Lets talk breaking the HTT 230 limit on k10 here

[msimax]

hey macci does aod work on 4x4 systems? my rig is down atm so i cant test

[KTE]

Could be. I'm just going off what tools/benches read/showed. They were afterall synthetic benches which showed it much faster, so it could be "untrue" of the original NB frequency. It only happened on a day or two, never happend agian. If I chose the settings again, no POST. I booted at a few similar settings.

RAM was in single channel those days. This is how the pattern went.


Compare Cinebench10:
2640/1920HT/1920NB/480/4-4-4-4-24 1T tRFC 75ns dual channel: 2214 / 8448


2640/2160HT/3120NB/480/5-5-5-5-24 2T tRFC 105ns single channel: 2235 / 8456


Booted:
HT ref. 200
HT LInk 9x 1800MHz
HT Fre. 1800MHz
CPU FID 06
CPU DID 1
CPU VID 28
CPU Fre. 2200MHz
NB FID 09
NB DID 1
NB VID 36
NB Fre. 2600MHz

Then I just AOD'd upwards and it went up along with HT/CPU/RAM IIRC. But I did the same a few times at different MHz those days.

[lukija]

Can you follow instructions given in this post.

http://www.xtremesystems.org/forums/...9&postcount=91

[KTE]

lukija:

I dd that back then. You want me to repeat it? I'm running stock 2.2GHz/1.8GHzHT/2GHzNB now, so would that matter?

[lukija]

Yes, you did. but it was at stock settings:

http://img220.imageshack.us/my.php?image=idud8.jpg

If you can repeat the same procedure but with NB at 3120MHz and CPU ad 2640 MHz. NB FID 09 and NB DID 1 should make NB run at 3120MHz (HTT=240MHz) but just to be sure, one must inspect values of registers. To be more specific, for this settings:

http://img507.imageshack.us/my.php?image=lmqy3.png

10x in advance.

[KTE]

Well as I've been mentioning, I can't even bootup at more than 2GHz NB or over 9x NB multi since 4-5 days. It'll fail. No chance.

Volts/cooling makes no difference in my case. I've had the chip at 2750MHz 1.392V and switched it off and taken out the core in my hand 8 seconds after and it was cooler than my palm, far cooler than my RAM/HD/NB. I've booted 270HT ref. max with Phenom 9500, that's where the limit lies for it (low multi) and around 262HT ref. bootup at 11x multi. The board showed no vdroop/vdrop before but now shows quite a lot. 1.562V BIOS is 1.489V Windows idle. And AOD resets the voltages after you apply them to bootup values.

[Brother Esau]

@Tony........Any word on new Bios for DFI 790FX-M2R???? Current Alpha Bios has issues 12/06/07 Alpha lots of them

[lukija]

@KTE

OK. If you somehow manage to reproduce settings in post #202, please, if you can, make WPCREDIT and CrystalCPUID shots.

[KTE]

lukija: I'll try right now.

But first of all I need some help here please.

I started the system up at stock (auto values) and it's downlocked itself. Any assistance you can offer?

[Tony]

@Tony........Any word on new Bios for DFI 790FX-M2R???? Current Alpha Bios has issues 12/06/07 Alpha lots of them

nothing yet...if i get anything i will post up

[Brother Esau]

Tony, by any chance do you have two ATI Cards you can run in cross fire?

The reason I ask is if you can valid a existing issue in bios at AUTO in PCI-E in Advanced Chipset Menu where it is perhaps not supplying enough power to the given cards or if there is some sort of IRQ Conflict with the PCI-E GFX CARD AND GB Lan controller?

[Tony]

I do not sorry

[Brother Esau]

I am getting another board this week because of some issues going on with the current one and I will see if this is duplicated with new board when I get it set up and running

[lukija]

lukija: I'll try right now.

But first of all I need some help here please.

I started the system up at stock (auto values) and it's downlocked itself. Any assistance you can offer?

