Trouble in A64 (Clawhammer only) land w/ a mobo--- need help

[BC]

Guys,
Here's a problem I can't figure out how to solve in Win/XP (32).

Corsair memory (512mb sticks) CMX-3200XL (2-2-2-5).... reports
via CPU-Z ver 1.23 that

a) DRAM speed = CPU/20 (aka... FSB/2) yet bios is set FSB=DRAM
in 1:1

b) utility program reports all available Corsair info as:

2-2-2-5-11-16 (the 11-16 is pushing it to 1/2 speed) hence,

*GASPING*... My DDR speed is being reported as pushing 113 Mhz ... NOT 226 as it should be. (as if the 1:1 lock was ignore by bios)

c) the board 'single channel' mode... and it is....

d) does not report 'interleaved', yet it is..... and split-integer/split-real benchmarks prove it. (near 8GB/sec which is close enough to where it belongs, but only 1GB/sec in non-split interger or real).... so interleaving is active... just not reported, so that's ok .

e) XP64 SCREAMS and performance is where it belongs.... but I can't run it because Markell does not yet make a 64 bit network driver for the Yukon.... :(

I have bios 1004.003 (up from older 1003.006(?) ).... Is this a BIOS foul up in misreading the Corsair and mis-setting the MB clocks?

I found a similar XP3200 which has nearly double the Int performance as me... meaning his mobo is working right.

I tried to set the DDR Delay Timer (the 16) down to something reasonable.... the highest value my mobo goes is 8... (which is the correct value).... but would rather have *ALL* the Corsair params before I
go doing this *AND* overclocking....

I don't have lab tools at home, nor do I have a mobo w/ jumpers...

The mobo is the ASUS K8V SE Deluxe ver 2.0....

I think the bios is messing up based on what it's reading from the Corsair,
or this Corsair just belongs in a DC board.... period...


Anyone have any answers for A) keeping the overclock and B) setting ALL the memory parameters (there are 9) and getting DDR speed *AT LEAST* back up at 200? This is MOST embarassing.... I could solve this in a blink with the logic analyzer and see who's messed up, but I have NO LA, schematics or source code and can't get either, so I have to do this one blind.

I suspect this might be related to Keith's 207 FSB issue, but am not certain.

But at this point, I'm WAY too close to the problem and am not seeing the obvious.

As usual, ASUS is totally non-responsive....
and there's no info at work about this mobo.

Suggestions?

Thanks in advance,
BC

[jlangner]

Have you posted for the RAM Guy? I had a similar problem when I just upgraded, albeit Newcastle and 939 but still same problem. What I found was that I had to have memory aligned a certain way since I had 2 256MB and 2 512MB sticks, I also had to enable 2T instead of 1T but that increased my Sandra mems from 2500 to 4800. I found this table for my Asus A8V MB.

http://www.x86-secret.com/pics/cpu/s939/banks.png

Also read this:

http://forums.amd.com/index.php?showtopic=18521&st=0&#entry149462


http://www.houseofhelp.com/forums/showthread.php?t=27819&highlight=jlangner

[jlangner]

Also found this from Corsair Help Forums:
http://www.houseofhelp.com/forums/showthread.php?s=&threadid=8048

"Testing parameters for XMS3200XL:

CMX512-3200XL Data Sheet
Motherboard make and model: Asus P4C800/-E and /or Asus A7N8X
DRAM Clock: 200 MHz
SDRAM CAS Latency: 2T
SDRAM RAS to CAS Delay (tRCD): 2T
SDRAM Row Precharge (tRP): 2T
SDRAM Active to Precharge Delay (tRAS): 5T
SDRAM Bank Interleave: 4 Bank
All other settings are motherboard defaults."

[BC]

Ok, so if I read this right... since I have a 3-slot machine.... I cannot run DDR 400... I MUST run in DDR-200 mode???? That's how I take it since I have 3x DS DDR-400 sticks... there is no 4th slot (clue #1). AND the bios is overriding me? :(

Notice is says bank interleave is 4??? I can't have multiples of 4 with 3 sticks, can I? (clue #2)

Thanks for the params from the forum, I found them as well, but how about TWR and the rest of the parameters???? and Clock delay?? that sucker is set for 16 per JDEC... it should be 8 and the board only lets me go up to 8. (clue #3)


Also. Since the 2T 'disable/auto' option disappeared with this rev of the bios... I'm going to back rev..... I smell a rat... (clue #4)

Plans are this (unless you suggest otherwise):

1.. Pull a dimm for now ( I have 1 black and 2 blue slots all full of DS.. I will experiement)

2. if no solution... I will back rev the bios and get my '2T' option control back. I think that's a part of the key and play with that.

3. then start pulling ram AGAIN... :(

4.. if still no luck... start slowing it down to 2.5 or even 3... e.g 2-3-3-8
and see if that works... I remember SOMEWHERE that I had the
option of 16 clock delay.. now only 8 is available... this hints the MB
is TRYING to go to the proper speed but can't for some overriding reason


AND, if all else fails..... I'll change my construction plans to be this:

a) find a processor (someone recommend one that works on this 32bit MB) for this ASUS board (and ram?)

b) I will go get a PROPER DC mb for the 3200.. and SS memory.

