Processor performance matrix

[Duster]

how are you guys getting your info out of console clients?

Any way we can get this in excel? if not i will be glad to put it in should be able to produce some good graphs and play more with the numbers

I will try and bring back some info when I go check on my comps tomorrow

 

[rodsfree]

nnnnngggggg!!! I guess I should have expected this kind of thing. Overclocked processors, processors under partial load, etc.

I'm working to include the latest information...unfortunately, I can't upload it from work, so you'll all have to wait until I get home tonight.

This is where EMIII would shine!

He's got a little speed utility that you run in your F@H folder and EMIII will use that actual measured speed in it's reports.

Too bad he's too busy, and I'm a user not a coder.

u_m - do you think that Larry would let somebody else work on his source code?

 

[unhappy_mage]

He's got a little speed utility that you run in your F@H folder and EMIII will use that actual measured speed in it's reports.
...
u_m - do you think that Larry would let somebody else work on his source code?

1) Why not just use that value along with the core type as your hash value?

2) No. If he hasn't open-sourced the project by now, I doubt he'll do it now. And I don't know enough VB to do something like this.

 

[Duster]

1) Why not just use that value along with the core type as your hash value?

2) No. If he hasn't open-sourced the project by now, I doubt he'll do it now. And I don't know enough VB to do something like this.

bah be a man and write it in C

 

[gwai lo]

I only have stats for my two home folders. The other one my Mom uses sometimes..but for the most part it's just folding with nothing running (killed off all unneeded processes including explorer). My rig usually folds without anything running except AIM. Both run QMDs.

Currently:

Processor - P4 3.0C @ 3.6
Project - p1912
Frames - 100
Points - 450
Time Per Frame - 15:59 (bare minimum processes)

Processor - Celeron D 2.4 GHz
Project - p1911
Frames - 100
Points - 450
Time Per Frame - 27:53

Hope I got this right...

Don't have info on school rigs..but they're stupidly bloated and all the frame times are way off from what they should be.

 

[Mohonri]

UPDATE: Latest version up, including all entries in before roughly 4:00 pm today, with the exception of OSUguy's Ginormous excel file.

If you want to put it into Excel, just highlight the whole table and copy/paste it into Excel. I don't play games with my HTML tables, so it should be pretty straightforward.

 

[sandmanx]

UPDATE: Latest version up, including all entries in before roughly 4:00 pm today, with the exception of OSUguy's Ginormous excel file.

If you want to put it into Excel, just highlight the whole table and copy/paste it into Excel. I don't play games with my HTML tables, so it should be pretty straightforward.

I just noticed you are listing cores as well, let me add mine:

The P3 850, and both the P3 1GHz listed are Cumine cores.

The Athlon XP 2400+'s listed are Tbred(I think they are all teh b variety above 1.8GHz, but all cpu-z reports is Throughbred)

The AXP 2600+@2.017 is a Barton.
The AXP 2700+@1.91 is a Barton, and it should be a 2600+. My bad, I get these damn performance rating mixed all up when they're non windows boxen.

 

[rodsfree]

1) Why not just use that value along with the core type as your hash value?

2) No. If he hasn't open-sourced the project by now, I doubt he'll do it now. And I don't know enough VB to do something like this.

Using it as a hash would work....
Larry's speed.exe utility creates a file called "speed.emd" which is just a test file.
Here is the contents of one of mine.

Intel: 3054 Mhz

I don't think it does the cores though. But it would probably be easy for a master coder like you to come up with a replacement for Larry's speed.exe that would do cores and save the data to the same type of speed.emd file that EMIII could read and use for the html code.

That's it.

Since it is just a text file it would be easy to merge it with something else.

Actually, I just checked EMIII options and it will create everything that we want, in an html format, on a scheduled basis - but it doesn't slug the data with your username.

Here is everything that it'll output on that html page - and it is a running history.

Row
Points 
Atoms 
Protein 
(Project)Run/Clone/Gen[Server] 
Time/Frame 
Points/Day 
Total Time 
Date 
Time 
Box Name 
CPU Info 
Frames 
Protein Image

If you put your user name into the box names then you could get around that.

