Dothan ram bandwide boost possible

[railer]

Well as title stay i was playing with my system and notice one strange thing that let me find one bug and exploited to get a boost in memory score for my 730. Thanks to Hipro adapter mod you could even get a better result, booting at high fsb could matter. So far i noticed 1sek PI increase and around 1K in Aqua cpu score. Others results could be way higher since i was booting at not so High FSB. For this to work some thing need to be done first.

You will need to boot with highest multy possible for your cpu. The FSB jumpers on adapter have to be set to 400FSB. Now forget about booting @200fsb. Boot at highest fsb possible for the condition. In my case lets say its 213x12 that max my 730 will do on air. So after I boot to windows i usualy change to 10x multy to maximize my ram bandwide and clock cpu to same speed > lets say 10x256 is faster then 213x12.

Here is the trick. I dont know why but when jumpers set to 400sb when you change from higher multy to lower you get increase in rambandwide. See my screenshots and isn't amazing that at 6x multy my 730 does 7.5K. Now onother part of the trick. The higher the FSB you have when swithicing to lower multy them more rambandwide you get. For instance if i boot 12x200 "useles for P4C800 since it has PAT anyway" and switch to 10x multy i get less score them swithching to same 10x from 2.550G. Yor comination shoul be best posible FSB + highest Multy. This way i get way better performance thern booting my sys @200fsb and working my way up. To put it simple the higher ther FSB speed i boot on when switching down multy the more Rambandwide i get


Now i have tried this with my 2 Asus boards and it worked same for P4c and P4p. I could have bad cpu but i highly dobt so. Can anyone test this on their sys and let me know if it worked for them? It could also be that i have completly buged setup. The funny thing is that when i increase the FSB after switching to lower multy the sandra and ctiaw does not register increase. I score same mem score as when just switching the multy down. But Aquamark and PI defintly gets a boost this way. Just wanted this to be public knowlege if indeed it works for others
http://www.xtremesystems.org/forums/attachment.php?attachmentid=41483&stc=1&d=1135249353

[kiwi]

7K at 210 ram with crappy timings is not possible
Is sandra reporting the same?

[railer]

proof that lower multy with 400fsb jumpers score better.Notice the FSB is the same yet on lowe multy better MEM score.
http://www.xtremesystems.org/forums/attachment.php?attachmentid=41481&stc=1&d=1135249198
http://www.xtremesystems.org/forums/attachment.php?attachmentid=41482&stc=1&d=1135249198

[Daytime Dreamer]

Dude, I'm willing to try this out on my system. However I have a question : Do you have 4X512 modules on your system? I have 4X512 old KHX Bh5 and booting at anything above 200fsb disables PAT for sure on my P4C800-e DLX. Cpu-Z may say "enabled" at the Performance mode at the memory tab but its wrong:slapass: If you run 865/875 tweaker it will say that PAT is disabled. Enabling it in tweaker and returning to CPU-Z will then say disabled (which is wrong again!)

Try running some sandra benches and you'll see you lose PAT above 200 fsb with all 4 banks filled. I boot at 200 FSB and get up to 212X12 = 2550. I could post some Sandra benches if you are interested

[railer]

7K at 210 ram with crappy timings is not possible
Is sandra reporting the same?
holy crap why sandra reports my bus 4x426 =1700?
Someone have to test this. There is a bug when 533 fsb cpu is booted with 400fsb jumpers swithching to lower multy does crazy things for me
http://www.xtremesystems.org/forums/attachment.php?attachmentid=41484&stc=1&d=1135249935

And DAydreamer i have 2x1024. Now with 4x 512 i havent test it. This could be benficail if you got high scaling ram. In your situation lower multy usefull for you? I dont know howe high you can scale 4x bh5? My pat is not disabled no matter what speed i use. CTIW never wrong on that see in uper screenshot
It just seams by sysoft benches i done that when i switch to lower multy my bus is locked. Even if increase fsb after Sandra does not report it.