THEN, I will get back to building my next machine...

So I will end up with 3 instead of 2... but at least they will all perform...

This DDR-200 crap (i don't care what the bios says) stinks.


Thanks again.... Please suggest as appropriate.
BC

PS: WHERE'S MY LINUX and WHERE'S MY XP64?? it works fine there. :(

[jlangner]

Shouldn't you set the ratio to 2:1. That is the way mine is set.

[BC]

Jeff & all,

Some interesting info has come my way... something I totally overlooked.

First, The initial Clawhammers are 128bit internal wide databus and strapping it to a 32 was a problem. I don't get a 64 bit burst, I get a 128 bit burst written each time to RAM.

Second. I have confirmed the ASUS BIOS 1004.003 for the K8V SE DELUXE is flawed. This has been confirmed by resetting all parameters in 1004 to default (auto) and attempting to boot... It would not. Reloading (via AFUDOS.. which is the only way to backrev) to the revious version did not give me the 1T / 2T control back yet, BUT has allowed me to boot and run with defaults.

Third, 'Turbo' mode on this thing ONLY applies to memory timing. Corsair XL is already aggressive enough at 2-2-2-5, and the ASUS cannot deal with the double sided 2-2-2-5-7-16 config (notice 7, not 8 now?)....

Fourth, the new A64's more correctly address dual channel vs single channel modes and are (per more indepth study from Jeff's article reference... thank you) are able to run in 'virtual DC' mode. The Claw is not. It must have DC mode if it is going to run with DS DDR-400 ram.

Fifth, The Windows XP 64 MB drivers for 754 pin machines do set the proper parameters for the counters correctly and override BIOS... hence the speed up which matches the 'virtual DC performance in split-integer/split-real' modes. (aka.. this thing will burst up to 8GB/sec, but not sustain it without being in 64 bit memory/OS mode). This machine will be a 'butt kicker' when I get a 64bit OS on it... like Susie 9.1 or XP64 w/ a net driver or different network card in there.

So, I would summarize status as saying I am not up to full speed yet, but am stable at 226 FSB , Standard performance on 1003.006 BIOS. I can burst at 6-8 GB/sec, but not sustain it for more than 64 bytes (one cache row) at a time.

The true Claw is not sufficiently smart enough to handle 32 bit and single channel bus combinations with Double Sided SDRAM. Either 64 or DC or SS is fine. All the newer cores are fine across the board.

The newer cores (NewCastle, etc) can handle the 128/64 bit split and memory management with auto-fragging better / 'virtual DC' mode correctly.

Therefore, I am going to do the following.....

a) Move this Claw Rev 1 over to a true non-ASUS, DC board and grab one more stick of 3200XL and have a 2G machine..... and run Linux until XP64 is out or a driver available.

OR

b) Build the new 939 pin on a true DC like you've done Jeff and spin it around... (Just watch that DAMN BIOS... they are rushing and making mistakes).. and buy single sided ram.


It's a shame... I have 1/2 of both worlds here and they don't work together. Had I picked up a 940 pin Opteron 100 instead, none of this would have happened because it's already DC.

Thanks for all the help... I owe you a big thanks and also to Alex who reminded me of some of the issues from back in 2001-2002 time frame.


This baby is back on the air at 2/3 speed for now and will be fixed as soon as Fry's gets restocked with the new product line....

... An alternative would be ... placing a 754 pin anything else w/ 32/64 bit cache design on it.. that would also work... but off topic.

Interesting day, huh??


BC

[BC]

We found it.... and you're going to laugh...

I've always questioned the spot on the Mobo for a dip switch that
was never installed...

After removing the Mobo, installing the switch, setting it to DDR-400
(where things belong)... reinstaling everythiing (including burning a floppy drive because I was off by 1 pin on the power.. thank you-know-who for spares).... I am at DDR-400... with virtual DC (the multipliers are set to what 64bit mode sets them to) and the Corsair is hauling butt!!!!!

I did crash the machine once in the process and am currently doing a defrag after the system restore (one registry file got corrupted in the floppy burn).

My only loss was Predictor non-reported jobs from 6pm -> 11pm... (not much... a whole 5 jobs)...

There is no other damage and Memory performance at 2002 Mhz right where it belongs ... I'm bank interleaving and CPUZ now says I am CPU/10 for DDR speed (which *IS* FSB speed)...

when the defrag finishes, I checkpoint, shutdown, reboot, shutdown again, (make sure it's solid) then start to OC and get back to normal...

So gang... lesson learned... what we put on the MB in the lab *should* be there... LOL..

I sure learned a lot about the 'commercial production' world and want to thank everyone.... Especially Jeff.... for the pointers to the commercial board info...

I also thank you all for tolerating this step-by-step saga, and hope it can help anyone in the future who is dealing with weird situations.


With Deepest and Greatest Regards and Thanks to all,
BC.

[WienerDog]

i just wish i understood what the heck you guy's were talking about

[jlangner]

Thanks - Just helping the "Team" lol.