Other than that - just need an upload location and some way to combine and sort the data, eliminate duplicates and such.

EMIII will even run other programs on a scheduled basis.....so if you had a command line ftp utility(suggestions?) it could just upload the EMIII stats file to a server location where all the rest of the data mining could happen.
Using a cmd script we could even have it run multiple programs like the ftp and Larry's speed.exe to keep up with the actual speed of the computer as it changes - or put that into windows scheduler.

Bingo, you'd have everything we want and more.

 

[Mohonri]

Sorry for the lack of updates. Our house was the host for all the meals over the weekend, and we had lots of family in town. I'm currently working on adding OSUguy's data (man, that spreadsheet's enormous!), and I've added a few core names. It won't be uploaded until tonight (uptight proxy at work).

One of the advantages of having data for lots of projects is that I can see what projects can be easily grouped together. For example, I'm beginning to see that projects 243-246 all give about the same ppd, even though they have different point values per WU. I am giving serious thought to combining groups of projects with similar ppd and consecutive (or near-consecutive) numbers in an effort to reduce the size of the spreadsheet.

In addition, once I have three data points for a certain processor on a given project, I'm not going to bother calculating more ppd for that CPU/WU combination. For example, I have tons of data for a Sempron 2200 on the 1700-1709 series, so I'm not going to bother more with it. If you have something that's significantly different than the numbers posted (>10% difference), please post your numbers.

In the meantime, keep posting your ppd/time-per-frame numbers so I can add them to the matrix. Thanks!

 

[OSUguy98]

One thing I noticed looking at my numbers awhile back, occasionally, EMIII will give some unreasonable completion time... like this one from my spreadsheet:

Row	Points	Atoms	Protein	(Project)Run/Clone/Gen[Server]	Time/Frame	Total Time	Date	Time	Box Name	CPU Info	Frames	Pts/Hour	Pts/Day	Pts/Week	Pts/Year
2508	600	9130	p1141_RIBO_FSpeptide_HEL_nospring	(1141) 106/19/10 [156]	21:47:54	2179:50:00	08/26/05	2:26:21 AM	Athlon 1700+ #2	Athlon 1700+ #2	100	0.28	6.61	46.24	2,412.85

So are you throwing out the outliers such as this?


Keep on Folding!! For the [H]orde!!

 

[Mohonri]

So are you throwing out the outliers such as this?

Heh. Yeah. EM3 does that to me every once in a while, but it's always my fault. Yes, I'm screening.

 

[Mohonri]

Alright! The latest is up. Big thanks to OSUguy for the massive amount of data, of which I used probably less than 15%. You can tell which processors are his...

Don't pay any attention to the P4 2.4C running at 3GHz. It's running two instances (HT), and has had some configuration changes, so it's numbers are unreliable.

Hopefully sometime soon I can start adding OCAU's stats.

 

[ElektronikSeraph]

Got some more info. Lots of p2305.

Athlon XP 2500+ (Barton) - XPPSP2
p2305 46pts - 8:33/frame; 8:27/frame

Athlon XP 2200+ @ 2.0GHz (Thoroughbred-B) - XPPSP2
p2305 46pts - 8:09/frame
p1477 364pts - 32:07/frame
p1141 600pts - 47:25/frame (measured twice on two separate crunches of this project)

Pentium M 1.6GHz (Dothan) - XPPSP2
p2305 46pts - 8:33/frame, 8:32/frame, 8:41/frame (all measured on separate crunches of the project)

'Preciate the time you're putting into this, although it would definately be cool to setup some sort of web database... create a username/password, then add your CPUs, then login every couple days, select your CPU, select the project, and input the time...

 

[molson3530]

Here are some of my WUs:
Protein Time Per Frame Points
p1155 0:04:22 241
p1149 0:30:02 600
p1159 0:04:35 241
p1140 0:30:42 600
p755 0:04:42 41
p754 0:06:23 51
p1406 0:35:38 266
p1141 0:30:05 600

All on an Athlon 64 3500 @2310 Venice core

 

[OSUguy98]

bump for you guys that haven't added your info (you know who you are!)


Keep on Folding!! For the [H]orde!!

