Phenom 9500 w/ MSI K9A2 Platinum

[gareth170]

gareth170: Using the exact same method you used to flash 1.2, just with the correct v1.1 BIOS file.

i used msi live update 3 which it made a automatic flashing dos disk with the 1.2 on it which i liked. but ive deleted the 1.1 setup files from msi live update 3 folder. where can i download it from again? and would it downgrade this way?

[jonspd]

no it will just go back to stock bios check MSI's website for offical bios.

[justapost]

With this board the options are either 1 or 2.
I haven't seen that option on this board.

You are talking about your bios options.
Check MSR C0010070 with CrystalCPU.
Bit 0:5 is CPUFID
Bit 6:8 is CPUDID

CPUDID should be zero if you set it to 1 in the bios.

About MaxAsyncLatency.

As i played with different nb multis the value varied between 48 and 57ns here.

Here are two Screenshots with differen MaxAsyncLatency settings.



Difference in Everest Latency is ~6ns.

The equation used to calculate this value is in the K10 guide. I'll try to find it.

[KTE]

Can I get a short list of the options difference you see with POD compared to 113 when running a black edition of the Phenom? did you take not of this when you where playing with the 2 different bios. Is the POD really a better performer then the 113? more buggy?

P0D had the errata patch and nothing more different for me. So I reinstalled 113. I'm running P0E now and it has different options than the last BIOSes.

i used msi live update 3 which it made a automatic flashing dos disk with the 1.2 on it which i liked. but ive deleted the 1.1 setup files from msi live update 3 folder. where can i download it from again? and would it downgrade this way?

On their website: http://global.msi.com.tw/index.php?f...2&type=utility

[jonspd]

P0D had the errata patch and nothing more different for me. So I reinstalled 113. I'm running P0E now and it has different options than the last BIOSes.

P0E has more options then the last bios 1.2?

[gareth170]

On their website: http://global.msi.com.tw/index.php?f...2&type=utility

noo i mean i have live update 3 already install. and when u download something from it , it saves it in a temp setup folder but i deleted the 1.1 bios setup files which automaticly flashs the bios in dos.. i can't redownload it from live update 3 because its seeing i got 1.2..

[jonspd]

the flashing dos disk it made take the 1.1 file from msi and replace the 1.2 on the disk if you said it made one with 1.2 on it then restart and type the flashprograms name in when in dos

[KTE]

You are talking about your bios options.

Yep.
I'm aware of the equations mentioned in the guide as I've mentioned before because we went over them in the "Breaking K10 230MH HT limit" thread. The equation I gave is a simplified version of the official one and specific to this boards BIOS programming. You only need BIOS HEX values to give you the speeds and so its easy for anyone to use and doesn't require any decoding.

About MaxAsyncLatency.

As i played with different nb multis the value varied between 48 and 57ns here.

Here are two Screenshots with differen MaxAsyncLatency settings.



Difference in Everest Latency is ~6ns.

Ah ok. I didn't use Memset to change timings with K10 so far because it was giving me problems. Tried the latest one now and it works OK, and the option for MaxAsyncLat appears. And yep, it produces big latency gains and even bandwidth gains, well spotted and thanks.

P0E has more options then the last bios 1.2?

For Phenom, yes. But it has many problems not present in 1.2 aswell.

noo i mean i have live update 3 already install. and when u download something from it , it saves it in a temp setup folder but i deleted the 1.1 bios setup files which automaticly flashs the bios in dos.. i can't redownload it from live update 3 because its seeing i got 1.2..

I don't use Liveupdate 3 TBH so I'm not sure what it allows and not. My IE 7 is also crashing so I can't try the Liveupdate (I hardly use IE). The way I flash BIOSes is the usually done way, using a floppy or USB drive.

*You need a DOS floppy disk to boot with (download the files HERE and unzip onto an XP formatted floppy disk).
*Set system to boot from Drive A: with the floppy disk inside.
*And then after its loaded into the DOS environment, you need to insert another floppy disk which contains the BIOS flash utility (EXE file) and the BIOS itself.
*Then you need to type at the command prompt: A:>flashutility biosfilename.ver and press Enter. For example.
--With Flash Utility Afud412.exe and BIOS file A7376AMS.113, you type:

afud412 a7376ams.113

*Press Enter.
*Don't do anything and wait for it to see if everything completes (percentages) and the flash succeeds. Then restart and load BIOS defaults.
*Done.

[gareth170]

the flashing dos disk it made take the 1.1 file from msi and replace the 1.2 on the disk if you said it made one with 1.2 on it then restart and type the flashprograms name in when in dos

u can't type anything because it flashs all automatic..

[jonspd]

Sry about that I'm used to awardflash.exe...

I guess kte summed it up tho.


Thanks for the info on the POE also want be trying that one

[KTE]

Thanks for the info on the POE also want be trying that one

It has too many problems and also the errata patch applied. It also only allows DDR2-1066 3-5-5-15 as the lowest RAM values.