Very weird:

CpuFid = 000110b = 6
CpuDid = 000b = 0

CPU Speed = 100 MHz * (6+16) / 1 = 2200 MHz

NbFid = 00110b = 6
NbDid = 0b = 0

NB Speed = 200MHz * (6+4) / 1 = 2000MHz

Obviously, target speeds are correct but current are mess.

OK. This MSR C001 0064 holds settings for P-state 0, C001 0065 for P-state 1 .... and C001 0068 for P-state 4. You might dump values for these 4 registers also. And two MSRs (machine specific registers) are interesting: C001 0070 aka COFVID Control Register and C001 0071 aka COFVID Status Register. I think there is our answer.

[KTE]

Yep, it booted up with the correct register values. So even CPU-Z picks up the speeds as they were set in registers. But something has downlocked eveything, including the RAM. Me thinks it's NB related again.

It booted fine this time so I'll wait to check values when it freaks out again.
Thanks for the help.

[KTE]

Here's something for ya lukija. I changed NB multi to 10x manually this time and it downclocked everything again.





C001 0067 and 68 are the same as 66.

[lukija]

According to your SS, CpuFid is always 6. It means 2200MHz speed. But somehow AOD reports lower clocks. To be sure, you can use CrystalCPUID and its speed measurement function. It has various ways to measure speed, try all of them and see if they are all the same. Pref F5, F6 .... F9 and see what happens. My point is to determine who is wrong since AOD displays lower clocks and CPU-Z displays higher clocks.

[KTE]

CrystalCPUID measures exactly like AOD lukija, that's why I never posted it although I was about to. I even checked it's real-time clock => exactly the same as AOD.

CPU-Z read BIOS values.
AOD, EVEREST, CrystalCPUID, Windows counters read the same values.

Synthetic benchmarks which rely on the frequency they obtain to give you a perf. score, all score very high (scaling as expected). But when you run a real-life application (like POV-Ray/WinRAR), it shows the CPU is slower than 2200MHz by quite a margin.

[lukija]

That is strange. Looks like CPU-Z uses different methods for calculating CPU clock, depending on platform. Small OT. When I was writing A64Info i had problems obtaining HTT speed. Best way is to read values from clock generator but it was difficult for me (no documentation, etc etc). Instead of that I measured CPU clock , decoded CPU multiplier (from MSR) and calculated HTT. In most of the cases this works OK. QPC method from calculating CPU speed works very well but sometimes, depending on HW, you can have problems when clock changes dynamically. Best way is to combine methods. In other words, measuring CPU speed should be considered case closed. So what causes your downclock? Two things - lowered multiplier, or lowered HTT. You may try for example to change HTT from 200MHz to 199MHz and return in to 200MHz and see what happens. Ideally would be to read values straight from clock generator, then we could pinpoint where is the source of problem.

[KTE]

If I change HTT values up/down they go in sync with whatever readings it gave before, i.e. 200MHz set BIOS but reading 180MHz Windows AOD, lowered 2MHz still gives 178MHz rather than 198MHz. This is one of the first things I tried.

If you notice, the system downclocks the memory too. This might be a ratio thing the way it looks. Something like downclocking HT ref. by 0.5x multi.

So 200MHz (11x CPU FID) would become 190MHz at 10x NB multi, and so on.

[lukija]

If I understood you correctly, HTT is underclocked? In your case it is 180MHz ? It would explain DDR2-720 speed ( mem divider is 5x).

[Lightman]

It might be BUG in BIOS and/or not yet finished C'n'Q kicking in....

My system never did that......

Hunt the BUG guys and good luck!

[KTE]

Yep lukija.

BIOS guys lay blame on Phenom (because 5000+ BE and X2's don't show these problems on this MB).

CnQ is always disabled here... thanks.

[Brother Esau]

Still waiting on a Bios in D.F.I Land

[lukija]

Yep lukija.

BIOS guys lay blame on Phenom (because 5000+ BE and X2's don't show these problems on this MB).

CnQ is always disabled here... thanks.

I wonder what is the location of clock generator? You can always check its registers. I tried to look at SB600 papers but there are only clockgens fro USB,SATA etc etc...