To all Phenom users with x64bit OSes

[Lightman]

Hello guys and girls!

I have issues getting my Phenom 9600BE above 2.4GHz stable in 64bit systems.
Nothing unusual you may think, but this is third Phenom I've put in my system so far with identical stable frequency limit!

My first Phenom was 9500 from 0743 week which was 3DMark stable up to 2.64GHz in Windows MCE. On XP x64 max. I could boot into system was 2.5GHz, but when I started any game or taxed system with some 3D raytracing stability was gone. Quickest way to hang up was Crysis x64 executable (1-5 minutes in game). I started lowering frequency to final 2.42GHz which was stable enough.
When I ordered my 9600BE I was hoping for a bit more seeing other samples on XS. My CPU came, I've checked production week (0749) and happily swap my old Phenom with new one. BlackEdition couldn't even boot into desktop . At that time I had only 64bit OS so I haven't tried normal XP or Linux with it.
Two days ago I've received another Phenom 9600BE, this time 0750 and in short:
2.42GHz under XP x64 and Vista x64 as max reasonable stable freq.!
2.6GHz easy boot into XP MCE desktop and to run couple of benches.
2.8GHz @1.35v stable enough to to run full MemTest suit

As a note I've tried all voltage possibilities GA MA790FX-DQ6 gives (up to 1.55V on CPU).
Phenom 9600 @2.5GHz/1.25V - not stable even for 3DMark or Cinebench
Phenom 9600 @2.5GHz/1.45V - look above ...


As you can see I need your feedback. Can you show me any Phenoms running 64bit systems clocked above 2.5GHz?
Is B2 revisions not optimized enough for higher frequencies when 64bit data path is used?
Is that a reason why AMD didn't even show Barcelona server counterpart above 2.0GHz?


Thanks for any help

[gOtVoltage]

You are not the Only one....My 9600BE is acting buggy as hell in xp64bit...it does play crysis well though...It dosnt seem to load drivers good when i raise the FSB... i only get 2400mhz windows stable ,Thats 210fsb or i use AOD..u.. disable..I believe CnQ and AMD driver is at fault in Xp64bit..This is truly a PHENOM X64bit windows problem..

I was trinking buying 32bit pro or vista..For now im keeping it at 2300mhz and waiting to see if Biostar gets a more optimum Bios soon .

[justapost]

Currently trying Windows Server 2008 64bit. If I install at 2.6GHz I get the famous prozessor interrupt bluescreen. Tried it with and without HPET. Now at 2.3GHz with TLB-Cache disabled it works (still installing).
Will try install at 2.6GHz with TLB-Cache disabled later to sort out Err. 298/256.

BTW: Glad to heard you finaly got an working BE.

Update: 2.6GHz with TLB-Cache disabled also bluescreens during install.
Debian lenny 64bit works fine at 2.6GHz, running linpack on four cores with no problem.

Second Update: 2.6GHz are not stable on debian 64bit. Got this output:

BUG: soft lockup - CPU#2 stuck for 11s! [events/2:17]
CPU 2:
Modules linked in: fglrx(P) rfcomm l2cap bluetooth ppdev parport_pc lp parport ac battery ipv6 cpufreq_powersave cpufreq_stats cpufreq_userspace cpufreq_ondemand cpufreq_conservative freq_table nfs lockd nfs_acl sunrpc it87 hwmon_vid firewire_sbp2 loop snd_hda_intel snd_pcm_oss snd_mixer_oss snd_pcm snd_timer snd soundcore i2c_piix4 snd_page_alloc i2c_core sky2 psmouse serio_raw button pcspkr tsdev evdev reiserfs dm_mirror dm_snapshot dm_mod ide_cd cdrom generic sd_mod atiixp usb_storage ide_core ahci ata_generic firewire_ohci firewire_core libata crc_itu_t ehci_hcd scsi_mod ohci_hcd thermal processor fan
Pid: 17, comm: events/2 Tainted: P        2.6.23-1-amd64 #1
RIP: 0010:[<ffffffff8021a2e4>]  [<ffffffff8021a2e4>] __smp_call_function+0x62/0x81
RSP: 0018:ffff8100378bbe10  EFLAGS: 00000297
RAX: 00000000000008fc RBX: 0000000000000003 RCX: 0000000000000001
RDX: 00000000000008fc RSI: 00000000000000fc RDI: 000000000000000b
RBP: 0000000000000000 R08: ffff8100378ba000 R09: ffff810001024168
R10: ffff810065efd7c0 R11: 00000000000000d0 R12: ffff810001021850
R13: 0000000000000000 R14: 00000002378bbda8 R15: ffff8100378bbea0
FS:  00002ad5bca6a250(0000) GS:ffff81007e2dc540(0000) knlGS:0000000000000000
CS:  0010 DS: 0018 ES: 0018 CR0: 000000008005003b
CR2: 00002ad5d7f54000 CR3: 0000000000201000 CR4: 00000000000006e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400