 

[mavalpha]

P1152_L939_K12M_ext_f
11min/frame
Banias P-M 1500
239pts

I'll try to add the rest next week- all of my P4's are school computers, the only ones at home are a P-M and a C-M, both 1.5GHz. I'm also still struggling to get a PCIe video card for my Venice, and I think I may be able to recruit another 3200 to mirror mine. That's not including a few mid-to-high-end s370 procs that I've been toying with, although they're insignificant by comparison.

 

[SidewinderX]

Pentium M 2.0Ghz (533mhz FSB) 1GB RAM

p2305, 46 pts.

1min 54s per frame

 

[Slayten]

P1150_RIBO_Semihelixfrom 1141

21m19sec/frame 600 points

AMD XP3000+ (Barton)@2.17Ghz
1024Mb Ram

 

[Mohonri]

Results from this thread have been included, although I haven't gotten around to adding all the info in this thread. Adding projects to the html isn't easy when openoffice has some weird regular expression quirks in its find/replace dialog, and when I niglected to install the python component, so I can't record macros. I've installed python now, but it's too late, so I'm going to bed. Hopefully I'll get the new projects into the table soon.

 

[Mohonri]

I've added a bunch of data from OCAU's thread on this topic (it's in blue in the matrix), and you can find the latest version here. I've pulled the overclocked data into a separate file here.

Next steps:
1) Start weeding out more projects that have little data in them.
2) Write a java applet that will allow me to easily add/remove/show/hide processors and projects, and data points, and will write my html for me. I don't like the jaw-droppingly bloated html files that Word or Writer generate. The file is already almost 100k as it is.
3) Combine projects 243-246 into one column--the score for them seems to be the same for each processor.
4) Include the rest of the data from OCAU. There isn't too much left, but it's still a "TODO:"

A couple other notes:
-On the subject of outliers, I'm not going to include them. The idea behind the matrix is to give a good general comparison, so "typical" run-times will take precedence. The fact that a p2063 is taking 60 hours on my P-M 1.4GHz (<50ppd at that rate) with -forceasm means that it's one data point that won't make it in the matrix.
-The Aussies like to use two instances when HyperThreading is available. This messes things up, so I'm not going to include datapoints from HT processors. At this point, I'm not sure what to do about Dual-core chips.
-I need suggestions for color coding. I'm thinking about something like this:
---Blue means I grabbed it from OCAU, Red means the [H]orde.
---As I get more data points, the color will progress toward white. Three agreeing data points is when a cell will become white.
---If I only have one data point, and it seems to be an outlier or otherwise weird, it'll be either grey or yellow.

Any other suggestions are welcome.

 

[sandmanx]

A small update:

The P3 1.0GHz in the overclocked table is a Cumine core. It's a 100MHz FSB with a 10x multplier running on a 124MHz bus to get to 1.24GHz for the record.

The latest on it:

p2305 - 100 frames - 16:32 per frame - 46 points - 40ppd

I'm really sick of these small units!

Edit: A form to submit data for those already established might make things a bit easier to submit more WUs in the future.

 

[Mohonri]

Edit: A form to submit data for those already established might make things a bit easier to submit more WUs in the future.

I agree. I just need to figure out how to set up Apache with Tomcat or something similar and write some servlets. I wrote a java-based webapp a few years ago, but wasn't involved in any of the software setup, so it would be a learning experience. With a steep curve. Unfortunately, it's not a high priority at the moment, so it'll probably have to wait.

 

[OSUguy98]

are you going to monitor which projects are active? sometimes, a project will be off of the psummary for a couple weeks, and then make a return....

a form to enter would be nice.... but could introduce error..... is it easier to get the number directly from the output from EMIII? Just email you the myproteins.html and have a script translate that to whatever you wished?

I'd like it duely noted that I have ideas, some of which may be impractical and stupid, so fell free to shoot them down


Keep on Folding!! For the [H]orde!!

 

[Mohonri]

are you going to monitor which projects are active? sometimes, a project will be off of the psummary for a couple weeks, and then make a return....

