Gigabyte MA790FX DQ6

[keithlm]

One thing I have found with this 9850 is the IMC is nowhere as good as my 9500 was. I'm now running my ram at 800. Must be something in my bit of silicon because it's on the same bios and I know the ram is good for 1150mhz.

I disagree with you: I don't think it is the 9850 IMC. I think it is the Gigabyte bios.

This same memory worked great on the DFI board I had for a few days and I had no problems. (Up until I ran Sandra and it hosed everything. But I think that it just mangled something in the bios... who knows.) But when I say "no problems" I mean that Prime95 ran for hours with no readily apparent problems. I even ran 4xPrime95 and then kicked off the AOD stability test at the same time. I let those both run overnight at the same time with no problems.

It's either the Gigabyte BIOS or they (Gigabyte) have done something wrong in hardware.

(I'd prefer BIOS.)
===========================

ANYWAY:

This morning I did the following:

1. Reset optimized bios settings.
2. Reboot.
3. Reflashed the F3 bios. (I did this since the note on the Gigabyte website says that the F3 bios updates the "backup" bios.)
4. Optimized settings.
5. Reboot.
6. Reflashed F4.
7. Optimized settings.
8. Reboot.
9. Reset some BIOS things. (Turned off misc stuff.)

RESULTS:
1066 was very unstable. I could get windows to post, but if I started Prime95 it would reboot immediately. I lowered the memory speed to 800 and it still rebooted immediately if I ran Prime95.

I also noticed that with the F4 BIOS the CPU-VID and NB-VID both defaulted to 1.3Volts. (They defaulted to 1.2375V and 1.1375 in BIOS version F5)

But F4 was too unstable. So I reflashed F5 (optimized, save, reboot, etc.) and everything is back where it was. I can run PRIME95 and it won't immediately reboot the machine. Prime95 (blend) will fail after a few minutes... but at least it runs. (If I decrease speed to 800 Prime95 will run longer but still fail. If I put in some old DDR2-400 ram it will run for a few hours before failing.)

BTW: I think my "main problem" is the G.Skill 2x2Gb ram I have. (With the powerchip IC. They have really funky BIOS timings.) If I had a more normal RAM I think I'd be a lot more stable. (But not as stable as I'd like to be.) But even though that may be some or even most of the problem... I still suspect that the bios has some issues.

EDIT: I learned more than I wanted to know about Prime95 today. It seems that the "small FFS" test will test the cores and that the "blend" test the memory, core and communication between both. Since memtest86 runs fine on my machine for many passes... and Prime95 small ffs runs without problems for long periods of time... then that leaves the the NB. Which is what I guessed previously.)

Looking at the congregated results here with DS5... I'd say GBT has a major BIOS problem.

Apparently they don't think so:

Gigabyte's response to my question about problems with the current (DS-5 F5) BIOS. (I asked them when they'd have a fix):

There are no known issue with the 9850 with the F5 bios, please check your memory and try testing with single module

So apparently we aren't having problems... it's all our imagination.

To be honest: I had suspected I'd get a lame answer like that.

[keithlm]

NEW RESULTS::: Lowered the NB speed and got stability.

Instead of playing with the different memory speeds... and timings... et al. ad nauseum. (I am now SICK and TIRED of memory timings.)

I didn't know how to lower the Northbridge speed:

In the AOD you have a setting called NB-VID... but it won't save if you attempt to change it. (You have to change it in bios. See DaveBurts post.) But increasing the NB-VID made my system VERY unstable.)

In AOD there is also a HT Multiplier which lowers the HT Link Speed but it doesn't touch the Northbridge speed. There is no Northbridge slider. Dang it.

NOW: In the BIOS there are two settings: Memory controller (9x, 10x, 11x etc.) and also HT Link Frequency (1.8, 2.0 etc.) Apparently BOTH of these must be changed at the same time. If you decrease the Memory controller lower than the HT Link frequency the BIOS will change it back when you reboot. But if you change the "Memory Controller" to 9x (1800) and the second to 1.8... and save it... the northbridge AND the HT Link Speed will be 1800.