Call Trace:
 [<ffffffff802150a4>] mcheck_check_cpu+0x0/0x30
 [<ffffffff8021494a>] mcheck_timer+0x0/0x7c
 [<ffffffff8021a60b>] smp_call_function+0x32/0x47
 [<ffffffff802150a4>] mcheck_check_cpu+0x0/0x30
 [<ffffffff802392f3>] on_each_cpu+0x10/0x22
 [<ffffffff80214967>] mcheck_timer+0x1d/0x7c
 [<ffffffff8026e4d6>] vmstat_update+0x0/0x32
 [<ffffffff80243481>] run_workqueue+0x7f/0x10b
 [<ffffffff80243db8>] worker_thread+0x0/0xe4
 [<ffffffff80243e92>] worker_thread+0xda/0xe4
 [<ffffffff80246b96>] autoremove_wake_function+0x0/0x2e
 [<ffffffff80246a79>] kthread+0x47/0x76
 [<ffffffff8020cc08>] child_rip+0xa/0x12
 [<ffffffff8021dcea>] flat_send_IPI_mask+0x0/0x4c
 [<ffffffff80246a32>] kthread+0x0/0x76
 [<ffffffff8020cbfe>] child_rip+0x0/0x12

[LIKMARK]

I'm a victim of bad Phenom clocks under vista x64 too, although I'm running it on an AM2 board. For now. Nothing is stable for me.. Thinking about buying Abit AX78 and reinstall XP 32bit too..

[Debello]

Been running the 9600BB since it was released on the MSI K9A2. At 2.5ghz or above the computer is completely unstable (same issues with a 9500). At 2.5Ghz I am able to run 3dmark or cinabench with out issue, but when the computer is idle I will hear the HDD spin up and then I get the BSOD. Which almost makes me thank that there is an issue with the Phenom, Vista 64 and the SB600.

I have played with the voltages everywhere from 2.5V to 5V, with out being able to get any sense of stability. I have a 780G mobo coming I am interested in seeing if the Phenom overclocks any better with the SB700 instead of the SB600

[MusicIsMyLife]

I´m running a Phenom ES in a review-rig. Currently I´m using vista 64 ultimate without tlb-fix. Installing vista at 2.4 ghz was absolutely no problem. Yesterday I tested max clocks (short-term-stability with 15 minutes prime x64) and i got these clocks:



I can run any benchmark (except crysis -> that doesn´t work for me, even at stock clocks ). Everything else works fine, no bluescreen, no freeze, no reboot.

But ok, it´s an ES...

[Lightman]

I was running 2511MHz on my old 9500 for almost 2 weeks. It was semi stable, but I had one or two blue-screens (famous interrupt error) during a day. I could run 3DMark06 test, but not full suit. Crysis was most affected by instability. Cinebench x64 or LW9 no problems, even AOD stability test was passing 8h stress!
I've even did couple of test @2.6GHz under x64 but only with 2 or 3 cores at these clocks and 1 or 2 below 2.4GHz.

There is one thing I haven't tested. How increasing CPU NB vcore affect stability. My board doesn't have an option to adjust it.