I noticed that exactly this happened not too long ago with some low-numbered projects. I'm keeping an eye on it...occasionally.

a form to enter would be nice.... but could introduce error..... is it easier to get the number directly from the output from EMIII? Just email you the myproteins.html and have a script translate that to whatever you wished?

I don't have any sort of script yet. It's all just manual data entry. Tedious, yes. Maybe I should set up a script where people can enter data, but it doesn't go onto the main matrix until I approve it...that goes back to my above post about learning how to use Apache, Tomcat (or an equivalent), etc. At this point, I would probably not use a database (mysql), considering that the amount of data is relatively small.

I'd like it duely noted that I have ideas, some of which may be impractical and stupid, so fell free to shoot them down

Hey, I appreciate the feedback. Good stuff.

 

[Mohonri]

I've finished adding the OCAU data (not a whole lot of it useful), and I've combined projects 243-246 into one column. I had an idea occur to me today. Maybe I should just do a few columns--one for Tinkers, one for gromacs, one for big gromacs (600 pointers & the like), one for each of the other cores. Fewer columns = easier to compare. What do you think?

 

[OSUguy98]

Would it be difficult to Have a tinker/gromac/amber/etc column that basically gave a range.... and then click on it to open another page that showed the individual projects that are tinkers/gromacs/etc? (Is that even possible? too much work/hassle? did you follow that? lol)

One thing you might want to add is some contact info for times like this morning when the forum is down..... and a date of when you last updated it?


Keep on Folding!! For the [H]orde!!

 

[Mohonri]

Would it be difficult to Have a tinker/gromac/amber/etc column that basically gave a range.... and then click on it to open another page that showed the individual projects that are tinkers/gromacs/etc? (Is that even possible? too much work/hassle? did you follow that? lol)

That's exactly what I was thinking. Is it too much hassle? It might not be too hard, if I get off my behind and write some code. I seriously need to write an application that will let me easily edit the contents of the matrix--it'll make life much easier.

One thing you might want to add is some contact info for times like this morning when the forum is down..... and a date of when you last updated it?

"Last updated"--great idea. I'll just put it at the top left corner of the table. As for downed forums...since this data isn't particularly time-sensitive, I don't know that I need people emailing me their data points. Performance for a given processor on a given WU isn't going to change if the forums are down for a while.

 

[DF454]

Protein 1168 p2360_Coiled_Coil
1294 Points
AMD Barton 2500 small OC to 1925Mhz (basicly running at 2600 speed)
100 Frames
Time/Frame 2:15:14
Time/WU 225:23:20
W2K

 

[darkstar7]

All of these processes were run on an Athlon64 X2 4400+ @ stock speeds (2.2)(Toledo core)

The Graphic Process, were run using the Graphic version and there was only 1 running
The # Console Process, were run in the Console version and the # refers to which process 2 running total, 1 per core

Project,Frames,Credit,Process,Avg Time/Frame,PPD
2304,100, 48, Graphic,0:02:50,243
2305,100, 46, Graphic,0:06:14,106
985,100, 23, Graphic,0:01:02,316
2071,100,166,1 Console,0:11:06,215
2072,100,206,1 Console,0:15:06,196
2304,100, 48,1 Console,0:02:54,238
1809,100,153,2 Console,0:11:11,196
2303,100,218,2 Console,0:22:15,140
2304,100, 48,2 Console,0:02:50,243
[ahh the joys of csv]

 

[Mohonri]

So I've finally started writing an app that will let me (relatively) easily add/remove/show/hide processors and projects. It's java-based, since that's the language with which I have the most recent history, and because it's cross platform, so when unhappy_mage or King_N see what a mess it is, they can happily make it into a usable application and happily run it on their Linux boxen.

Gee, I'm rusty. It's been way too long since I programmed--I had to go back and look up how to declare an array, yet somehow remembered this:

BufferedReader inputFile = new BufferedReader(new FileReader(inputFileName));

I remember how much that bothered me my freshman year--having to type System.out.println("blah"); instead of cout << "blah" << endl;
...
ahh, the memories
...
ok, back to work.

 

[Codegen]

'nother one.

Pentium 4 1.8ghz Williamette

P2071

166 pts

37 mins 47 secs.

heh heh heh, sex.