Apparently the BIOS "memory controller" == AOD's Northbridge and the BIOS HT Link Frequency == AOD's HT Link Speed.

I had a scary close call. I decided to raise both to 2200 just to see how unstable it would be. Hey who knows... it might have been more stable. Anyway BOINK. SPLAT. I had to remove the Phenom and boot with an X2-4800+ before I could get the Phenom to boot again. SCARY STUFF.


OKAY YEAH!

Using 1800Mhz NB/HT I can run Prime95 (Blend) without getting errors at least in the first few hours. This is running the memory at the FULL DDR2-1066 speed. (Before even if I put in old DDR2-533 memory it ran longer but eventually rebooted. Using 1066 memory it would error after a few minutes.)

In addition, because I'm impatient:: while 4xPrime95 (blend) ran for about an hour I kicked off the Everest stability test. That worked okay but it really slowed down the Prime95's. Then I ran the NVidia "Cascades" demo which mostly uses the GPU. All of these ran together. (Actually I'll kick them all off together when I go to bed. Maybe run the AOD stability test at the same time. They'll ALL run slow if you run them all together... but they should run without errors and without rebooting.)

NOTE: If I try running multiple stability tests using the default 2Ghz Northbridge/HT the system WILL reboot after a few moments.

SO I THINK I HAVE FINALLY FOUND STABILITY.

Sure I don't like the lower 1.8Ghz NB/HT... but at least this system is now running stable with this CPU, this MB and this memory. NOW I have a "zero" point and can "play." I know there is no "unknown" something that makes the system completely unstable.

So why do I need a lower NB?

Possible Reasons:

1. A Bad CPU
2. The BIOS needs updates
3. G.Skill F2-8500CL5D-4GBPK memory doesn't play well with AMD



Oh... pretty... and BEATING the crud out of the system. (This is a dual monitor... doing a screenprint will show both screens as if it was one screen. OH NOES... only 11.2 FPS in the game... THE TRAVESTY!)

Oh wait... I need some HD Tach in there to test the RAID hardware card. :-)

[Daveburt714]

NOW: In the BIOS there are two settings: Memory controller (9x, 10x, 11x etc.) and also HT Link Frequency (1.8, 2.0 etc.) Apparently BOTH of these must be changed at the same time. If you decrease the Memory controller lower than the HT Link frequency the BIOS will change it back when you reboot. But if you change the "Memory Controller" to 9x (1800) and the second to 1.8... and save it... the northbridge AND the HT Link Speed will be 1800.

Yep... 1st Rule of Phenom. HT Link Spd can't be higher than NB Spd, it's actually a good thing that Bios won't let you do that..

I really hope that stays stable for ya! Running the HT Link at 1800 is not that big a deal, but lowering you NB to 1800 will have a more detrimental effect... If all goes well and you want to get a little more out of it try raising your NB to 10x while leaving the HT@1800.

[keithlm]

Yep... 1st Rule of Phenom. HT Link Spd can't be higher than NB Spd, it's actually a good thing that Bios won't let you do that..

I really hope that stays stable for ya! Running the HT Link at 1800 is not that big a deal, but lowering you NB to 1800 will have a more detrimental effect... If all goes well and you want to get a little more out of it try raising your NB to 10x while leaving the HT@1800.

Can't do that.... it's unstable with the NB at 10X.

But I agree that all of this is probably the BIOS and not the memory. They made some pretty drastic changes between F4 and F5 for the NB voltages... so I'm sure they have more tweaks in that area.

You know: actually I might lower both NB and HT to even further to 1600 so I can play with increasing the HT multipler (What was FSB) to as high as I can. At 230Mhz the CPU would be 2875 and the NB would be 1840Mhz. (I think the memory can handle handle that.)

That would be nice... because if I can do that with just the FSB then when Gigabyte actually gets around to fixing the BIOS... I might be able to up the CPU multiplier and get 3.2Ghz. (3.4Ghz???) Of course first they have to admit that there IS a problem. (Which right now they are denying... RIGHT.)

BTW: as you can see in the picture I posted above... while slamming the system with 3 programs that can EACH use 100% of all 4 cores... the CPU temp only went up to 46C-48C. The Zalman 9700 ROCKS.