Phenom 9500 w/ MSI K9A2 Platinum

[justapost]

Guys, you'll have to excuse me with time because my systems have all failed to boot. No idea why. It happened when I tried changing PSU.

Phenom is not booting at all in MSI or Abit board with 3 different PSUs, same problem as before when you switch power off (defaults), stuck in reboot cycle. Spent 4 hours on it.

Gigabyte PSU is totally DOA on all boards. Tried 4 boards, MEM/CPU doesn't even start.

Two C2 systems also no boot with those 3 PSUs but luckily an older P4 worked with the Antec and the Seasonic which I'm posting off right now. Spent hours troubleshooting this.

At least the P4 idles 55W AC.

Achim: Download GBT 550W PSU guide here, there is no guide but on CD in box and you might find trouble installing it (it has many things): http://america.giga-byte.com/FileLis...s800a-d1_e.pdf

Sounds like a weird weekend start over there. Thank you for the guide, with luck the psu will arrrive today. Have you tried what's described in there at the last page.

sidenote, sounds like the monitoring can be done on a different pc, as i hoped.

Celeron system idles at 43W AC here atm.
M2A-VM with phenom 9500 and onboard graka was ~75W AC.
With the BE2400 ~55W AC.
All setup's used different hardware but it was always onboard gfx, one hd, one dvd drive and no extra fan's. All had power saving features enabled.
K8 has p-states i guess maybe the msr tweaking works there also.

[KTE]

I notice the HT can be set now, I'm wondering, does changing that also raise the NB/IMC multi to match?

Nope. You can change NB multi through WPCREDIT/BarEdit though.

I think what I'm gonna do is bug MSI's customer support thing, and see if I can get em to put together a bios that allows for NB/IMC clock and voltage changes like the ones that had the p-states section. Since apparently thats how that ACI.101 bios came about.

That BIOS was the Chinese release only, across the rest of the world, they didn't release it until someone had already spread the link online. The P0J is better. I have asked them twice now for a better BIOS and the options personally. I will post back what they say, so far 3 days and no reply.

Sounds like a weird weekend start over there. Thank you for the guide, with luck the psu will arrrive today. Have you tried what's described in there at the last page.

It's Phenom and the PSU.

Phenom hasn't booted for a day now with 4 PSUs in 2 boards when I cleared CMOS. At the memory stage (C1 LCD) it just reboots.

PSU I'm going to try again now. Had no time since last post here.

sidenote, sounds like the monitoring can be done on a different pc, as i hoped.

Yep.

Celeron system idles at 43W AC here atm.
M2A-VM with phenom 9500 and onboard graka was ~75W AC.
With the BE2400 ~55W AC.
All setup's used different hardware but it was always onboard gfx, one hd, one dvd drive and no extra fan's. All had power saving features enabled.
K8 has p-states i guess maybe the msr tweaking works there also.

This is with a power meter or with the GBT PSU?

I am going to cross-check power values. Phenoms are like the Penryn in one sense.
-You can't simulate or compare power between CPUs, as in simulate SKUs.
-Your one sample cannot be generalized for all the same SKU CPUs out there unless you can repeat it with another 2 retail samples with different VID/Volts for the same TDP/MHz. Overall, they are only pointers and valid for your sample. Inherently they have variables.
-Reviewers need to start publishing raw data in a separate link available as a download. This way we have too many inconsistencies and errors of theirs go unspotted and waste too long conjecturing wildly what could or could not be.
-Take Penryn for a good example. Penryn has a GHz limit at 3GHz for a good TDP/power/temps. You can see it's extremely tweaked at that, a very high bin, hence it's power/volts/MHz are best possible on that core/step. Nothing else will come close. If you tried to simulate lower or higher models, it will be completely inaccurate since it requires lower volts/power/TDP than any other model for the same frequency. Hence the overclockability. Jump one stage higher at 3.2GHz SKU and you see things turn shoot up with power/temps/volts. Showing the core MHz:TDP limit. Hence why no 65W TDP 4GHz Penryn retail as some delirious individuals hyped, ever. I had said they had a top bin of 3.4GHz internal since June '07, max and it had v.high TDP. But one sided individuals could never accept these facts until too late >
-The same thing exists for Phenom but at lower MHz. At 2.6G it reaches a very high TDP. The power draw/volts/temps for it will be much higher than what anyone can get with a lower model CPU esp. the 9600 BE that are tweaked for oc lower temps/power/volts. 9600 BE will be lower in these parameters than the 9500/9600/9700/9900. The core/step TDP:MHz limit seems to be near the 2.5GHz range. After this, things just shoot up with diminishing returns and poor yields. The FX range of CPUs, if released will improve on these parameters further if higher binned. But they might also require higher current for MFG stability and thus have higher temp/power than a comparable 9600 BE at equal settings and speeds.

