32m SPi Low Clock Challenge

[NewBeetle]

Really nice NB-san
I have seen your thread about AData and i was praying for you to don't try LCC 32M But now, i'm defeated...I'll try to go under 3894 this afternoon

Thank you, Benji Tshi-san.
I am also praying for you not to break my LCC 32M results.
Now I have changed CPU ratio & Ram setting.
And I have got a very little improvement.

8.xx Category (update)

M/B: ASUS P6T Deluxe (BIOS:0703)
M/B Bed: V-TEC ARC Bed Rev.3.3
CPU: Core i7 Extreme 965 (3836A767)
CPU cooling: Alphacool NexXxoS XP Silver Socket 775
Chipset Cooling: stock + FAN
Mem: A-DATA DDR3-2133X 1GBx3
VGA: MSI NX7900GTO 512MB PCI-EX
HDD: Western Digital WD360ADFD (AHCI)
PSU: SilverStone OP1200
OS: Windows XP SP3

[Benji Tshi]

I'm under U12P, i can't fight with you...I'll try with my SS later, but actually, with 3x1 Go D9GTR, i can do only 1000 8-7-6.

Nice run anyway, but trfc 72. Elpida don't want 60 ? Or perhaps it's the IMC ?

[cheapseats]

32M initial value range:

Low-range = ~ 6.5s (9.xx), and ~ 7.0s (10.xx)
High-range = ~ 7.0s (9.xx), and ~ 7.5s (10.xx)

i found the 'switch' between High or Low for 32M initial values: the 16K time recorded in the pi_rec.dat2 file.

for Low-range 32M initials, a sub-0.109s 16K time must be present in pi_rec.dat2 before the Spi exe is started (from same folder).

--------------------------


to activate Low-range 32M initials:

(1) delete pi_rec.dat2 from an existing SPi folder, or unzip new Spi folder (running 32M at this stage will give High initials)

(2) using the local Spi exe, run a 16K time below 0.109s (ie 0.094s or less). The created pi_rec.dat2 will contain the Low initial 'instruction'.



All further instances of Spi opened from the same folder will produce Low-range 32M initial values, but the already open Spi is not affected.

to test this before closing the original Spi, a second instance of Spi can be started from the same folder.
the original will give High 32M initials, and the second instance will give Low.




(3) failed runs can cause 32M initials to switch back to High. It may not be possible to restore Low initials to a pi_rec.dat2 which has suffered a failed run. Instead, repeat steps (1) and (2), or replace 'corrupted' pi_rec.dat2 with a Low 16K copy.....

cheap

pi_rec.dat2 (16K = 0.094s)

[Zeus]

Great find cheap!! I had a feeling it had to be something along those lines as a fresh install of SuperPi usually gave good initial values but after some unfinished or errored runs the initial values raised.

Shame i sold my RE and 8600 now.

[3oh6]

sorry about the lack of updates guys...reviews, full time work, classes, life is just too busy right now. i am definitely not working on my i7 efficiency but will be getting back to benching soon enough. i will try to get an update done this week, but for anyone wanting to help out, all you have to do is quote the original post and you can get the BBcode

[ToyTen]

4460.1......ToyTen.............7.59.922....21x212. 4....849.6mhz.....6-6-5-16-1T....2/8..X58



Can't do much better at this point, maybe will i update this score later, time is missing here too

[George_o/c]

I'll be joining you soon Jody, with i7 920 Finally

[dami1stm]

Hi my friend, I resumed the race with my timezone and i7 965 XE on RADBOX + 3X1Go Cellshock PC12800 7-6-6-18 (GTR) and my Asus P6T WS.


TZ 8M / 3885.9Mhz - dami1stm - 8mn59.875s - 27x143.9 - 3x1Go @ 1007.8Mhz 7-6-6-16 - 2:14 - P6T WS Pro X58




TZ 9M / 3464.8Mhz - dami1stm - 9mn59.782s - 24x144.4 - 3x1Go @ 1010.6Mhz 7-6-6-16 - 2:14 - P6T WS Pro X58



Maybe my next CPU will have a better memory controller ... If so, I would take some Elpida chips memory ...

[George_o/c]

@cheapseats: Can you explain a bit that tweak you gave us ? I got a little confused

[cheapseats]

@cheapseats: Can you explain a bit that tweak you gave us ? I got a little confused

sure George, though i'm hoping someone can confirm it working . (i guess Zeus might have done - cheers bud )

there's really only one thing to do: create or paste pi_rec.dat2 (including 16K time = < 0.109s) in the Spi folder. Then open SuperPi.

also to mention - it's a different effect on 32M initial values than running 16K directly before each 32M run (which i do). Running the sub-0.109s 16K time for the Low/High switch is only needed once to generate the pi_rec.dat2 file.

ignore the bit about running 2 Spi's, it was just to show that the 16K result in pi_rec.dat2 causes the 0.5s difference in 32M initial values, and only before SuperPi is opened, not after. (ie: the open Spi which runs the 16K will not switch to Low 32M initials)

seems like a bug in SuperPi rather than being hardware- or OS- based, but i find it consistently 0.5s faster to finish from Low 32M initials. Although the Loop times will appear slower, the Loop 24 to finish stage is faster.