 

[gnewbury]

Isnip
1) What processor you're running (Type, core, speed)

2a/3a) Instead of 2 and 3, just post the ppd for your processor on that specific WU.

Running 2 instances:
AMD X2 3800+, Manchester Core - 2Gb RAM
instance 1
p1481 - 325ppd
p770 - 93ppd
instance2
p2051 - 101ppd
p2353 - 253 ppd

Seems like a pretty wide point range.

 

[Mohonri]

Seems like a pretty wide point range.

I'm still trying to decide what exactly I want to do about dual-core processors, for this very reason--the two instances affect each other because the two instances share the memory controller/bandwidth, even if they have their own processor core. Here are two options I've thought of:

1) use only total ppd at any given time for both instances. Unfortunately, this number will also vary widely, since running two bigpackets at one time will perform worse than a bigpackets + a tinker, running two tinkers will give some different ppd, and things will be again all over the place.

2) Only use ppd numbers when the same project (or series of projects) is running on both cores. For example, running two Tinkers, or two 600-pointers, or two QMDs.

what do you guys suggest?

 

[gnewbury]

Just a suggestion -
How about recording benchmark result, as reported in FAHlog.txt?
for instance, cpu, speed, br, and present EMIII points per day for protein:
compaq 2.8Ghz - br 7176 - 175ppd - p1912
AMD Barton 2500 (2.2Ghz?)- br 6892 - 129ppd - p2365
AMD X2 3800 (2Ghz) - br 6636 - 98ppd - p762 (did about 350ppd on a 1481)
AMD X2 3800 (2Ghz) - br 6636 - 128ppd - p768
AMD Applebred 1.8 - br 5968 - 136ppd - p1481
and I just noticed the benchmarking is hosed on Linux, because I've got an
AMD 2400 (1.66Ghz ) - br 672 - 112ppd, p1159

/edit - I just noticed that the proteins I'm getting on my X2 3800 are deadlineless - even though I've specified big packets and advmethods.
WTF is going on?

 

[OSUguy98]

I vote #1.... I think it would get messy to do option 2.... you'd have to have people keeping track of what WUs were done with what WUs...... with #1, you could just say "expect to get anywhere from _____ to ______"

Would it be better to do pph rather than ppd? I guess that comes down to what people can better understand (have a concept of/used to thinking).... EMIII now gives pph, ppd and ppw, so it seems like it's just a matter of personal preference....

Do the console and graphical client differ enough that the results need to be seperated?


Keep on Folding!! For the [H]orde!!

 

[Mohonri]

Do the console and graphical client differ enough that the results need to be seperated?

I don't believe so.

Believe it or not, I'm not sitting on my hands. I've been working on writing my app (it's been a long time since I programmed, so I'm making all the noobie mistakes) at night when I can't sleep, and the hard part is done. Once I'm done with it, I'll be more actively adding new data points, and more importantly, new projects.

 

[OSUguy98]

No hurry... take your time.... I just made some improvements to my spreadsheetl, I found a thread in the programming forum that pulled prices directly from a Newegg URL...... tweaked it to work for the shipping and now, nice and easy updates To the point that I'm considering adding prices from another site (like ZZF).... make an if statement and post the lesser of the two......


Keep on Folding!! For the [H]orde!!

 

[OSUguy98]

So how go things with your matrix?


Keep on Folding!! For the [H]orde!!

 

[Mohonri]

Sorry for the lack of updates. Been out of town and busy with other stuff. Now that I'm on vacation, I should have time to finish my app up. In the meantime, I'll add the info I have (it's a fair amount). Thanks for the boot in the behind!

 

[OSUguy98]

not trying to kick you in the behind, so much as wondering how things were going.... I get sidetracked alot... so when I think of something, I check on it.... and in the case of threads like this, bump it to keep interest/ideas/etc/etc/etc... I should have an update posted for my spreadsheet tonight/tomorrow.... I'm working on it right now, but I can't post it until I'm at home.... I'm still trying to figure out how best to incorporate your grid into my PPY column, or my prices into your grid....


Keep on Folding!! For the [H]orde!!