Hence, my measurements are only going to apply to my testing and for comparisons, since by now, I know more about the core and its working and holes than the guys who spent 4 weeks or less reviewing them.

Operating 95W TDP CPU, no matter which CPU, will always have the same internal temps. The TDP is what decides the temps, nothing else. Thus 95W Phenoms should have temps the same as Q6600 G0. One easy way to test is let it go on until TCC is activated to throttle/shutdown. This you need to know, to have some calibration of your temp feedbacks and to map things out. Measuring TCASE is not like measuring TDIE as some claim, the metal case is there for a reason, it dissipates heat and has a thermal resistance which reduces the temp delta below and above. This is why internal core temps like the DTS are always higher than TCASE. Infact, at idle without CnQ, 2.3GHz with no HSF, both my 9600BEs measure 23C on IHS @ 21C ambient. Probe used was n-type industry and this was done at my work lab too. That's IHS temp is no way my internal core temp.

I'l go back to testing now, AX78 has very impressivve BIOS.

[KTE]

Here's the problem with the Gigabyte Odin =>

Shorted it, started it up and overrided the OVP and UVP. It's on but not connected to anything and I've linked the USB output to this computer through USB, so I get the idle voltages on all rails. Check this out:




You seen the +5V rail there? It's way out of spec. that is far beneath the lowest allowed without damaging the system. DMM reads same value as that software. So faulty PSU.

[dr_drache]

i'm wondering, arn't I supposed to have 2000MHz HT? not 1800MHz...

[KTE]

Nope, 1800MHz for all released Phenoms.

[dr_drache]

blah, i guess i read something somewhere wrong..... .blah... it's weird, i kinda want to get one of these performance bios's to work, so i CAN overlclock, not that i really want to... it annoys me that i don't have the ability to overclock much.

[KTE]

The higher models are 2GHz IMC and these were supposed to release 2G before bugs were found.

[dr_drache]

as in the 9700 and the 9900s? blah, i kinda wasted my moneys on my BE.

[Mathos]

The higher models are 2GHz IMC and these were supposed to release 2G before bugs were found.

Actually they were supposed to work with an auto detecting adjustable IMC, like the original spec had. Was suppose to auto detect and adjust the IMC clock to within 200mhz of the core speed to avoid the uneven divider problems that the K8's had, which would allow you to run your ram at full speed regardless of processor clock. It was also suppose to be able to clock down, and go into Idle mode like the cores do, with C&Q 2.0. Thats what the whole separate power planes thing was supposed to be for, all that does right now is make it easier to OC.

And a side note, my system has proven to be gaming stable so far, on the auto detect voltage through the P0J bios, been messing about with Portal and Sims 2. Even played Half life 2 a bit, not gonna hit that too heavily though until the new 3870 comes in. At which point I'm gonna see if my system stays stable after putting the new card in, like it didn't do last time. Granted last time I could of just plane had a dud card.

[dr_drache]

how do i get it to 2g?
i have a question, i WAS using SB600 raid for my raid, but i was having... "lockups" whenever the system would read the cd-rom (take 30 sec to show my computer, wouldn't update to new disc when replaced) so i switched to promise. could that be part of my issue with BSOD on the P0J?

is there a way to just switch raid controllers without a reinstall? (like i guess, turn on raid on the SB600, boot, install driver, shutdown, switch sata cables..) would it then be a good idea to place cdrom and storage HDD, on the 2 promise sata ports?

[Mathos]

how do i get it to 2g?
i have a question, i WAS using SB600 raid for my raid, but i was having... "lockups" whenever the system would read the cd-rom (take 30 sec to show my computer, wouldn't update to new disc when replaced) so i switched to promise. could that be part of my issue with BSOD on the P0J?

is there a way to just switch raid controllers without a reinstall? (like i guess, turn on raid on the SB600, boot, install driver, shutdown, switch sata cables..) would it then be a good idea to place cdrom and storage HDD, on the 2 promise sata ports?

It might be, can't remember if something like that was brought up on the MSI forums. If you're running a raid array, you'd need to switch to the promise controller, rebuild a new array, and reformat. I've been considering getting an SATA DVD RW drive, but am worried about it working properly with the SB600 on AHCI mode. But thats part of the problem I see with everyting being migrated over to SATA. Granted the average Joe isn't likely to worry about raid, an enthusiast will.

You can't OC the NB/IMC with bios P0J. You'd have to use bios 1.1b3 or something similar. But 1.1b3 seems to be the best for that. But, not the most stable, why I kinda wish we could get a cross between 1.1b3 and p0j.