With regards to ES samples I think they are using different manufacturing parameters to make them clock faster. In other worlds if you will measure power consumption of AMD Phenom ES and retail Phenom at the same clocks I'm sure ES will use significantly more power. I saw this behavior in one or two reviews. I suppose AMD decided to pursuit performance per watt as a logical thing to do now till buggy rev. B2 will be replaced by B3 or even 45nm successor.
Because ES samples are built using faster switching transistors (more gate leakage) they can clock a bit higher in x64 as well. If thats the case then we are dealing with slow path that needs to be addressed by new revision. What is strange is why clocks are not scaling at all with voltage increase and even seems to be scaling a bit with voltage decrease? Hot spot? Anyone with chilled water or sub-zero cooling to investigate that?

@Justapost I'm glad as well. This is another I bought, my previous one is still at Scan RMA.

[JumpingJack]

Well, I cannot contribut too much ... I have been successful with the 9600BE installing 64-bit Vista and Ubuntu 7.10 distribution, the Vista was prime95 stable as well as Everest stress stable (~ 20 hours as a recall, I was not systematic and did not write down much info, just playing).... I do not recall if I tried an OC.

Jack

[KTE]

Aye, glad you got the Phenom you've waited for.

There is one thing I haven't tested. How increasing CPU NB vcore affect stability. My board doesn't have an option to adjust it.

Lightman, how are you clocking high without feeding NB volts? Or are you dropping NB speed when you clock HT high?
Most Phenoms will need NB volts to clock decent using HT and all would require them to clock past 2.2GHz NB stable. My present one would crap out at 2GHz all first 5 weeks but now 2.214GHz at 1.250VID (1.232V) is something I can run fully stable. This again is weird, I made it very clear when I first got it that it would not boot plus 2GHz NB for 2 weeks and now after over 6 weeks of hard use and testing it runs fine 2.214GHz 1.250VID. I found it by chance too, didn't look for it because I didn't expect it from previous testing.

With regards to ES samples I think they are using different manufacturing parameters to make them clock faster. In other worlds if you will measure power consumption of AMD Phenom ES and retail Phenom at the same clocks I'm sure ES will use significantly more power. I saw this behavior in one or two reviews.

Gotta agree with you there. Something is iffy with ES samples which clock higher at lower volts, the power use in them is not like the ones I've had and that's 5 now. You can see the one I have in total system power use at 2.626GHz here after tweaking: http://www.xtremesystems.org/forums/...12#post2810312

I must admit after almost 7 weeks with this chip (or more), I'm still not even started on it. The problem is, it reacts totally different at the beginning, to the middle, compared to after 6-7 weeks. So you have do redo everything again as I'm doing now for the nth time. Stable first 3-4 weeks is not stable later, with all my chips so far and of some others. The biggest oddity is, you might have 2.8GHz 100% stable for the first 3 weeks in all conditions you honestly tested but after that, one day, at a random attempt, you may find for 5-6 days that no matter what volts you cannot even boot the same MHz with the same OS, no modifications, same drivers, same everything. I am totally stumped, first time for me and I can't figure it out no matter how hard I've tried.

Then suddenly she makes a live connection to my dream to reveal to me she's still virgin and after some more... so I go at her once again.

I will mention one problem in stability though.. ATi drivers. For some reason or another, they crap out or become slightly corrupt after a few crashes and lockups, which make your system very unstable even if the CPU/RAM is stable.
In the link above that's my lowest volts fully stable so far, i.e. no problems whatsoever and I've tested very stringently this time due to the problems with many users including myself before. I've ran Gutsy Gibbon and Feisty Fawn on them before. I'm about to try running some VMs and then I'll try installing something 64b at those MHz to check it out. Here is what is my latest testing and it's fully stable so far (waiting for the idling freeze to occur): With CnQ.





This was not possible before (without the freeze) for over 7 weeks! But now I've been testing low volts for about 9 days and they are working much better than high volts.

All I can say is, just don't give up and make quick conclusions but work with it. Because it has shown consistently not to always behave the same and proven my findings wrong repeatedly. It's very strange and hard to predict.

Hot spot? Anyone with chilled water or sub-zero cooling to investigate that?

