The CDT and copywaza lab

[[XC] gomeler]

Just curious, has anyone run a few 32M runs with some HDD monitoring applications to see what sort of hits are being addressed to the disk? This will solve all your questions of how important the drive is. You could also artificially cripple the drive by loading up the drive with some throughput tests while running SPi but this would also affect the memory available to SPi, skewing the results. If I had a second machine I'd try to run the HDD tests myself but all I have is my webdev server right now.

[dinos22]

- CDT basicly is an improved Copy-Waza. It's build on the same fundamentals, but is more elaborate. The average gain depends on what frequency you're running, but at 3.6 - 4.0G, the gain should be +/- 6 to 10 seconds (if copy-waza and cdt are done correctly).

it has to be a bigger gain because OPB did 12m 39s at 3.6GHz with CAS5 even

[hipro5]

it has to be a bigger gain because OPB did 12m 39s at 3.6GHz with CAS5 even

Quoted and I also repeat......
You MUST see ~10 - 15sec LESS than Copy Waza according to the letest OBP's 32M at 3.6GHz...

I didn't know about that BUT it's good that works for both of us.

I'll re-run if you want doing copywaza again.

But how big a CW file?
I tried 4GB already, two different drives, 1st partitions.

Yes make it ~4GB and cut/paste from one HD to another for 3 times.......Also put a fixed Page File at 384-384.... Thanks....

[mrlobber]

Well, even if kind of "semi-working" CDT gave 6 seconds vs a good copywaza, I'd be satisfied Eagerly awaiting your results and explanation massman.

[elmor]

ELMOR: have you enabled the Largesystemcache reg key? Without that, CDT and Copy-waza are useless. (Beardy: that's the setting we spoke about on MSN. Of course it's obvious :p)

Yes, of course I use that

[JD1985]

Anyone mind posting a link to the copy waza tweak having some troubles finding it and would be interested in trying it.

[BenchZowner]

Just curious, has anyone run a few 32M runs with some HDD monitoring applications to see what sort of hits are being addressed to the disk? This will solve all your questions of how important the drive is. You could also artificially cripple the drive by loading up the drive with some throughput tests while running SPi but this would also affect the memory available to SPi, skewing the results. If I had a second machine I'd try to run the HDD tests myself but all I have is my webdev server right now.

[KTE]

I want you to pay special attention to what I'm going to say next guys...

OPB... you FRACKING GENIUS!!

OPB... damn what the hell?!

My God heeeeh!

I've no words.

I've no words.

I've finished my testing for today. Let me get some tranquilizers first.

[hixie]

I want you to pay special attention to what I'm going to say next guys...

OPB... you FRACKING GENIUS!!

OPB... damn what the hell?!

My God heeeeh!

I've no words.

I've no words.

I've finished my testing for today. Let me get some tranquilizers first.

May i ask whats going on?

[massman]

So, like I said, I did a few 32M runs yesterday, made an excel sheet in which I put the time of every loop and the time difference between a loop and the preceding loop. I can share my excel-file if someone wants it ... not that special, though .

During the testing I've switched to a faster sata drive with more cache. I gained a bit more than 1 second.



This is the pic with the two segments I talked about. The first area will be explained later, the second one is quite important. It shows the difference in time between the loop24/output and loop23/loop24 ("/" is not the symbol of division). Normally, loop24/output should be about 1 second faster than loop23/loop24, in my case I can't do this. My times would be 2 seconds faster.



The first segment I talked about is very interesting. It shows an almost lineair decrease.
Line3: ran superpi from C, while pagefile on D. Used 3x632 files and copied from D to C first. IDE hdd for page
Line2: ran superpi from D; while pagefile on D. Used 3x632 and copied from D to C first. IDE hdd for page
Line1: ran superpi from D, while pagefile on D. Used 1 file exactly 3x632MB and copied from C to D first. Sata HDD for page.

I need to do some extra testing before I can jump to conclusions. But there are a few things that catched my eye.
- The faster you get that lineair decrease, the better the time.
- Running superpi from the same harddisk as the pagefile is on will be slower if not properly tweaked, but if well-tweaked = faster.
- Pi from page HDD = slower loop24/output.
- The size of the systemcache is important (the higher, the better), BUT available memory must be the same size of the systemcache to get the best results!