[dr_drache]

It might be, can't remember if something like that was brought up on the MSI forums. If you're running a raid array, you'd need to switch to the promise controller, rebuild a new array, and reformat. I've been considering getting an SATA DVD RW drive, but am worried about it working properly with the SB600 on AHCI mode. But thats part of the problem I see with everyting being migrated over to SATA. Granted the average Joe isn't likely to worry about raid, an enthusiast will.

You can't OC the NB/IMC with bios P0J. You'd have to use bios 1.1b3 or something similar. But 1.1b3 seems to be the best for that. But, not the most stable, why I kinda wish we could get a cross between 1.1b3 and p0j.

that's my issue, my dvdrw is a sata drive. current config is, raid 0 array on promise, 500gb and sata dvdrw on sb600, ide mode. never thought of going AHCI (call me passivly stupid) currently on bios 13b3. but with P0H, i get nvidia BSOD, P0J i get storage controller BSOD (goes too fast, but it's right in that area), and i get a x00000101 BSOD or a lockup every time i OC over 12x and/or 208 HTT. this is with bios .(insert bios version here) yes, it's the secondary processor error, and others. (x64 vista, FTL)

[justapost]

C1 errors, hmm. That's what my defect not jet rmaed 9500 also does.
Have you moved the setup (cpu+hd) from 790FX to 770 chipset without windows reinstall? If so have you had issues with monitoring apps?

Beside it's bad that the psu is defect the software seems to be usefull.

My power consumption examples where all measured with the power meter, the psu did not arrive yesterday.

[KTE]

as in the 9700 and the 9900s? blah, i kinda wasted my moneys on my BE.

No, BE is better for oc, uc and power/temps. You can easily get 2G on NB, stock volts it should be. Mine does 2214MHz stock 1.250VID fully stable.

All Phenoms were 2G IMC, they were downclocked for better stability after TLB errata was discoovered. That's whay the ones reviewers had were 2G NB.

Actually they were supposed to work with an auto detecting adjustable IMC, like the original spec had. Was suppose to auto detect and adjust the IMC clock to within 200mhz of the core speed to avoid the uneven divider problems that the K8's had, which would allow you to run your ram at full speed regardless of processor clock. It was also suppose to be able to clock down, and go into Idle mode like the cores do, with C&Q 2.0. Thats what the whole separate power planes thing was supposed to be for, all that does right now is make it easier to OC.

That was moved over to 45nm around early May period, they had too many problems with their original aims because of shortage of time. These are just a few things they skimped on at 65nm because of IMC bugs since January which led to all their launch delays. Others included a larger L3, 4 ccHT links and so on.

how do i get it to 2g?

Use BIOS 1.13B or use WPCREDIT/Bar_Edit to oc NB.

i have a question, i WAS using SB600 raid for my raid, but i was having... "lockups" whenever the system would read the cd-rom (take 30 sec to show my computer, wouldn't update to new disc when replaced) so i switched to promise. could that be part of my issue with BSOD on the P0J?

is there a way to just switch raid controllers without a reinstall? (like i guess, turn on raid on the SB600, boot, install driver, shutdown, switch sata cables..) would it then be a good idea to place cdrom and storage HDD, on the 2 promise sata ports?

AHCI mode has problems working with SATA drives on the SB600 ports. Otherwise they should work fine.

But AHCI mode gives me same perf. as IDE mode with SATA II NCQ drives, so I use IDE mode.

But 1.1b3 seems to be the best for that. But, not the most stable, why I kinda wish we could get a cross between 1.1b3 and p0j.

Yeah, same here. Wanted P0J stability, support and compatility but with 1.13B BIOS OC options and performance. P0J is much slower than 1.13B.

that's my issue, my dvdrw is a sata drive. current config is, raid 0 array on promise, 500gb and sata dvdrw on sb600, ide mode. never thought of going AHCI (call me passivly stupid) currently on bios 13b3. but with P0H, i get nvidia BSOD, P0J i get storage controller BSOD (goes too fast, but it's right in that area), and i get a x00000101 BSOD or a lockup every time i OC over 12x and/or 208 HTT. this is with bios .(insert bios version here) yes, it's the secondary processor error, and others. (x64 vista, FTL)

Vista x64 FTW!

IDK and will never know about Vista 64b esp. with SP1, because it is so problematic.

Can you ceck if you have any BSoD dumps and attach them back? I can check to see what the error is rather than guess.

C1 errors, hmm. That's what my defect not jet rmaed 9500 also does.