[justapost]

Yep.
I'm aware of the equations mentioned in the guide as I've mentioned before because we went over them in the "Breaking K10 230MH HT limit" thread. The equation I gave is a simplified version of the official one and specific to this boards BIOS programming. You only need BIOS HEX values to give you the speeds and so its easy for anyone to use and doesn't require any decoding.

Must have missed that, so for the record the bios labels 2^CPUDID as CPUDID.

Ah ok. I didn't use Memset to change timings with K10 so far because it was giving me problems. Tried the latest one now and it works OK, and the option for MaxAsyncLat appears. And yep, it produces big latency gains and even bandwidth gains, well spotted and thanks.

I mentioned it because the max async latency seems to change in a non linear way with different ref HT's and nb multis and that can causes non linear results in benchmarks.

[KTE]

DID 1 stands for 0 and DID 2 stands for 1. I'll check on the MaxAsyncLat soon. So far I've tried keeping it at 44 at all multis if HT ref/RAM is kept constant.

[justapost]

I'll check on the MaxAsyncLat soon. So far I've tried keeping it at 44 at all multis if HT ref/RAM is kept constant.

Found the place in the K10 dev doc. MaxAsyncLatency is labeld MaxRdLatency in the doc (verified it with wprcedit). Chapter 2.8.7.8.4 describs how to calculate the value used during ddr2 training and a method to optimize the value after dram training. In case you are interested and don't already know.

The value varies up to 2ns between reboots here that makes ~100 MB/s difference in everst mem read.
44 seems to be the max i can use here. I gain ~300 MB/s in mem read compared to the 50ns the bios detects.

Thank you for the mem-tuning link, gona try that method soon.

[Daveburt714]

I'm not sure what the last 4 characters of the stepping code means. I assume it has something to do with the IMC?

But the "MPMW" seems to be weak for OC'ing. I've been reading these threads for a few weeks now and was really confused by some of the numbers you guys were getting, when I couldn't even get close regardless of how many V's I gave the chip... (I have a 9600, 0745MPMW)
I kept thinking I was losing my touch, or missing something....
At least I don't feel quite so bad now.

I've had the chip for 2 weeks and this is the best I've been able to achieve while retaining Stability. This is a shot with SMP folding for well over 72Hrs...



If you folks should find a magic bullet for MPWM, I'd sure like to know!
If not, this may at least give some of you a rough point for a stable OC...

EDIT: Oh BTW, this is with 9x NB (1962Mhz) and 8x HTT (1744Mhz)... HTT may run at 9x but I haven't tried upping it yet, I was just looking for something solid...

[KeZzZu]

at last with my wc setup im able to run 3dmark at 2.7ghz though, i need 1.44vcore...
card's @ 850 mhz gpu or 800 mhz gpu mems are stock card one 1177mhz and card 2 1126mhz... enjoy :P
gonna run again tomorrow becose flashed clock fix biosses in.

phenom 9500
msi k9a2 plat
ballistix 8500 2x1gb
2x3870 powercolor zerotherms... which are not working even at stock speeds (mem)


3dmark 06 14.4k
http://xs123.xs.to/xs123/08026/3dmark06349.jpg

[KTE]

Found the place in the K10 dev doc. MaxAsyncLatency is labeld MaxRdLatency in the doc (verified it with wprcedit). Chapter 2.8.7.8.4 describs how to calculate the value used during ddr2 training and a method to optimize the value after dram training. In case you are interested and don't already know.

The value varies up to 2ns between reboots here that makes ~100 MB/s difference in everst mem read.
44 seems to be the max i can use here. I gain ~300 MB/s in mem read compared to the 50ns the bios detects.

Yeah, it calculates a total aggregate based on latencies and NB clocks so indeed it would definitely make a large difference to the DCT performance. I've tried 40ns vs 50ns at 1066 5-5-5-15 2T, good gains.

As for NB DID, my BIOS gives the following options:
Bits - Divisor
000b Divide-by 1
001b Divide-by 2
010b Divide-by 4
011b Divide-by 8
100b Divide-by 16

Anything chosen >1 doesn't POST (clear CMOS needed).

I'm not sure what the last 4 characters of the stepping code means. I assume it has something to do with the IMC?

IDK if what applies to K8 applies to K10h number decoding since they are totally different. One thing does apply, the year/week and the stepping '0747' and 'CAAWB' for instance.

But so far one other member says "MPMW is crap" based on 2 results. That's not right to do, it's too early to say this or if the latter letters make any difference at all.
-Firstly you'll need to have a database of similar week different last 4 letters to compare them.
-Secondly you need to have records of 'MPMW' and some other of any one particular week.
-Thirdly you need other week batches.
-Fourthly you need to match and variate steppings.
-Fifthly you need many samples of each chip to get a close to accurate rough idea.

So far we don't have these. 9600 BE could in itself be just a bad clocker. How do we know? Based on retail user results so far, it looks like a bad clocker, worse than locked Phenoms. Especially on HT ref. Jack looks to have a similar week/step to me, so we'll wait to compare chip potential.