Unfortunately, I don't really want to guess where in the process the problem lies nor delve on what's been covered many times in the appropriate sections, it would be too random and juvenile a guess even though I know pretty much the little data we do have on K10h, but the problem with clocks is in the chip itself. Cold doesn't make a difference to mine for better clocks than air. I would assume it has something to do with the respective embedded channel stressors and the chip materials used at various gates. SiO2 dielectric will just be too thin, almost 3 atom layers thin at this level, which will leak immense current and also deplete (degrade) very quickly, experiencing the tunneling effect and in this situation, the electrons are not only in a particle behavioral state but as waves too, so there's no guarantee of controlled behavior over such a transistor - EM is also worst here (especially with increased voltage). A thicker metal gate and dielectric replacement would allow them to decrease FO MOS, latencies of various CPU subsystems, a combination of mix alloy stressors to increase FET drive ability and increase frequencies while keeping gate and source-drain channel leakage lower than at this node. That's my opinion.

I would install 64b to test right away if I had a spare drive here. Unfortunately I don't, all 7 are booked with crucial data. Maybe I can partition this one more and install Linux distro dual boot 64b now... hmm. Have left playing with Linux for a whole now. What's a good 64b Linux distro, good support for Phenom+RD790 and little hassles installing?
If it has a good manageable bootloader, it would be better. Don't want to mess with bootup hassles at this stage.

[Boschwanza]

Just set up XP64 on my second hard disk, installing with 2.65 GHz without problems. I will test a little bit and give more feedback

EDIT:

XP64 failed booting on 2.7 GHz, 2.64 GHz works fine without any crashes. Will try some prime tomorrow.

[Grandpa]

I have 9600BE and running duel boot, Windows XP 32 bit is 8hrs prime stable at 2.639 203 X 13 If I go anything above stock on Vista 64bit it will not boot all of the way or it will freze as soon as it boots into Vista. It does the same thing on both the MSI K9A2 Platinum motherboard and the DFI LP UT 790FX-MR2 board. So it is a 64bit issue.

[Daveburt714]

I experienced the same thing Lightman, with 2 chips and 2 boards..

Using Vista64 my 9600Std (0744MPMW) seemed to hit a limit at 2.5Ghz (250x10x8x8) on an MSI K9A2 (may it rest in peace )... I bought a 9600BE (0750CPMW) thinking the chip was the problem, hit the same wall under Vista64..

I changed the OS to XP32 and easily got to 2.6Ghz stable before the board died (during beta BIOS flash).

I've since bought a Gigabyte DS5 and the BE seems to run smoothly @ 2.66Ghz under XP32 (213x12.5x9x9). I haven't had a chance to put the 9600Std in the new board under XP32 yet, but I have a feeling it might be a better chip than I initially thought...

The only bad thing about XP32 is you dosen't get the benefit of DX10 and I can't use all 4Gb of my Memory.

*Side Note: WGA is a PITA, Guess I'm going to have to pick an OS and stick with it...

[KTE]

OK. Just dual booted Ubuntu 64b at 2.639GHz. One way to check what the problem is to install it with a speed. If it's the OS itself, installation will pass easily but not the OS loading/functioning.
Flew through the install without problems. Booted too. But my GFX is not supported, working in 8x6 is horror and every other second, a certain application kept freezing up with low power load. Very buggy OS even at stock with TLB Fix.

I had to restart in the middle around 4 times and that kept destroying the HD/OS with errors. Next instance I had it sorted, 2.639GHz would always hang at loading Ubuntu at 135W AC load. That's 100% 32b stable speeds. Voltage upto 1.45V didn't make a difference whereas in 32b Linux/Windows, I ran it at 1.248V. But 2.626GHz would work error-free and it ran for nearly 9 hours now without fail. 64b LiveCD ran perfect and Memtest latest version passed 1hr before trying.

Gutsy 64b had higher idle and load power consumption.

[justapost]

OK. Just dual booted Ubuntu 64b at 2.639GHz. One way to check what the problem is to install it with a speed. If it's the OS itself, installation will pass easily but not the OS loading/functioning.
Flew through the install without problems. Booted too. But my GFX is not supported, working in 8x6 is horror and every other second, a certain application kept freezing up with low power load. Very buggy OS even at stock with TLB Fix.

