狂少's CDT-tweak

[v_parrello]

I made 13.812 sec. at 3.6 GHz CPU and DDR3 platform (DDR3-1864 6-5-5-12).

For me it is really hard to think about a tweak able to make a gain of 0.344 sec. in superPI 1M with DDR2-1233 5-5-4-10. It is really a big challenge, I can not figure out how this is possible.

But before to give my final opinion a I have to try if this works for me. It would be really great!
I do not know if I will be able to reach 13.468 sec. but for sure if the tweak works I would make better of my best 13.812 seconds.

[Enoc]

no screenies but i'll say this...a few months ago he explained part of the tweak in coolaler.com, he got it on a video, part 3-5..if a remember well...

so like 3 weeks ago with my dfi p-965s running 514x7 1:1 4-4-4-4, 3-19-2-9-9-3-7-3 boot from bios, 24/7 windows xp pro-sp2, i got 13m.17 seconds on pi...nothing tweaked...
with copywaza 1.2gb file ,realtime, large cache sytem=1, 512mb pagefile in disc d, PI on disc d, so cut and paste file D: to C: and C: to D: , got 13m and 10seconds..

with thesame as above, i took a file and made 3 folders of 650mb files,and 1 of 180mb, all pics...put the 3 650mb on a folder, and left the 180mb file alone, i first compressed the 180mb file(it's to get up the virtual memory faster), this is all on D:, after i compressed the 3 other files, after finished compressing cut it to C:, waited untill i got the best virtual memory ,system got stable...got 13m:04 seconds...

food for thought...for me is good enough..

.....It's VERY EASY you know......I will run the 32M at 3800MHz and after the resault is out and I'm ready to screen capture it, I'll DOWNGRADE my CPU frequency via clockgen/setfsb to show me 3597MHz and YOU ARE READY
EDIT: You must be dreaming or so IF you think that DDR2 (as OPB 32M above) can beat DDR3.....NO WAY....

as for this...i want to see you with 5:6 divider running 652mhz with his timmings 32m stable with 2.35 volts on a asus p5k deluxe....

for him to be running 3800 mhz as you say he would need 542mhz fsb(652 on mem) on the
P5k deluxe, and guys with this board know how easy it is to get it a 2.35 volts ...seriously Hipro let the animosity aside and accept facts...or there is a program that changes the ram volts in windows ? or a program that tricks aus probe and cpuz vcore report ?

just focus on the tweak...

[hixie]

Windows XP Pro SP2, small amount of OS tweaks, and some services disabled.
No other special tweak
time is 1m 43.157. First picture

OPB's CDT-IV with exact same OS as before.
Time is 1m 42.984. Second picture

Hixie tweak rev 1.0 (that i made up a couple hours ago at work) with exact same OS as the previous two.
Time is 1m 42.859. Third picture.

Seems like my tweak is work working better than OPB's CDT-IV

[hipro5]

no screenies but i'll say this...a few months ago he explained part of the tweak in coolaler.com, he got it on a video, part 3-5..if a remember well...

so like 3 weeks ago with my dfi p-965s running 514x7 1:1 4-4-4-4, 3-19-2-9-9-3-7-3 boot from bios, 24/7 windows, i got 13m.17 seconds on pi...nothing tweaked...
with copywaza 1.2gb file ,realtime, large cache sytem=1, 512mb pagefile in disc d, PI on disc d, so cut and paste file D: to C: and C: to D: , got 13m and 10seconds..

with thesame as above, i took a file and made 3 folders of 650mb files,and 1 of 180mb, all pics...put the 3 650mb on a folder, and left the 180mb file alone, i first compressed the 180mb file(it's to get up the virtual memory faster), this is all on D:, after i compressed the 3 other files, after finished compressing cut it to C:, waited untill i got the best virtual memory ,system got stable...got 13m:04 seconds...

food for thought...for me is good enough..




as for this...i want to see you with 5:6 divider running 652mhz with his timmings 32m stable with 2.35 volts on a asus p5k deluxe....

for him to be running 3800 mhz as you say he would need 542mhz fsb(652 on mem) on the
P5k deluxe, and guys with this board know how easy it is to get it a 2.35 volts ...seriously Hipro let the animosity aside and accept the facts...or there is a program that changes the ram volts in windows ?

just focus on the tweak...

