32m SPi Low Clock Challenge

[jcmarfilph]

very nice fafi Alex.

Congrats..

Pinoy Powah ^_^

[dduckquack]

wew more and more of us playing this game... wish i still have the time to do spi 32m

[Benji Tshi]

It seems that my previous posts was lost during XS migration....i was too busy to submit it again, but here i am with my first sub7, made with my friend Casimir37 (more to see here)

6.xx
5075.2......Benji Tshi & Casimir37............6.59.797....29x175 ....1050mhz.....7-7-6-20....2/12..X58

[Benji Tshi]

And here is the data which was lost...

7.xx
4400.4......Benji Tshi............7.57.093.....24x183 ....1100mhz.....7-8-7-20....2/12..X58



First bench with Perfect Storm Elpida Hyper, and not enough time to do it...I will try it again soon i hope, and i'll aim ~4360
Under SS, -45°


8.xx
3868.9......Benji Tshi............8.59.931.....20x193 ....967mhz.....6-7-7-18....2/10..X58


8.xx
3453.6......Benji Tshi............9.59.625s.....18x192 ....959mhz.....6-7-7-18....2/10..X58


Done with the same Perfect Storm, but aircooling (so uclk is lower like memory frequency)

[pastelidis]

Ι thinh Βenji that the Perfect Storm are Perfect...!

[Benji Tshi]

Nooo...., you're sure ??

[cheapseats]

4424.3......cheapseats.........09.59.953...7.5x590 .....944mhz.....7-6-5-15....5/8...X48

all extra speed compared to my 09.59.906 :: 4431.5mhz run is from using a different pi_rec.dat2...

now it seems that there are at least three 32M run types determined by the pi_rec file, which affect Inital value,
Loop times, Loop 24 to finish and overall speed. the same difference also shows without waza.

XP sp2 / LSC on / core 1 affinity / realtime / explorer closed / 512-512 page file / 1 disk / 2 partition /
maxmem 680 / SpiTweaker v1.0 CDT (680MB file) D>C (SPi C) / pi_rec.dat2 cheap 'speed' version




did this run first on 0401 and got the same finish time as 3oh6's 4424.3mhz

click for larger image

[TheKarmakazi]

awesome work cheapseats. Any clues on what to look for with the different pirec.dat? Were you running 16K to get it or another size?

I was actually trying to take this timezone, but you beat me to it

[cheapseats]

thanks buddy i only beat 3oh6 by 0.016s, so it hasn't moved the target much.. the cpuz clock will drop to 4424.2mhz but not everest so i wanted to go a touch faster to nip in ahead

the pi_rec i'm using now has 16k = 0.078, 1M = 11.5x, 32M = 9m19x, and it's been sitting unused on a spare disk since february.. just used the disk again for some tests and the runs were faster than i expected. Thought it was the effect of a 2 disk setup at first, but after testing it turned out to be the pi_rec. The 4424mhz runs i posted were with 1 disk using a copy of the fast pi_rec from the spare.

don't know what's in the file that is relevant to the outcome, but @ 9.xx clocks its ~1s faster than if i use my usual pi_rec (16k=0.078s). Both files give Low initials, but one of them is 0.5s faster up to Lp 24 and 0.5s faster Lp 24 to finish.

been trying to find out what was causing my slow down in Lp 24-to-finish since i replaced my OS a while ago and no idea that i'd lost another 0.5s on top too check out how my 24 to finish time tightens up between the 4431mhz run (~20.6s) and 4424mhz run (~20.1) both with same setup etc - just different pi_rec.

hope i can work out how to produce another fast pi_rec from scratch, otherwise i will just post the file i already have

btw, there is a hardware-based way to get 589x7.5 or even 588, but it needs quite high vdimm to run all the subtimings and tRCD5

here's my test run without waza or subtimings (or fast pi_rec). haven't finished a full-fat run yet though.

589x7.5 / 942mhz / 7-5-5-15 / tRFC 60 / vDIMM 2.25V set

[Moonman]

5982......Moonman............7.59.218.....10x598 ....957Mhz.....7-6-6-24....5/8..X48

Had some fun with RE
Memory chips are D9JNJ and these timings are as low as they can go. 2.2 Vdimm
But I guess it ain't bad for ~40$/2GB

[kintaro]

12.xx
3395.5......kintaro............12.56.256s.....7x48 5 ....970mhz.....8-7-7-22....1/2..X48

my first attempt with CSX DIABLO.

[kintaro]

11.xx
3720.6......kintaro............11.58.396s.....8x46 5 ....930.2mhz.....7-7-7-21....1/2..X48

[kintaro]

10.xx
4131.1......kintaro............10.57.104s.....8.5x 486 ....972.1mhz.....8-7-7-22....1/2..X48

[sofos1990]

i7 965+REX2+Adata 3*1gb+some reg tweaks
9,xx

[kintaro]

12.xx
3347.2......kintaro............12.58.252s.....6x55 8....929.8mhz.....7-8-5-20....3/5..X48

[marcolecorse]

7.xx
4643.9......Marcolecorse............7.51.172....21 x221 ....884mhz.....8-8-8-20....2/8..X58

[Maduk]

great run Marco very good i920

[coop]

11.xx
3693.9......coop..............11.59.453s.....7x527 ....633.2mhz.....5-5-4-8....5/6..P45


hardware noted below, maxmem 600, lsc, 1gb waza, pagefile 500/500, realtime

[IMG][/IMG]

[kintaro]

12.xx
3276.1......kintaro............12.59.359s.....6x54 6....910mhz.....7-6-5-15....3/5..X48

[kintaro]

12.xx
3264.5......kintaro............12.59.953s.....6x54 4....906.8mhz.....7-6-5-15....3/5..X48

this is my best run so far. i think Diablo's have no more to give.

[Zeus]

Here 's sub 19 minutes for the green team:



Much room for improvement and i'm sure DDR3 will slaughter this score but i'm prepared for a little challenge.

Hoping to get sub 18 min soon.

[kintaro]

12.xx
3264.5......kintaro............12.58.641s.....6x54 4....906.8mhz.....7-5-5-15....3/5..X48

[Stelios]

Memory speed reported in cpu-z is wrong due to bugged divider ....
In reality it's about 816 ,mhz .

[Moonman]

5864......Moonman............7.59.703.....9.5x617. 3 ....987.6Mhz.....7-6-6-15....5/8..X48

[Moonman]

5030.7......Moonman............8.59.813.....8.5x59 1.3 ....946.9Mhz.....7-6-5-15....5/8..X48


Made a video of last two iterations.
http://www.youtube.com/watch?v=VjkVC5wAqOQ