That's same as both of my 96BEs. That's the CPU IMC, I'm 100% on it. It faulters with initializing MEM, even 1GB 800 1.8V single channel. C1 code is when memory is being initiated. I can also tell because we have Tracer RAM which signals LEDs to show what is happening. If RAM loads up it'll keep blinking and then show a smooth pattern of flashes all across. If not, it'll either get stuck at a stage, either not turn on or switch off the LEDs. I like this RAM, it is very useful for memory troubleshooting.

Watch this. I've tried Phenom now and it won't boot in the AX78, same error. I don't have any phase or cool chemicals here to throw on it under a pot, the closest thing here is a refrigerator. So I'll have to work some insulation and leave it in there. I bet when mine goes subzero, it'll boot... I think

Have you moved the setup (cpu+hd) from 790FX to 770 chipset without windows reinstall? If so have you had issues with monitoring apps?

I tried both a new and an old install. Fact is, the new one didn't start when installing and the old one did and showed nothing wrong. Same drivers.

Beside it's bad that the psu is defect the software seems to be usefull.

Achim, it is definitely very useful. Saves a heck of a time! One little dilemma appears, as in why when you change the voltage rail output, i.e. from 5.00V to 5.15V, the actual rail voltage does not move.

My power consumption examples where all measured with the power meter, the psu did not arrive yesterday.

Ah ok. Returned mine today. It's worse for me since I pick it up from uncles place which means someone at their branch has to contact Gigabyte directly or distributor if they got it through that and return it.. then wait for a new one, whereas then I have to go and get it.

[jonspd]

KTE I meant to ask is the 8.3's worth while are for updating the chipset drivers?

[KTE]

No idea. You'd have to test it and find out.

I can't because my Phenom doesn't boot. It is supposed to be a major release fixing much though.

[jonspd]

hummmmmmm

Guess I might give it a shot if I'm off next weekend...

[dr_drache]

http://www.mediafire.com/?jp4djdusnkw

here are the 3 dumps i could find.
i don't have a debugger, so i couldn't look at them quite yet.

[Mathos]

Yeah I think when I upgrade my Ram I'm gonna go with some Crucial Ballistix, either this:
http://www.newegg.com/Product/Produc...82E16820148183
This:
http://www.newegg.com/Product/Produc...82E16820144120
or maybe 4 of these:
http://www.newegg.com/Product/Produc...82E16820148092

I'm waiting to install 8.3 until the new 3870 arrives, don't see much point in using them with the 2600xt right now.

AHCI will have close to the same performance as IDE mode, but it will have less cpu usage. Hopefully with B3 they'll at least manage to ramp up the IMC to close to core clock. Otherwise it looks like Shanghai/Deneb will have to be the savior for k10.

I'd got up to 2.4ghz NB/IMC completely stable at stock VID. Only issue I had was the Vid card thing. Works just fine with the 2600xt in. Was running 2.0ghz at 1.1v VID stable. But I do like the OC stability with P0J, 2.6ghz core with auto voltage, and it's showing the cpu VID as 1.250, before to do the same speed I was having to use 1.2625 VID. Which means I'm running .008v lower than before, on the actual voltage. And thats just the voltage the bios auto detected for it. Note I've also got C&Q turned on at the moment as well, another thing, the temp calibration seems to be a lot better now. At idle with my Nirvana at 2.6ghz I'm having just around ambient, and with full load under everest stress test it tops out at 53c after 5 mins, thats with full stress test except for HDD. I may be able to get that lower if I reseat the HSF again, didn't do as good of a job this time as I did the first time with the paste. Granted I don't know if thats 1316beta of Everest detecting more accurately or if it's improved with the bios.

[justapost]

That's same as both of my 96BEs. That's the CPU IMC, I'm 100% on it. It faulters with initializing MEM, even 1GB 800 1.8V single channel. C1 code is when memory is being initiated. I can also tell because we have Tracer RAM which signals LEDs to show what is happening. If RAM loads up it'll keep blinking and then show a smooth pattern of flashes all across. If not, it'll either get stuck at a stage, either not turn on or switch off the LEDs. I like this RAM, it is very useful for memory troubleshooting.

First i thought damn i paid ~20€ for a bunch of led's but yepp they are usefull. With my 9500 stucking at C1 the LED's where off, so i assumend a defect IMC.

Watch this. I've tried Phenom now and it won't boot in the AX78, same error. I don't have any phase or cool chemicals here to throw on it under a pot, the closest thing here is a refrigerator. So I'll have to work some insulation and leave it in there. I bet when mine goes subzero, it'll boot... I think

Hmm, had mine in the refrigerator for a while, tried to mount the cpu as fast as possible but still no luck booting.

