PCMark05 Tweaking Thread

[xXSebaSXx]

I forgot to tell you... it's case sensitive.

[Mikecdm]

At first none were caps, then i thought it might be the problem, but I didn't realize that the M was supposed to be capitalized.

 

I also forgot to thank CherV for sharing, Thank you.

[El Gappo]

[pro]

LOL good one, i like it

[dinos22]

 

:D:D:D:nana:

 

from what i hear, it's going to be exactly like that

[Gluvocio]

 

WOOT ! I get badly pwned

Much congrats PRO

 

I'll search an answer trying to use my sempron 145 image decompression on server 2008 I hope be lucky

Edited May 19, 2012 by Gluvocio

[pro]

Nah mate not owned, I just wanted to have a play with some older hardware for a change, heaps of fun

 

I was hoping to play with sempron 1 core as cedar mill is pita but maybe some other time

[I.M.O.G.]

lol... if only there were more crocodile dundee memes

 

By the way, since I said something about it getting quiet sharing tweaks, I guess its only fair I help... This is a small one, but has been tested well and proven consistent in my benching and a handful of others.

 

Slam the mouse pointer in the top left corner when TW starts. Ensure the mouse pointer is all the way in the corner, do not move it.

 

Compare that to the rabid mouse tweak - top left has been better every time than mouse movement in my testing. Down with the rabid mouse!

 

[ATTACH]1267[/ATTACH]

[Christian Ney]

[xXSebaSXx]

ROFL...

 

Sad but true... We all thought the "Rabbid Mouse Tweak" worked; and it does to a certain extent.

 

Here's the reasoning behind why it doesn't and why getting the pointer out of the way is best.

 

- Try the TW test and leave the mouse alone right after you hit the "Start Benchmark" button... that should leave the pointer somewhere in the middle of the screen. You'll notice that the windows move around fast, but slow down when they get closer to the area where the pointer was left.

 

- Now run the test again and move the mouse like a mad man... You will now see that the windows move faster, but some of them still appear to hit a slow patch at certain points.

 

- Now run the test and move the mouse pointer to the corner so that no windows can go over it. You'll get a higher score.

 

I'm not sure, but I'm willing to bet the windows that come in contact with the pointer are the ones that get slowed down and throw your score off. And the reason the "rabbid mouse tweak" kind of worked is because by moving it like crazy the tendency is to rotate the pointer around the edges of the screen (you may not see it, but the momentum you put on the mouse when shaking it like crazy will make the pointer get closer and closer to the edges of the screen) and being far out to the edge means that it has less chance of hitting a moving window so they all move faster... You still hit one or two every once in a while and they slow down. Just move the mouse out of the way and see.

Edited May 19, 2012 by xXSebaSXx

[I.M.O.G.]

Lol!

 

Credit to Sebastian, for saving my mouse arm from repetitive stress injuries. Hopefully the Sebastian mouse tweak can help some others as well.

Edited May 19, 2012 by I.M.O.G.

[CherV]

http://www.hwbot.org/submission/2285099_cherv_superpi_core_i7_3770k_6sec_125ms?recalculate=true

 

I got a good CPU for PCMARK 05.....!! ^3^

[Gluvocio]

well ... we're talking about trasparent windows , but I see classification starting from High Text edit score.

I understood that even if you do 200k trasparent windows , it's not enough

I did some test last days... Text edit is a powerful monster to equalize.

I'm still far from your scored.

Much congrats to CherV , Pro and Vapor. who can do it

[SteveRo]

^^ What!! Five new pages on this thread and no text edit tweak yet ...

 

ahhhhhhhhhhhh!!!

Edited May 20, 2012 by SteveRo

[Gluvocio]

Cherv distract us explaining trasparent windows ....

 

ahahahaha... I joke ! You did a nice explaination, CherV ... I think it will be useful to spread this benchmark

[zeneffect]

ROFL...

 

Sad but true... We all thought the "Rabbid Mouse Tweak" worked; and it does to a certain extent.

 

Here's the reasoning behind why it doesn't and why getting the pointer out of the way is best.

 

- Try the TW test and leave the mouse alone right after you hit the "Start Benchmark" button... that should leave the pointer somewhere in the middle of the screen. You'll notice that the windows move around fast, but slow down when they get closer to the area where the pointer was left.

 

- Now run the test again and move the mouse like a mad man... You will now see that the windows move faster, but some of them still appear to hit a slow patch at certain points.

 

- Now run the test and move the mouse pointer to the corner so that no windows can go over it. You'll get a higher score.

 

I'm not sure, but I'm willing to bet the windows that come in contact with the pointer are the ones that get slowed down and throw your score off. And the reason the "rabbid mouse tweak" kind of worked is because by moving it like crazy the tendency is to rotate the pointer around the edges of the screen (you may not see it, but the momentum you put on the mouse when shaking it like crazy will make the pointer get closer and closer to the edges of the screen) and being far out to the edge means that it has less chance of hitting a moving window so they all move faster... You still hit one or two every once in a while and they slow down. Just move the mouse out of the way and see.

 

hey thats 3rd lazyman way bottom left = best results.

[knopflerbruce]

Just wondering... I'm testing an areca 1882ix w/4gb cache on some old crap setup, but it feels like it's not making any use of the cache... thought i'd have to cripple XP startup even on old k8 singlecores, but I haven't been past 150 yet, usually stuck at ~135-140. First run is usually lower, about 80 or so (that's with 4 m4's, using just one i get the same score during the 2nd run, but 1st run is like 35 (!) ). Any of you gurus know what's up?

[I.M.O.G.]

Write back cache enabled in controller BIOS options when you created your volumes on the card? If you have write thru enabled, perf will be much worse.