NOTE: Initial time was 14m32s, after some testing I went to 13m59s, now I'm at 13m56s.

[mrlobber]

May i ask whats going on?

I think he just got THE TWEAK working full scale Cant wait the report.

[hixie]

I think he just got THE TWEAK working full scale Cant wait the report.

Could it be the CDT-V tweak OPB mentioned before? If thats the case i'm looking forwards to CDT-VI and CDT-VII.

I'll see if i can do some tests on a I-RAM, latter this week.

[KTE]

Copywaza is an old method - well known and I've ran it countless times before. Gets me about 8-5 seconds drop usually (at a Pi test which runs in 15-10 minutes).

May i ask whats going on?

I think he just got THE TWEAK working full scale Cant wait the report.

Its tricky but it ONLY works if you get it just right and boy does it work well. Damn what a tweak, this is more a miracle!

Took me around 15 attempts over two days to learn it right and EVEN still I've not perfected it no where near to the screenshots I saw of OPB. The guy will slaughter anyones Pi times with this tweak at the same clocks, timings, hardware, and settings. Make no mistake about that.

[KTE]

Nice to see this back on track again, great work saaya.

Nowhere near Robert, that's crotale I do not have so much more info to add but I can sum up what I've experienced so far.

First time I tried the CDT Tweak I had those crazy results that I posted in the previous thread. However after reformat I was never able to get System cache about the same as Avaliable memory again and hence no boost (that's what I believe gives the boost) except a few seconds which could probably be aquired with hipros method aswell. I started using maxmem = 600 and I actually got worse times with that than with 620 so one could do some experimenting there.

Currently I'm playing with P5E3 WS Pro as I got kinda frustrated over not getting this to work again, but I'll get into it once again when I have some time to spare.

//Jon

Sorry Jon I get mix up between you two all the time.

Try and make sure you get above 500MB system cache and above 410MB available memory before you run Pi. Below that there wasn't much difference for me.

[before]

Took me around 15 attempts over two days to learn it right and EVEN still I've not perfected it no where near to the screenshots I saw of OPB. The guy will slaughter anyones Pi times with this tweak at the same clocks, timings, hardware, and settings. Make no mistake about that.

lol

So what about teaching us dude? I eventually tried it and not succeeded in beating my cw method... may be you know something else, something we don't know. I'm not sarcastic, honestly if something exists, I really would like to learn it.

[dinos22]

lol

So what about teaching us dude? I eventually tried it and not succeeded in beating my cw method... may be you know something else, something we don't know. I'm not sarcastic, honestly if something exists, I really would like to learn it.

i'm with you on that one Xavier heheh

[massman]

lol

So what about teaching us dude? I eventually tried it and not succeeded in beating my cw method... may be you know something else, something we don't know. I'm not sarcastic, honestly if something exists, I really would like to learn it.

And how does you CW method work?

[KTE]

lol

So what about teaching us dude? I eventually tried it and not succeeded in beating my cw method... may be you know something else, something we don't know. I'm not sarcastic, honestly if something exists, I really would like to learn it.

I followed the same routine you did from reading here late on. Kev did all the work and my utter thanks to him for putting up with **** when he could do and new what no one else here did. Rather than keeping it hidden he spoke openly and tried to help everyone get and beat his times. Salute to that.

If I get hold of a Phenom straight after release and someone else has one to test as comparison then I'll show you the time difference it can make if done properly. Pretty damn sure about that. Just run it as good as you can and try to keep variations out of it as little things can make it fluck. You've seen my 8M results above with a Presy Celeron (similar to your 32M) and I'll show you my new test results for CopyWaza and CDT. Just at work yet.

Did you get your system cache and available memory above 450MB or 500MB each before running Pi? You need to do that, read kevs words properly honestly. He 'aint kidding. You need to balance both values and have them as high as possible or it doesn't quite show its true nature.

There's no doubt the tweak influences hardware caches, buffers and prefetch. It decreases latency very much (I've only tested on a slow CPU yet, it may give less drops in time on a faster core).