A friend of mine can have his DDR2 rams at OPBs timmings up to 1350MHz....

[massman]

Enoc and Hipro5, would you please stick to the subject, which is the tweak and use the PM function for other OPB-related discussions? Thanks

Hixie: are you sure you're applying the OPB tweak correctly? It seems kinda weird that your modification works better than a tweak which has been created during months (N/O).

[Eldonko]

Looks to me it is similar to copy waza except it also works for 1M and it works a few s better for 32M. Whats the point in arguing about it, if you don’t believe it works or you think your tweak is better just don’t use it and move on.

[M.Beier]

I'd like more evidence on the 1M matter... If that 0,2-0,3 sec is true, well thats just amazing... and simply, too amazing...? Thats evens more then 300mhz cpu juice more on a fully clocked 1M run with C2D..?

[Misko]

First I just want to say that I do not know OPB and that I do not have nothing against him (all dough I do not have nothing against no body here or at OCX), also my personal thinking that maybe he has some special tweaks and on another hand maybe that he has not, up to day status by my opnion: He is 100% clean (valid results) until someone get some proof against it.

But some things get me wondering how many different super tweaks OPB has? What about this one I do not know the name, let we call it "turbo at last iteration", as I mentioned before, generally, the time of last loop takes about 80-84% at the average time of 1-20 loops but he can take that too only 66% if he wants (like on nvidia i680)



OPB result compared to "normal" one


With this (only this tweak) he can shave 5sec on ln2 or lets say around 7sec on 3.6Ghz:

Then with CDT again he can shave I do not know exact value but lets say VERY MUCH, if he can make that 1M on 3.6ghz. So if he use all his tweaks together he will be lets say half MINUTE faster then Gautam. Also he did only once that last iteration speed up, why he is not using it anymore? And he is gone from WR chasing why that if he has all knowledge and tweaks?

My proposition is that someone (best someone from Team Japan) go to OPB and sit there while OPB benchs (of course if OPB is OK with that) Team Serbia (and I think many more Teams will do the same) can give some money to that person so he can pay his trip so we finally know for 100%.

Misko

[massman]

The "turbo at last iteration" is part of the CDT tweak, I think. I remember a screenshot of one of the runs explaining that the true superpi super times can only be reached with a good last 4 loops (the so-called bonus loops).

[BeardyMan]

A friend of mine can have his DDR2 rams at OPBs timmings up to 1350MHz....

I have done it thighter then OPB at 1353Mhz


srry for offtopic

[Misko]

The "turbo at last iteration" is part of the CDT tweak, I think. I remember a screenshot of one of the runs explaining that the true superpi super times can only be reached with a good last 4 loops (the so-called bonus loops).

hm he didn't have turbo on at some of his previous WRs (like WR on 975x):

[hipro5]

My proposition is that someone (best someone from Team Japan) go to OPB and sit there while OPB benchs (of course if OPB is OK with that) Team Serbia (and I think many more Teams will do the same) can give some money to that person so he can pay his trip so we finally know for 100%.

Misko

Just FORGET about this proposition of yours.....
The same proposition was done half a year ago and HE (The LORD), din't accept it....So don't bother with this any more....

IF he was OK as he want's to be(?), he whould accept someone at his benching place to CLEAR things up.....All other things are just EXCUSES.....PERIOD.......

I'd like more evidence on the 1M matter... If that 0,2-0,3 sec is true, well thats just amazing... and simply, too amazing...? Thats evens more then 300mhz cpu juice more on a fully clocked 1M run with C2D..?

[Gautam]

This bickering isn't going to get us anywhere.

eva>> what size copy-waza are you using?

I use MAXMEM=512 with a 5.2GB copy waza. I only copy the file once. The gain is about 15 seconds quite easily.

Stephen>> Conroe's are able to fit the 1M calculation in the L2 cache, which is what reduces their margin of error and makes them incomparable with AMD. The tweaks that work on AMD don't work on Conroe. We need a Conroe for a fair comparison.

However for the folks asking for 1M, I can guarantee you won't see anything there. 32M is the test of choice with Core 2's since it enlarges the effects of all tweaks much more. (PiFast does as well)

elmor>> I want to see matched settings as mine. Notice how low my FSB is. I also want to see a third party running the calculation. You've got experience running his tweak and you have the hardware to run at 400x9 2:3 4-4-4-4 I'm sure. I think most of us also would put our faith in you. Give it a run and let's see what happens...