I had to restart in the middle around 4 times and that kept destroying the HD/OS with errors. Next instance I had it sorted, 2.639GHz would always hang at loading Ubuntu at 135W AC load. That's 100% 32b stable speeds. Voltage upto 1.45V didn't make a difference whereas in 32b Linux/Windows, I ran it at 1.248V. But 2.626GHz would work error-free and it ran for nearly 9 hours now without fail. 64b LiveCD ran perfect and Memtest latest version passed 1hr before trying.

Gutsy 64b had higher idle and load power consumption.

You need the fglrx drivers for the hd2600xt, last time i tried ubuntu that packages where available. Also compiz fusion worked flawless and without the need to edit xorg.conf the first time.
Have you inspected the logfiles if an app freezes? (dmesg or /var/log/kern.log)
It's stable at 2.5GHz I assume.
As for power consumption there is no k10 cnq driver for linux yet.

[KTE]

You need the fglrx drivers for the hd2600xt, last time i tried ubuntu that packages where available.

Yea I removed the old fglrx drivers as they did not mention support for 2600XT and 2600XT is known to have problems with Ubuntu 64b (not 32b, 32b is flawless) and tried installing the new downloaded package from AMD's site. No go, it says file/directory not found while it's there.

Have you inspected the logfiles if an app freezes? (dmesg or /var/log/kern.log)

Too many files are corrupt and unrecoverable after reboot and after running fsck they're either not there or empty. The problem is mainly with two apps, if I don't touch Add/Remove/Synaptic or Restricted Driver Access, I get no crash/bug. So I can't install or search much.

It's stable at 2.5GHz I assume.

The hangs happen at stock and at any MHz but 2.626GHz is max stable if I don't use those above apps.

As for power consumption there is no k10 cnq driver for linux yet.

Yeah but without CnQ, the power consumption is av. 11W AC higher idling in Gutsy 64b than XP 32b. There's also a lot of background happenings which make power be from 122-135W AC regular while XP 32b idles 108-115W AC even with Fx and all basics running.

I'll need to check if there's any software to support reading system speeds in Linux now.

[justapost]

Yea I removed the old fglrx drivers as they did not mention support for 2600XT and 2600XT is known to have problems with Ubuntu 64b (not 32b, 32b is flawless) and tried installing the new downloaded package from AMD's site. No go, it says file/directory not found while it's there.

The amd package's install script has an option to build packages for distributions. Build those for ubuntu the apt-get install the module-assistant package and use it to build the fglrx kernel module.
You can use dpkg -i [packagefile] to install the packages.
Chances are good that you need a few development packages before the fglrx packages build successfull, at least x-dev should be required.

To install an package use apt-get install [packagename], to search use apt-cache search [packagename].
To get the restricted packages it might be as easy as commenting out the corresponding line in /etc/apt/sources.

[KTE]

That's what I've been using, sudo apt-get update and then sudo apt-get install linux-restricted-modules-generic restricted-manager allowing access to the restricted packages, and then sudo apt-get install [packagename] but the ones I can find aren't being installed and my card is being shown as unsupported. The problem seems to be with the xgl server and opengl not being initialised. I'll debug it when free but I don't like being stuck in 8x6 mode for long. If I get the output for fglrxinfo with LIBGL_DEBUG=verbose it should give me the required details I'm after. Removing xgl server with sudo apt-get remove xserver-xgl and then if I restart X, it should start working... hopefully.

I think I also forgot to add the x capability to sudo chmod +x ati-driver-installer-8[..] (sigh)

Thanks.

[justapost]

That's what I've been using, sudo apt-get update and then sudo apt-get install linux-restricted-modules-generic restricted-manager allowing access to the restricted packages, and then sudo apt-get install [packagename] but the ones I can find aren't being installed and my card is being shown as unsupported. The problem seems to be with the xgl server and opengl not being initialised. I'll debug it when free but I don't like being stuck in 8x6 mode for long. If I get the output for fglrxinfo with LIBGL_DEBUG=verbose it should give me the required details I'm after. Removing xgl server with sudo apt-get remove xserver-xgl and then if I restart X, it should start working... hopefully.