 

With 3 or more maxiops, on that controller I got 800+ XP Startup. I think sometimes it was over 1000 by quite a bit.

 

Xp startup would always be huge, on any number of cores - it wasn't sensitive to platform, the way virus can changes depending on number of cores and power. You can pm me if you want to talk settings. There is definitely something wrong with there, but if it isn't write back cache, there are a lot of other config options. Probably something when you created the volumes/array. I have used this exact card and know it well.

Edited May 21, 2012 by I.M.O.G.

[CherV]

@GLUVICO

 

If Pro agreed to disclosed PRO's Text Edit tweak, will you disclosed your memory latency tweak ?

[pro]

nah thats cool, i can work out memory latency and im sure gluvicio can work out text edit

[CherV]

For Text editing test , Futuremark use the WordPad application provided with the current Windows installation and perform search & replace operations on the document which is an 85 Kilobyte text.

 

WordPad consists of two main parts, the WordPad.exe program itself and its Wordpad engine which is actually part of the Windows operating system.

 

FMPad.exe is using Wordpad engine which Microsoft has steadily enhanced the WordPad engine over the years while making few corresponding enhancements to the WordPad program.

 

to boost the performance, you need to know the serect of Wordpad engine.

 

Hint :

1. Disable Rich Text Format support in Wordpad engine

2. Disable Unicode support in Wordpad engine

3. Force Match whole word Search and Match case Search in

4. Disable Error Level output in Wordpad engine

 

No DLL Modify required.

[Antinomy]

Thanks for sharing, CherV!

 

The most I've got through was finding reference of MS Sans Serif in FMpad.exe and deleting it from the OS. Now half OS goes with no text at my work. Might even not reboot so it's two weeks powered-up.

Anyway I wanted to reinstall it.

[CherV]

Thanks for sharing, CherV!

 

The most I've got through was finding reference of MS Sans Serif in FMpad.exe and deleting it from the OS. Now half OS goes with no text at my work. Might even not reboot so it's two weeks powered-up.

Anyway I wanted to reinstall it.

 

you're welcome.

[xXSebaSXx]

well ... we're talking about trasparent windows , but I see classification starting from High Text edit score.

I understood that even if you do 200k trasparent windows , it's not enough

I did some test last days... Text edit is a powerful monster to equalize.

I'm still far from your scored.

Much congrats to CherV , Pro and Vapor. who can do it

 

It all has to do with the math behind how the final score is calculated...

 

I'll explain:

 

- PCM05 score is calculated by taking the Geometric Mean of the 16 sub test scores and multiplying that number by 87.

So a final PCM05 score would look like this:

PCMarks = Geomean(st1, st2, st3, ... ,st16) * 87

The Geomean function is the nth root of the product of n scores... So for PCM05 the final score would be expressed as follows:

 

PCMarks = 87 * (st1 * st2 * st3 * ... * st16)^(1/16)

 

And if we apply mathematical properties of exponents we can re-write it as

 

PCMarks = 87 * st1^(1/16) * st2^(1/16) * st3^(1/16) * ... * st16^(1/16)

 

The fact that we're taking the 16th root of each and every sub-test score is what normalizes the results and "prevents" anomalies when a particular sub-test score is "too high". But there's a "tweak" here as well. The nature of PCM05 is that sub-test scores range from the 9s (memory latency) to the 100Ks (Transparent windows). Taking the 16th root of any of the sub-test scores will reduce that score to a number slightly greater than 1... And now that we are getting close to 100K mark on TW test; it will even go a bit past 2.

 

Example:

 

Memory Latency Score: 9 --> 9^(1/16) = 1.14720269. I like to think of the difference between the number above and 1 as the weight of that particular test's score towards the final PCM05 score (or how much it will affect it).

 

The reason TW has gone from being the most important test to now being second to Text Edit is the following.

 

A TW score done by someone like me (I know how to get 60K on it, but I haven't figured out the 100K tweak yet) when compared to the 100K scores that Pro and CherV can get would look like this:

 

60000^(1/16) = 1.98899849 vs 100000^(1/16) = 2.05352503

 

1 - (1.98899489/2.05352503) = 0.0314223294 --> ~3.14%

 

Or that an improvement from 60K to 100K in Transparent Windows (67% Improvement on sub-test score) would only give me a 3.15% increase in the effect that particular score has on the final score.

 

 

But if we look at the Text Edit sub-test:

 

A normal score for this particular test (without the new tweak) is in the 300 to 400 rate. See Gluvocio's score here. He has the highest 4x score that is not using the Text Edit tweak and his score on it is 355.38. If we compare that score to Pro's #1 score for 4x CPU here; his Text Edit score is 36836.75.

 

355.38^(1/16) = 1.44350506

36386.75^(1/16) = 1.92778659

 

That's a little bit more than a 1000% improvement on the sub-test score, but we don't care about those... We want to see what the change in "weighted" score is.

 

1 - (1.92778659/1.44350506) = 0.251211173 --> ~25.12%

And this is what makes the Text Edit test the most important nowadays... Going from "non tweaked" to "tweaked" would give you a 25% improvement on your final score. We can see that if we compare Gluvocio's final score (64K) vs Pro's final score (87K) and it does come quite close to that number... The actual difference between those two scores is ~35%, but that's due to Pro also beating Gluvocio in other sub-tests and not just on Text Edit.

 

Ok... I hope this helps some of you better understand the ins and outs of this benchmark.

 

Seebs

[pro]

yeah and thats exactly what im seeing,

 

once you get to 90k+ trans windows, it really does NOTHING for your score, but, when you start raiseing cpu frequency, you get massive gains for every 100mhz because that trans window is waiting there to catapult your score forward when the other subtests increase