Of course many runs already show the Low type of 32M initial, but before now mine weren't deliberate.

this table covers the behaviour i found that is consistently repeatable/predictable:

the exception is after some failed runs (not all): 32M initials can switch back to High 'permanently' (until the pi_rec.dat is deleted or replaced for that Spi folder)

[George_o/c]

Wow, that's amazing, thanks a lot pal

[3oh6]

update done...a couple notes added to the first post:

Requests:
1/ Please only post one score per forum post...makes it easier for updating

2/ DO NOT use links to ImageShack with thumbnails...i will not go to that site, your score will not get added. full size images embedded in the post only please.

[cheapseats]

guys, just a 'minor' update/correction to this, but the general behaviour is still the same.

my earlier pre-conditions for 32M initials are not correct: Pi size and time are both irrelevant.

i found that any result for any Pi size, if present in pi_rec.dat2 before SuperPi is opened, can reduce 32M initials by 0.5s.

re-tested with runs from 16K up to 32M and they all produced the same effect on 32M initials. Don't know if there is a threshold time for any particular Pi size, but i also tested @ 3Ghz with same results.


new tests/results:

for High 32M initial:

1) delete pi_rec.dat2, open Spi (new pi_rec.dat2 created), run 32M (High initial showing)

2) delete pi_rec.dat2, open Spi (new pi_rec.dat2 created), close Spi, open new Spi, run 32M (High initial showing)

3) delete pi_rec.dat2, open Spi (new pi_rec.dat2 created), run Pi size X, run 32M (High initial showing)

for Low 32M initial:

4) delete pi_rec.dat2, open Spi (new pi_rec.dat2 created), run Pi size X, close Spi, open new Spi, run 32M (Low initial showing)


doubtful that this was really a sensible use of time, but the aim for me was to find a consistent way to get one or other type of 32M initial

cheap

[3oh6]

doubtful that this was really a sensible use of time, but the aim for me was to find a consistent way to get one or other type of 32M initial

cheap

copy wazza...different copy wazza tweaks and OS setup provide the two different types of runs your outlined.

[cheapseats]

copy wazza...different copy wazza tweaks and OS setup provide the two different types of runs your outlined.

i'm not sure if we're talking about different things what i found is independent of OS tweaks and CW.
it seems it's only caused by the pi_rec.dat2 file and Superpi itself.

[cheapseats]

4461.9......cheapseats..........9.59.578...7.5x595 .....952mhz.....7-6-5-15....5/8...X48

(Edit 03/03/09 - 3oh6, please ignore this run in next update lower clocks achieved !! )

finally got my first D9GTR kit..

beat my previous best 9.xx (with D9JNL), but without using maxmem or CW, and some slack left in tRFC.

LSC on / core 1 affinity / realtime / explorer closed / system managed page file size /
1 disk / 1 20GB partition / no maxmem / no CW

[cheapseats]

3982.9......cheapseats.........10.59.344.....7x569 .....910mhz.....7-6-5-15....5/8...X48

(Edit 29/04/09 - 3oh6, please also ignore this run in next update )

might be able to improve on this as my SpiTweaker settings + maxmem etc are still out of line
with what's been recommended here. Worth about 4 seconds to this run as they are

LSC on / core 1 affinity / realtime / explorer closed / 515-515 page file /
1 disk / 2 partition / maxmem 680 / SpiTweaker v1.0 CDT (680MB file) D>C (SPi C)

[cheapseats]

4436.4......cheapseats..........9.59.359...7.5x592 .....946mhz.....7-6-5-15....5/8...X48

sub 60 tRFC = scary vDIMM on this kit so i will leave that for later... only 2 days old

LSC on / core 1 affinity / realtime / explorer closed / 515-515 page file /
1 disk / 2 partition / maxmem 680 / SpiTweaker v1.0 CDT (680MB file) D>C (SPi C)

[cheapseats]

3557.6......cheapseats.........11.59.594.....6x593 .....949mhz.....7-6-5-15....5/8...X48

LSC on / core 1 affinity / realtime / explorer closed / 515-515 page file /
1 disk / 2 partition / maxmem 680 / SpiTweaker v1.0 (680MB file) D>C (SPi C)

[cheapseats]

3269.9......cheapseats.........12.58.750.....6x545 .....908mhz.....7-6-5-15....5/8...X48

LSC on / core 1 affinity / realtime / explorer closed / 515-515 page file /
1 disk / 2 partition / maxmem 680 / SpiTweaker v1.0 CDT (680MB file) D>C (SPi C)

[cheapseats]

5040.0......cheapseats..........8.59.688...8.5x593 .....949mhz.....7-6-5-15....5/8...X48

cheap sub 9 : Q820A606 + Noctua NH-U12P + 1 x NF-P12 fan + weather cold enough

LSC on / no core affinity / realtime / explorer closed / 515-515 page file /
1 disk / 2 partition / maxmem 680 / SpiTweaker v1.0 CDT (680MB file) D>C (SPi C)

[hiwa]

Tweak
OS : 2K3
Maxmem=600, Pagefile=512/512
Priority CPU1, Affinity Realtime
Only 3 Services (rpc/shell hw detect/event log)
waza
reg tweak nda

[George_o/c]

@ToyTen: Sorry for the long waiting man, but I was kind of busy for the last one and a half month or so

You got PM though, and I can assure you - you are going to like it Keep benching my friend

Nice work guys, I will join you in a while, now that I got the time to bench again

[hiwa]

+ some reg tweak

[Shahryar_NEO]

@ Hiwa : http://hwbot.org/listResults.do?cpuM...true&limit=100 amazing score !!! with that clock !