.9V on for a long while and nothing happened. Far too many others did, no one but one or two brand MFG workers reported these dying in moments. No common user has reported such problems to date and most run 2.2-2.4V every day.

The problems was also when they ran low CPU volts and plus 2.2V on RAM adjacently and upping the CPU voltage would offset this. You had 1.4V n CPU, so this is why it can't be and you only had it for a very short while.

Your case is very different to theirs TBH. They had done nothing but plus 2.2V RAM and killed their new chip in moments as they said, but you've had yours for long and ran it at high temps/volts/speeds above stock, especially the temps many times were high (I would take those as +60C). That means it's almost impossible for us to rule out how and why that no-boot occurred.

[Oldguy932]

No, that was done with a regular 9600. My mobo actually locks the speed of the nb at 9*htt if I set the nb multi higher than 9. I realized that the BE was completely unlocked after I bought it, and it saddened me even more when it only hit 235htt max and it needed the 8x nb multi just to do it.

The point of increasing the nb multi isn't to make the nb faster, its to make the L3 cache run faster. Try it again for the heck of it, and I'll get some screens of everest and the cpu-z latency thing showing the difference.

[KTE]

No, that was done with a regular 9600. My mobo actually locks the speed of the nb at 9*htt if I set the nb multi higher than 9. I realized that the BE was completely unlocked after I bought it, and it saddened me even more when it only hit 235htt max and it needed the 8x nb multi just to do it.

Wait a minute... you can run NB at 8x with 9600 BE? Are you sure you're not getting them mixed up?

The 9500/2x9600 I had all run 9x NB multi max but are downward multi unlocked.
The 2x9600BE I had both are downward multi locked but upward unlocked.

Same MB, the same BIOS and the same as I've seen with many other users so far.

The 9500/9600 show high NB multi only in CPUZ (false). They never change above 9x actual.
They then show downclocked CPU/RAM speed in AOD/EVEREST but not in CPUZ.
CPUZ will not verify those speeds because they are all over and not true.
Same reason why you can't change HT to higher than 9x on any MB.
The actual CPU/RAM/HT speeds are how CPUZ shows it, but the NB speeds are as AOD shows them and not as CPUZ.

We've been through this a few months back so I can recognize the speeds clearly. That's why I know your 1M run was at 3GHz rather than 2.5GHz as its a known AOD bug. Like this: http://img219.imageshack.us/img219/7721/jsjssk9.png

False NB speed, correct everything else.

The point of increasing the nb multi isn't to make the nb faster, its to make the L3 cache run faster. Try it again for the heck of it, and I'll get some screens of everest and the cpu-z latency thing showing the difference.

CPUZ latency is synthetic and shows faster latencies based off CPUZ readings, nothing else will show any gains with 9500/9600 NB multi above 9x. Run a real-life benchmark and you'll see no change in NB speeds above 9x multi for Phenom 9500/9600 but you'll see big changes for 9600BE above 9x multi. Because for 9500/9600, they're not changing but they are for 9600BE.

Already done many tests many pages back in this and a few other threads.
The last time I tried NB oc was with the BE:
http://www.xtremesystems.org/forums/...&postcount=958
http://www.xtremesystems.org/forums/...&postcount=961
http://www.xtremesystems.org/forums/...&postcount=964

Do some real application tests and check the results. Higher NB will be better results. Easy way to check is, run EVEREST mem/cache bandwidth and it'll show a high NB compared to a low one easily.

Max is 9x NB on Phenom 9500/9600, that's why you can't move HT above 9x on any MB with those CPUs. And max you'll have with stock NB volts is lower than what you get with the 9600 BE with added voltage/multi. Phenoms that can have plus 9x NB multi can run higher HT multi easily, like 9600BE. If you have a 9600BE, try it.

My 9600BE boots 250HT ref. max at 9x NB multi.
9500 did 272HT ref. max at 6x NB multi.

[justapost]

Yeah his 0743 died but 0744 didn't but I had a 0743 at the same time too and ran 2.2V-2.9V on for a long while and nothing happened. Far too many others did, no one but one or two brand MFG workers reported these dying in moments. No common user has reported such problems to date and most run 2.2-2.4V every day.

The problems was also when they ran low CPU volts and plus 2.2V on RAM adjacently and upping the CPU voltage would offset this. You had 1.4V n CPU, so this is why it can't be and you only had it for a very short while.