I mean, how many users have you seen get more oc out of 9600BE than the standard edition?
How many retail users have you seen 2.55GHz stable on 9600BE?

Many have this with 9500/9600 though.

When I see ES/PR junk like this it really ticks me off because they're try to get the feeling across that one one random sample they received reached 2.8GHz stable easy and yet that's totally untrue for retail so far. I'm sorry but not talking to any AMD reps for which step/week to buy and just going out and picking any one sample of the shelf is what we do and what applies as user centric results. I mean even Fudzilla reported on 18th November something far more accurate than most I've seen around: No plus 2.3GHz parts until B3. The >2.4GHz B3 Q1 part is where what I've heard doesn't agree but the rest has been true for a while.

Also, I'm not sure if I've mentioned it but you'll need to know K10 (AM2+, and AM3 actually) are limited to providing a maximum 110A to the cores and 20A to the northbridge. Not very healthy for oc if you start at 2.2GHz at 1.232V 95W TDP.

But the "MPMW" seems to be weak for OC'ing. I've been reading these threads for a few weeks now and was really confused by some of the numbers you guys were getting, when I couldn't even get close regardless of how many V's I gave the chip... (I have a 9600, 0745MPMW)
I kept thinking I was losing my touch, or missing something....
At least I don't feel quite so bad now.

If you folks should find a magic bullet for MPWM, I'd sure like to know!
If not, this may at least give some of you a rough point for a stable OC...

EDIT: Oh BTW, this is with 9x NB (1962Mhz) and 8x HTT (1744Mhz)... HTT may run at 9x but I haven't tried upping it yet, I was just looking for something solid...

Generally 9500/9600 are clocking far better than 9600BE so far. My 9500 did 265HT ref but this 9600BE struggles over 210. Though I haven't yet underclocked the NB/HT to test max (BIOS problem), I have enough to show me the basics when 1.504V 1.48VID 220x11 fails bootup and destroys the Windows registry.



KeZzZu: Its better to either post thumbs when its not something everyone passing-by will need if they miss it, or just to crop only the info you need to show in the pics (the tools/values).
9500s are going very good so far... in comparison to the rest.

[KeZzZu]

KTE : fixed now. gonna rerun today with pencil mods. 15k barrier should be easy. also 3dmark06 wasn't up to date. patching it right now.

[KTE]

Thanks.

What volts did you need to bench 2.7G 3D on air?
And what was the water and ambient temp?

[KeZzZu]

Thanks.

What volts did you need to bench 2.7G 3D on air?
And what was the water and ambient temp?

that run was with watercooling here specs:
500l/h pump
black ice stealth 240 with only one fan.
thermalright xwb-01 cpu block
ambient air temp +20 c'
water temp unknown

Cpu vid 1.25v
vcore 1.4x (just to be sure it can handle heavy loads.)
cpu idles +24 - +28 c' load with superpi +32 c' and all cores load unknown
havent checked it yet, run was made at 6 am and after that i went to bed... so that's why im gonna take some more ss and info about systems ^^ thats's all info what i remember about from run

[KTE]

So you couldn't bench 2.7G 2k6 on air but could do it on water?

[KeZzZu]

So you couldn't bench 2.7G 2k6 on air but could do it on water?

yup, because boxed cooler couldn't handle heat at 2.7ghz also i have ace in my hands... using normal thermal paste between cpu and block... i have some liquid metal pads here should be few degrees lower temps then :P

[KTE]

That means there's a bad current leakage problem at ~1.4V.

For a 2.2GHz 1.23V 95W TDP chip, the power needed to be dissipated at 2.7GHz 1.44V is 160W providing core current supply is kept constant. So no wonder you need liquid.

[KTE]

K10 MCT seems quite limited which outlines another problem with having an IMC that's 1066 constrained. If you didn't already know, my RAM does 675 5-5-5-15 PL5 with 2.3V (benched) and 600 4-4-4-4 PL5 at 2.2V (32M) on P35 easy, but on Phenom/790FX, they fail and reboot at 562 5-5-5-15 all timings high and relaxed at 2.2V or 2.3V. Heck even my bad ProMOS IC's (Corsair PC6400) did 645 5-5-5-18 at 2.35V! So this is quite damn low for oc'ing. And nevermind "extreme", 2.2/2.3V is more like stock.

[KTE]

I'm linking two very good files for Phenom overclocking to the first post in this thread here

/One is the new improved Prime95 25.6 for torture testing, power measurements and for testing which core is stronger (it is still beta, but so are most good things )
/The other is K10Calc by lukija (A64Info developer). It is very simple yet allows you to input simple BIOS OC values to show you the CPU/HT/NB speed in different combinations.

Prime95 25.6 looks like this (my recent 2.3GHz CPU, 2GHz NB, 2GHz HT, 1066 RAM 35 minute stress test) ->



And K10Calc looks like this (values placed in there correspond to Phenom 9500 stock settings) ->