[Planet]

Hipro5 what speed do you install the operating system?

[massman]

I actually have no idea when he started using the tweak (can't ask either has he ignores my pm's ), I can only tell what I've read at the Coolaler boards translated online (so crappy english).

Can't find the pic atm, but the superpi 32M run was divided in three parts, two of them which can only be slightly faster with tweaks, the last one (bonus round = loop 20 to 24) can be a lot faster using the CDT tweak and copy-waza.

Does anyone actually know how superpi is run by the system? At what point it uses the memory/cache/virtual memory?

@ Beardy: lower timings =/= faster. Subtimings have to be set at the correct level in order to be really efficient. Imho, a lot depends on how you set the CL and tRAS and the timings of the latency between both.

[eva2000]

I'm not taking anyone's side as from my results http://www.xtremesystems.org/forums/...1&postcount=38 neither copywaza or CDT-IV give me such enormous gains compared to untweaked run...

If a person can't replicate a tweak such as CDT-IV, doesn't mean it doesn't work or is a cheat. I can't replicate any form of copywaza so doesn't mean it doesn't work heh

for copy waza i've tried all sorts of sizes from 400MB through to 2.8GB heh

[massman]

Hipro5 what speed do you install the operating system?

That's a tweak I mentioned at OCX once (the old forums, before the server crash), but no one believed me. Johnny Bravo tested it and he got the same results

[hipro5]

Hipro5 what speed do you install the operating system?

Ahhh.........I've heard of that too........and out of curiosity (maybe to get faster resaults), I have install them EVEN at 5300MHz.....NO GAIN AT ALL......Sorry.....

[Gautam]

I'm not taking anyone's side as from my results http://www.xtremesystems.org/forums/...1&postcount=38 neither copywaza or CDT-IV give me such enormous gains compared to untweaked run...

If a person can't replicate a tweak such as CDT-IV, doesn't mean it doesn't work or is a cheat. I can't replicate any form of copywaza so doesn't mean it doesn't work heh

for copy waza i've tried all sorts of sizes from 400MB through to 2.8GB heh

Note that 5.2>2.8.

[BeardyMan]

T

Stephen>> Conroe's are able to fit the 1M calculation in the L2 cache, which is what reduces their margin of error and makes them incomparable with AMD. The tweaks that work on AMD don't work on Conroe. We need a Conroe for a fair comparison.

Makes sense yes

this run made by GHZ's has been done at same dividers but note the timings used
I'm sure that's a well tweaked system, so the tweak should pull of a fully 0.3secs then if applied well.
I would like to test it but i'm on a AMD rig
so any volenteers like gautam suggested

[Misko]

Just FORGET about this proposition of yours.....
The same proposition was done half a year ago and HE (The LORD), din't accept it....So don't bother with this any more....

IF he was OK as he want's to be(?), he whould accept someone at his benching place to CLEAR things up.....All other things are just EXCUSES.....PERIOD.......

Didn't now for that one, but I do not see anything wrong with that proposition. I just see fredyama like to travel a lot so he would be perfect person for that one.

[BeardyMan]

I

@ Beardy: lower timings =/= faster. Subtimings have to be set at the correct level in order to be really efficient. Imho, a lot depends on how you set the CL and tRAS and the timings of the latency between both.

I think it's obvious wich subtimings are the most thights it's there to compare lol

[massman]

I think it's obvious wich subtimings are the most thights it's there to compare lol

Note how I say that it's not about which timings are the tightest . You're subs might be tighter, they will not be faster in se.

Back on the tweak, I presume testing should be done, but correctly. By the way, difference between Server03 and XP?

[elmor]

elmor>> I want to see matched settings as mine. Notice how low my FSB is. I also want to see a third party running the calculation. You've got experience running his tweak and you have the hardware to run at 400x9 2:3 4-4-4-4 I'm sure. I think most of us also would put our faith in you. Give it a run and let's see what happens...

The thing is I don't. I might have a E6750 (this one is stuck at 400mhz fsb) that can do atleast 450x8 later this week.

[eva2000]

Note that 5.2>2.8.

hmm tried 3.7GB was 2 seconds slower HA HA... will try larger file and see what happens