I think I also forgot to add the x capability to sudo chmod +x ati-driver-installer-8[..] (sigh)

Thanks.

No more need for the xgl server at all, the newer fglrx packages support AIGLX, if you add the fglrx module to the white-list in the compiz executable (it's a script).

My cpu softlockup was fglrx related, had one at 2.5GHz today, stuff run's fine at 2.3GHz atm.

Gonna inspect that lockup issue later, have to reorganize stuff on three 500GB hd's today that might take a while.

[alexio]

Could it be that Phenom limits current draw at a certain level? That would explain why it's so sensative to vcore tuning. Maybe if that's the issue there's a way around it (like drive strength tweaks with RAM).

If the current draw is indeed limited (in some parts of the CPU?) that may also explain why some day the CPU is stable at certain settings, the next it's not and the day after that it's stable again. A max. current limitation means the CPU is a lot more sensative to temperature changes.

To remain energy efficient AMD has every reason to limit current draw in (parts of the) CPU. I heard that the 9500 often clocks higher than the 9600BE. Maybe AMD lowered the roof for current draw to prevent people damaging the CPU as the BE was a rushed-out product using a very new architecture.

Just some wild theories....

[Lightman]

So it looks like very few Phenoms can run above 2.5GHz in 64bit mode but majority will not.

I hope to get my other 9600BE this week from RMA. Another sample to test .

Anyway thanks all of you! Good to know that I'm not alone with my issues

BTW I've tried 1.25V for 2.5GHz but I couldn't boot into windows. Looks like 9500 was better in that regard.

@Boschwanza - can you say what week of production and stepping?
@Alexio - interesting theory

[KTE]

Could it be that Phenom limits current draw at a certain level? That would explain why it's so sensative to vcore tuning. Maybe if that's the issue there's a way around it (like drive strength tweaks with RAM).

DS strength is mainly important to RAM/IMC stability but the currents are only limited to 110A for all cores and 20A for the IMC. This'll remain true for AM3 CPUs too. You're on a similar length to what Achim and I suspected earlier, as we saw values of Idd and IddDiv differ at different settings and they're set at bootup only. They are the currents for the CPU/NB at any setting.

Achim has a very similar setup to mine and his power draw idle/load is much more than mine even while I'm at higher volts/clockspeeds. The difference is in two separate TDP classes. Not really tested why yet... but I've told my uncle to reserve me a Gigabyte 550W Odin to check out the amps my CPUs using today. Good that he agreed.

So it looks like very few Phenoms can run above 2.5GHz in 64bit mode but majority will not.

You sure about that?

Take a survey, I think you'll find many can run above 2.5GHz easy on 64b. However it is fudged up, and mainly for Vista 64b. I can't boot above 230HT on it using the Ultimate Edition 64b.

Have you tried a Linux distro? It'd be wise to test a 64b one and compare against Vista.

Achim I'll be troubleshooting it in a bit. On lunch break ATM.

[Lightman]

@KTE - I've played OpenSuSe 10.3 x64 and it was not stable/booting at 2.5GHz+ on my old 9500. With 9600BE I could not even install it at 2.5GHz. Haven't tried yet @2.4GHz because I'm away today and tomorrow (welcome Sheffield ).

BTW I have MSI K9A2 Plat rev. 2.0. Anyone tried this one? What's different from older rev.?

[KTE]

The 96BE was 100% 3/4 day stable at 2.5GHz when you tried installing/booting OpenSuSe 10.3 64b?

I'll try Vista X64 again. Just need some days to test max stability. If I can get more than 2639MHz I have now.
Sheffield... I'm not too far off since yesterday morn myself.

K9A2 Plat rev. 2.0: not tried it, what are the specs and components on it?
There's nothing much on it around either so I don't know. Would be interested to see the board and compare.

Pics please

[justapost]

Hmm tied latest fglrx driver and vesa did not solve the CPU softlockup. Used different kernels and i'm still geninng the same error.
Must be the cpu i guess no lockup's at 2.3 and 2.4 GHZ.

I think rev 2 has no sas support.

[madfaze]

esata was removed from the IO panel