Better timings. CaN it be that this cpu has locked fsb? It seam wheni change multy fsb is applied and then never change
http://www.xtremesystems.org/forums/attachment.php?attachmentid=41485&stc=1&d=1135252682

[Daytime Dreamer]

Here are my benches at 212 X 12 = 2552

Notice that CPUID and Tweaker report the exact opposite as far as PAT is concerned. Tweaker is right and NOT CPUID (i figured this out by running some benches)

I boot at 200 FSb and then use Clockgen to increase the speed by upping the FSB.

Screens

http://img459.imageshack.us/img459/834/cpuidtweaker5gz.jpg (http://imageshack.us)

http://img459.imageshack.us/img459/1521/cpuidtweaker22kv.jpg (http://imageshack.us)

[Daytime Dreamer]

And the Sandra memory bench

http://img459.imageshack.us/img459/4664/sandra212x129xi.jpg (http://imageshack.us)

[saaya]

di rather say youve found a bug with the benchmarks and dont really get such a high boost... sounds weird...

[railer]

di rather say youve found a bug with the benchmarks and dont really get such a high boost... sounds weird...
So you teling me that all 3 benchmarks programs are wrong? For what i know CTIaw is never wrong. I tested with 2 dif MB and 2 dif operating system win xp and win 2000 server. I have come to conclution that it's not program related at all. It's just a buggy cpu that i have. It seams that whenever i apply Multy change the FSB get's manualy set. Then even if i increase frequncy .i.e core speed the FSB stays the same. It is very weird but its the way it works. I have tested this on 10x multy. After i swithch from 12x to 10x i get 5.6k in rambandwide. Increasing the cpu speed to what cpu can take max does nothing on rambandwide at all. Now Sisoft does not report FSB changed , it stays locked , only cpu core speed is increased. So EIST works kind of very diferent with this cpu.

[saaya]

sorry, but sisoft sandra is a pos :D
please write it on a cd, uninstall it and then cut the cd into tiny pieces and burn it :D

it doesnt read the fsb correctly, sometimes it doesnt read the fsb at all but takes a guess what it might be :rolleyes:

and do you remember the cpu clockspeed bug on the p34 cpus reaching 4.3ghz and above? the system timer would freak out and the system would report very odd speeds and act incredibly slow.
this could be something similar.

or maybe this has to do with the multiplier bug on dothan cpus hmmm

[kiwi]

Is that a fresh Win install? If not, do it and rerun tests. :)

[saaya]

is this a 730es by any chance? it isnt right? hmmm

[railer]

sorry, but sisoft sandra is a pos :D
please write it on a cd, uninstall it and then cut the cd into tiny pieces and burn it :D


hehe :slapass: Agree
Realy i have done all 2 dif Mb ,2 dif ct479, probably 3 or so fresh installs including windows 2000 server! This frkn cpu just works the same in all configuration. Once i use EIST to lower multy it set FSB and never changes it once i up the clock. Same rambandwide test after droping multy and increasing clock. To bad i sold my high 290mhz ram and got 2x1024kit which can only do 266 on air. The funnly thing is that Aquamark does get a higher cpu score if i do whati described above.

For now i havent put this setup under -25c cold air in my chillbox. I get a problem with Asus RAID controler. When temps below -20 it refuses to work and my windows dont load. I tried to disable it in bios but then i cant install windows since it does not find any drives. Anyone knows a way around this? I was thinking of flashing my p4c and bootblock with p4p bios since p4p dont have Raid. But i'm having issues with AWDFLASh it cant recodnize the p4p bios somehow..........
Eh this is the weirdest setup i had in loong time.

Check the screens. After changing from 12x to 10x from 12x213. Yet cpuz report the bus change the rambandwide didnt change. At least Sisoft reports it same.
http://www.xtremesystems.org/forums/attachment.php?attachmentid=41534&d=1135408562
and same rambandwide @ overclocked



http://www.xtremesystems.org/forums/attachment.php?attachmentid=41535&d=1135408562