BRB

[dinos22]

KTE
it would be nice to see your conroe 3.6Ghz times
that's the best indication and comparison points for this

[Zeus]

Man i hope you're right.

I 'm not sure you can replicate the very same thing with a C2D setup though.

Last night i really tried hard, got 527984 for available memory and 532088 for Systemcache, which to me is close to supreme but i did not see the accelaration in the third group of loops as it should.

I'm not done testing myself as i still have faith in this tweak and hope to get it working properly on my C2D setup.

Reading all OPB's ramblings on this subject makes me feel like a complete n00b, the study he made in SuperPi calculation time is nothing short of awesome.

Just trying to comprehend all this complicated stuff takes some time.

[mrlobber]

...You need to balance both values ...

That's the goal I've failed to reach with my 3.6Ghz C2D testing so far.

[KTE]

Sorry dinos22 I don't have a C2 here, can get one probably in 10-12 days (and your tempting me now seriously ) but I don't see a need for the hassle and am not planning on getting one soon. Got a load of old parts to sell for now. Maybe get a Penryn late March then to test>sell but for now, waiting for Phenom if all goes right. FWIW that Celeron is also my work system, used to be mine but I sold it to work place (where I still use it) and I only have a personal lappy for now (this isn't my home country either).

Zeus: I doubt it'll make anywhere near the same difference with faster cores (more powerful cores) but it certainly will make a difference over a top CW. I need to run some tests myself on various platforms first. I have a P4 chip lying around somewhere but again its completely in parts, some missing and no conroe either.

BTW the forums timing out for me repeatedly.

[massman]

Did you get your system cache and available memory above 450MB or 500MB each before running Pi? You need to do that, read kevs words properly honestly. He 'aint kidding. You need to balance both values and have them as high as possible or it doesn't quite show its true nature.

The higher the better?

I'm running with systemcache over 560MB and Available memory over 550MB.

[BeardyMan]

The higher the better?

I'm running with systemcache over 560MB and Available memory over 550MB.

no, otherwise you wouldn't been setting 600 or 680 as a max, then you could just leave it like that 2048/4096 etc..

I've done some more testing ,i lend a seagate 160GB 16Mb hdd yesterday 7200rpm and compared it to the same disk but 8Mb and a maxtor 5Gb wich has i believe no buffer

maxmem 680 + no pagefile + 632mb file + performance set to systemcache + 160Gb 8Mb = average gain of 0.7secs in SPI 1M on my AMD rig.

maxmem 680 + no pagefile + 632mb file + performance set to systemcache + 160Gb 16Mb = average gain of 0.9secs in SPI 1M on my AMD rig.

maxmem 680 + no pagefile + 632mb file + performance set to systemcache + maxtor 5gb = average gain of 0.2/3secs in SPI 1M on my AMD rig.

the best HDD woudl be this,
150 GB, 10.000 rpm, 4.5ms Zugriffszeit, 16MB Cache,

10k rpm instead of 7200
4.5ms instead of the 8.9/8.5 and ofcourse 16mb cache.

[KTE]

OK. You saw my old CDT tests right HERE.

Stock 8M best time was around 10m25s
With best CopyWaza was around 10m20s
With best CDT was around 10m15s

You can't beat those with that setup, doing those things, that way.

New testing:



Changed 2nd HDD-> Now both have same cache/size.
Different monitor and res is smaller.
Same version of OS (a fully working one although this is an older install).
Disconnected all USB, optical, drives etc that are extra, disabled everything extra in BIOS.
Page-file kept consistent at 384-384MB all throughout testing: http://i222.photobucket.com/albums/d...n/pagefile.png
You can see my normal services running here: http://i222.photobucket.com/albums/d...n/Services.png
You can see my bench setup services running here (1): http://i222.photobucket.com/albums/d...ervices-oc.png
Processes/Memory/Cache at bootup: http://i222.photobucket.com/albums/d...bootupmain.png

As you can see from the stock results, today this system was faster than ever, even yesterday (by 4-5 seconds). More Windows tweaks.