I tried both a new and an old install. Fact is, the new one didn't start when installing and the old one did and showed nothing wrong. Same drivers.

I still suspect hw io caused my cpu to fail. Moved cpu and hd from the m3a to the sapphire mobo and had a few glitches with everest (always caused my system to freeze), whom where gone after a fresh reinstall.

Achim, it is definitely very useful. Saves a heck of a time! One little dilemma appears, as in why when you change the voltage rail output, i.e. from 5.00V to 5.15V, the actual rail voltage does not move.

Sounds like a sowftware problem, hope they fix it soon.

Ah ok. Returned mine today. It's worse for me since I pick it up from uncles place which means someone at their branch has to contact Gigabyte directly or distributor if they got it through that and return it.. then wait for a new one, whereas then I have to go and get it.

Mine is on the way,i expect it tomorrow.

[KTE]

dr_drache: sorry, there may have to be some waiting since I don't have any tools on the system I'm using now (around 25 miles away from last post destination).

AHCI will have close to the same performance as IDE mode, but it will have less cpu usage.

You're right, it should. But it didn't with mine. They both averaged the same. IDE also performed quicker on the Phenom system than AHCI on SB600, posted details of my test a 2-3 months back. It also performed quicker than IDE mode on my P35 ICH9R system.

I'd got up to 2.4ghz NB/IMC completely stable at stock VID.

What was your mem bandwidth and latencies?

Note I've also got C&Q turned on at the moment as well, another thing, the temp calibration seems to be a lot better now. At idle with my Nirvana at 2.6ghz I'm having just around ambient, and with full load under everest stress test it tops out at 53c after 5 mins, thats with full stress test except for HDD.

Move over to 1319 build, 1316 was only a special debug build for someone with ATA problems. 1319 does have Phenom improvements but not over 1314 AFAIK.

As for temps, as I've mentioned before, those aren't your temps. but Tcontrol value, which is a similar value to your temps only at around 50C, lower than your core temp below this value and higher than your core temp above this. hence it is perfectly normal for it to read negative values at low ambients and plus 90C at near maximum temperatures.

Achim: await the results. Make sure you have a spare power cable for the PSU, they don't ship a power cable for it.

BTW, anyone checked out this system mod by Bill and the team for AMD to use at tradeshows? http://www.planetamd64.com/index.php?showtopic=34129

Just two pics (plenty more there):

Phenom 2.6 9900
2x 3870 X2
CoolerMaster Copper Sphere CPU cooler
WD 400GB HDD
2GB OCZ PC2-8000 memory
GB[T] 790FX board
750W PCPC power supply

Iz3D 22" widescreen monitor
Thermaltake 7" 5.25 Bay LCD
Razer Lachesis Mouse
Razer Lycosa Keyboard
4 Black SATA II cables (right angled/straight)
Pioneer BD Reader + DVD +/-R/RW SATA
MS Windows Vista Ultimate

The system will make it's debut at the World Cyber Games (modding division) in Cologne, Germany and be featured at Computex in Taiwan.
...
Vista is what AMD wanted.. Remember, AMD will use this PC for displays at Tradeshows around the world. MS probally wouldn't like to see AMD using an older OS!
...
We missed the deadline for Cebit, but AMD is still excited about using it Computex. I'd love to see the reaction from Thermaltake at that show. Since it missed CeBIT, AMD gave us permission to play with the system in the meantime. I'd like to get guru in video conference and overclock this 9900!

Check my AMD test system wallpaper out

[dr_drache]

@KTE, no rush, no rush at all,
would going form 1800, to 2g make much of a differnce?
i know, a simple 10% OC wouldn't make a great differnce, but it always helps,
just wondering if HT speeds would make same type of differnce or be more noticable.

[Mathos]

Just default timings on that cas5 G.Skill memory I'm using.



Bandwidth in sandra was around.... 10580mb/s or something like that. One thing I noticed though, my memory bandwidth at 2700 x 2000 was higher except for the L3 cache, than at 2600 by 2400. Makes me think the performance jumps after 2.6ghz, would love to see 2.7 by 2.4 or 2.6, just to see if performance goes up or down.

At 2.7 by 2.0, L1 was at ~86000mb/s 43000 and 43000, L2 was at ~21000, 15000, and 20000 respectively. Though the regular memory and L3 numbers were lower. So thats almost a 4gb/s jump in L1 performance, and close to a 1bg/s jump on L2 for a whole 100mhz cpu speed difference.

[Morais]

Kinda OT: Guys, do you think MSI plans to make a new board with 790FX + SB700? Maybe a K9A2 Diamond

[madfaze]

hoping they will