Your case is very different to theirs TBH. They had done nothing but plus 2.2V RAM and killed their new chip in moments as they said, but you've had yours for long and ran it at high temps/volts/speeds above stock, especially the temps many times were high (I would take those as +60C). That means it's almost impossible for us to rule out how and why that no-boot occurred.

It where the 9600's dying immediately the 9500's required more time. Can be that issue is somehow related to the 790FX chipset and that's why I had no problems with 2.2-2.25V on the M3A.
The chip is rated for max 1,5V and 70°C. It's really odd that he died at ~50° and 1,375V. But nevermind i joined the 9600BE-bingo, just receiped the confirmation that the 9600BE i ordered few hours ago is already on his way. With a little luck i'm on k10 again tomorrow.

A DMM with a temp probe just arrived. If i put the probe on the BE2400's IHS I measure ~27°C. Smart Guardian reports 42° C. That's only a 15° difference whould mean my phenom ran at 85°C minimum when SG reported 100°C.
Must say I expected prime to fail before the cpu dies if that high temps where the cause.

Once we had an air con problem on a hot summer in our server room.
CPU temps where above 85° every now and than case temps above 60° where also no rarity and the servers powered off on a daily base. A few HD's died but no cpu or other component.

[KTE]

Achim: if you measure IHS temp, you'll have to modify the IHS and install HSF on top sitting with a good flush contact and the probe underneath at the geometric center of the chip within the IHS. So the temps will be higher than taking any side measurements. X2 works differently to K10h since the core DTS can give different temperature feedbacks and the BIOS interprets them differently.

SmartGuardian is +20C wrong pretty easily looking at the various values. The next time you get the Phenom, open HWMonitor as well to checkup the various readings. Plus your CPU temp in EVEREST only showed ~53C actual around -10C with M3A readings and what every other Phenom gets. Every Phenom I've seen so far gets far lower temps than those at similar volts/speeds.
Measure the RD790 temp when you get Phenom, that was giving it a 52C reading. From personal experience I know RD790 is a 10W TDP chip and never reaches even 40C at <30C ambient, so those are near 32C actual temps it was reading as 52C, a +20C offset.

I don't think the temps/speeds killed the CPU IMHO, but I'm not sure if feeding the CPU/RAM high volts and priming it did. What's the max CPU VID and voltage you've ever fed it?

Kinda strange why it doesn't boot.

BTW, your sig reads you have a Phenom 9500 running a Phenom 9600 multi.

[justapost]

Achim: if you measure IHS temp, you'll have to modify the IHS and install HSF on top sitting with a good flush contact and the probe underneath at the geometric center of the chip within the IHS. So the temps will be higher than taking any side measurements. X2 works differently to K10h since the core DTS can give different temperature feedbacks and the BIOS interprets them differently.

That's why I said minimal. If you look at my 100°C screeny you see that smart guardian seems to read from what everest reports as aux and seems to be an externel sensor in or below the socket so the reading should be cpu independant.

SmartGuardian is +20C wrong pretty easily looking at the various values. The next time you get the Phenom, open HWMonitor as well to checkup the various readings.

Must use somthing different than speedfan (kills pci functionality on dfi mobo) hwmon will be a good replacement.

Plus your CPU temp in EVEREST only showed ~53C actual around -10C with M3A readings and what every other Phenom gets. Every Phenom I've seen so far gets far lower temps than those at similar volts/speeds.

That cpu temp on the sapphire mobo is indeed strange. The mobo allows to specify an offset for temp reading. I left it at the default 0. Never checked that offset in the cpu registers on the m3a.

I don't think the temps/speeds killed the CPU IMHO, but I'm not sure if feeding the CPU/RAM high volts and priming it did.

Kinda strange why it doesn't boot.

He died as mysterious as he lived.

What's the max CPU VID and voltage you've ever fed it?

Ever? 1,525V on the M2A-VM in the first week, later never more than 1,475V on the M2A-VM (max stable voltge) and not more than 1,425V (few minutes) on the sapphire.
A few days before I just had verified my oc results on the m2a-vm and could not find any signs of long term degradation.

BTW, your sig reads you have a Phenom 9500 running a Phenom 9600 multi.

Ups, fixed don't have a 3GHz phenom running.

[Oldguy932]

I'm just wondering on this, but has anyone noticed an effect on their max htt, max nb speed, or max cpu speed from using chilled water or phase on their cpu? I remember reading about macci doing some dice runs, but does that mean I won't have a cold bug? I have a vapochill, I just need to order the am2 mount for it and I can do some testing with it. If no one knows...I guess I can be the guinea pig.