Straight jump to the testing. Saaya asked (not verbatim) "Can you reproduce the results?"
With testing a millisecond bench, how many people can get their fastest record results every time they try it? I think we can agree, that you run it time and time again but they are usually one offs a bit better than the rest. But he means to reproduce the gain by a tweak. (Spi -1ms makes a lot of difference as loops go on and even how fast you hit Enter makes difference )

I can replicate all of them with each technique.

Still I complained "I can't get above 345MB/345MB". That I need to do to replicate Kevs results with that technique.

Experiment? You have to count everything, even your fudged results. These tests were ran straight one AFTER another -> no extra results missing from here. This is all that I ran. So nothing is hidden.

1st CDT test I ran this morning before work (383MB/380MB ): 10m16.266s
http://i222.photobucket.com/albums/d.../8M-10m16s.png

2nd CDT test ran (381MB/381MB): 10m 14.043s
http://i222.photobucket.com/albums/d.../8M-10m14s.png

You have NO idea how chuffed I was at that time. It is impossible with that system without CDT. Don't go away yet though...

3rd CDT test ran (386MB/384MB): 10m 10.768s
http://i222.photobucket.com/albums/d.../8M-10m10s.png



4th CDT test ran (386MB/384MB): 10m 9.746s
http://i222.photobucket.com/albums/d...n/8M-10m9s.png

(sorry messed up the window size and it wouldn't expand up later )

Crazy gains. 10m25s to 10m9s. At that time I still did not manage to get high cache (etc). This was my best 8M/32M time uptil now (the rest were ran on other days higher clocks): http://i222.photobucket.com/albums/d...i-complete.png

Just wait till what happens next... after a while.

Now to CopyWaza for a brief moment:

I already KNOW what I get with CW. We're not testing that here because its an old tweak. One thing I messed up is I used a 3.74GB file instead of 4GB. I doubt its a huge difference (from previous experience on this setup): http://i222.photobucket.com/albums/d.../winrar4gb.png

3 runs and the slowest was 10m 18s and the fastest was this @ (383MB/153MB): 10m 16.647s: http://i222.photobucket.com/albums/d...s-copywaza.png

THEN, fed up, I tried CDT again 3 times. I wanted to get 500MB/500MB like Kev...

5th CDT test ran (420MB/524MB): 7m 44.188s
http://i222.photobucket.com/albums/d...n/8M-7m44s.png

Holy fork

It dropped 7 seconds EACH loop consistent!

You need to know one thing.. I CANNOT run the first result below 7 seconds in ANY case what-so-sever (Error non convergent hundreds of time) UNLESS I did CDT and the System cache went above 500MB.

I doubted that result a bit

Still not perfect like Kevs, so I tried again for a run->> look at the time upper right corner for all shots.

6th CDT test ran (431MB/530MB): 7m 27.533s
http://i222.photobucket.com/albums/d...n/8M-7m27s.png



Simply unbelievable. That's like you testing 100 ES Phenoms being sore clockers at 3GHz max and cold bug -50C and then suddenly, you chose to test another one and it reaches 7GHz no cold bug

That is far far below the best. I was just shell shocked completely. What is happening? It dropped 7-10 second each loop!??! (yes I ran a mobile stop watch because I couldn't believe it )

Now I NEVER ran CDT again, but straight after the above finished (and I recorded it) I ran 8M again.

7th CDT test ran (411MB/520MB): 7m 27.604s
http://i222.photobucket.com/albums/d.../8M-7m27s6.png



These are not the fastest. It can be beat by using this method better although I ran out of time. Someone explain to me what is happening?

Also, keep your eye on the 32M time (stock of today). Tomorrow morning I'll go in again and test 32M after CDT.

I believe you can possibly gain anywhere from 10-30 seconds with a C2 in 32M if done perfectly. Maybe more, but I've not tried.

BTW the clocks of the CPU/RAM are not highest, they're just what I prefer. They can be thrown higher to 3550 stock volts (which is 2/3 benchable) and 3700/435 2-3-3-6 stock volts on all components, max. The clocks above I used are ALL only altered in BIOS at bootup (no PLL tweaking) and because they're full stable (no errors possible).

Hope you had a good read.