• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Folding Benchmarks

H

Haitch

Limp Gawd
Joined
Mar 8, 2011
Messages
383
As you may know, I've been working with pjkenned to expand on Musky's benchmark tool. The idea is that we take those captured work units, run them for a given number of frames, then upload them to a publicly accessible MySQL instance where anyone can browse/download/crunch the data thats there.

After a couple of false starts (what works great on an i7-860 didn't work properly on an i7-2600k - what worked greeat on those mis-reported on a dual Opteron .... ) I think it's at a point where more people can test it - I need to know what works, what breaks, when and how it breaks.

The benchmark is currently Windows only, but once I have the quirks ironed out of that, I'll be porting it to Linux - specifically Ubuntu - but can look to other distros as well.

Once the benchmark versions are built, I'm then looking to build a "FAHLog watcher" - rather than upload data just on benchmark units, parse the FAHlog.txt file, and everytime you finish a WU, the same WU & System data gets uploaded, for every WU you process.

The info I'm currently gathering is:

FAH Info:

RunDate, UserName, BenchLive, WU, NumFrames, MaxTPF, MinTPF, AverageTPF, BenchVer, FAHVer, FAHArgs

CPU Info:

CPUVendor,CPUCoreName,CPUCodeName,Freq,CPUSpec,CPU Multiplier, CPUFSB, Stepping, InstructionSets, CPUPackage,CPUID,eCPUID,NumProcs,NumCores, Numthreads, L1DataNum, L1DataSize, L1InstNum, L1InstSize, L2Num,L2Size,L3Num,L3Size,

Memory Info:

Memory, MemFreq, Numdimms, MemType, MemChannels, CASLatency, RAS2CASDelay, RASPrecharge, CycleTime, RowRefreshTime, CommandRate

Motherboard:

MoboVendor,MoboModel

OS Info:

Platform,Scheduler,Kernel,Architecture,FileSystem

Result Verification

Authenticity Hash

If there any suggestions of additional information that would be useful, desirable to have if possible, I'm open to ideas. For request to access the beta, PM pjkenned either here or at http://forums.servethehome.com

H.
 
Looking forward to this...

Not to load you up with too much for the first version, but what are your thoughts on handling data from non dedicated machines?

ie - someone who folds 24/7, but uses machine for websurfing etc during the day, resulting in a distorted average.

Also, can it cope with stopping & starting? - HFM often logs partial frames making the minimum frame times pretty useless.
 
Mibw,

There's not a lot I can do about non-dedicated machines. I scrape the raw TPF values, report the min and max, toss them, and average the rest. With bench app I only have access to the raw data for that machine. I can build statistical representations from that data from the DB though- eg for an i7-860 at 3.6 GHz the mean TPF for WU x is, with a STD. Variation of y seconds - meaning that at that freq, 90% of machines will do a tpf of between Y and X with Z% confidence. The more data points we have - the shorter the gap between X and Y become, and the higher the value of Z.

H.
 
Last edited:
I was thinking about that. I am no statistics dude, but I imagine tossing the upper half or third might also be reasonable insurance against partially used machines.
 
I don't want to get away from dedicated folding to bench right now, but I'd be very interesting when you get to the point of parsing legit logfiles. Very cool project, best of luck!
 
MIBW - My Stats classes were many year ago, so doing the analysis isn't my forte, but more data points from independent machines makes for better data than more analysis of a single machine.

Lord_Of_Doors - yeah - running benchies on my machines is killing my PPD, but if I can pull all your live data with 0% impact on your folding, and produce real data - its a huge opportunity

H.
 
hmm, next time i see pjkenned on IRC ill ask him for it and give you guys some data points for a 1090T @3.9 , 2600k will have to wait till you get a linux client :D
 
Haitch can hook you up also. I won't be on IRC for a day or two. Today was a 7:30am to 11pm work day and this week will be the same.

On the scrubbing, usually I would advise taking the top/ bottom quartile out, but non-dedicated folders may be an issue. The bigger threat isn't the box that plays games 6 hours a day, since that is easy to spot. Instead it is the one that watches a youtube video or two and does a few other miscellaneous things making it 10% slower so it will not get weeded out. That being said, in this, one may want to throw out the top 5% and the bottom 45% since, in theory, the higher the ppd the more likely it is a dedicated folder.
 
Could you not just add a tickbox of some sort to the machine information for "Dedicated" or "Not Dedicated", and filter the results that way?
 
I could, but is it really meaningful? If its a non-dedicated box, how non-dedicated was it while the bench was running? For the constant monitoring it would be somewhat more meaningful; but the question of how non-dedicated would remain.

H.
 
The goal is to basically have a script people can run that they do not need to self-report information. Even "dedicated" could mean something different as someone could have a stock PC that did not get a clean OS re-installation. In that scenario it is not impossible to lose 10% performance from pre-installed software (virus scanners and etc.)
 
I have zero dedicated boxes, so dedicated PPD results aren't indicative of the PPD I get.
 
I will help you out as soon as you get to linux... and I have a wide variety of systems to test on...

but dedicated windows ppd is not what I am interested in lol...
 
Patriot said:
I will help you out as soon as you get to linux... and I have a wide variety of systems to test on...
Click to expand...
Me too but please kept it script based with no GUI interface as I don't have X11 installed on any of my seven folders. :cool:
 
One other thing we need to do is figure out the display. Once we think of what we want I may either pick up a PHP/ MySQL book or find someone on oDesk or something.
 
So there's no way to see results at this point?
Posted via Mobile Device
 
Not a lot of results in the database to see. You can always pull from the MySQL DB which is easy to do. I posted the information you'd need on the beta forums.
 
Last edited:
Haitch said:
MIBW - My Stats classes were many year ago, so doing the analysis isn't my forte, but more data points from independent machines makes for better data than more analysis of a single machine.
Click to expand...
Hey that sounds kinda like how F@H works :p
 
pjkenned said:
Not a lot of results in the database to see. You can always pull from the MySQL DB which is easy to do. I posted the information you'd need on the beta forums.
Click to expand...

yeah... I even have some code that gives a php table output from mysql queries...
I could sanitize it and post it....
 
may i be worthy said:
Looking forward to this...

Not to load you up with too much for the first version, but what are your thoughts on handling data from non dedicated machines?

ie - someone who folds 24/7, but uses machine for websurfing etc during the day, resulting in a distorted average.

Also, can it cope with stopping & starting? - HFM often logs partial frames making the minimum frame times pretty useless.
Click to expand...

I'll add std. deviation to the min/max/average. A low std dev indicates that the tpf were very consistent, the more varaiability, due to being used, the higher the std dev.

For the live scraper, I'll only be using WUs that were folded continuously, if the client is stopped and then restarted, the WU will not be reported. Should have the Windows version of this ready tommorow.

H.
 
Have the Windows BenchMark and Live version running - testing the live Monitor now. If anyone else would like to test it, email me from an address that can take a .exe/.zip. (email to me at: hamish@haitch.net) The monitor runs independently of the fah client itself, so if it it crashes, it won't impact your folding. CPU utilization of the monitor is basically 0 - it runs a very intense CPU load on startup, then goes quiet - on my test machine it's almost 8 minutes of CPU time during startup, 0 seconds in the four hours since.

Have the bench compiled under Linux - but it needs some more work. With Linux it may actually be easier to do the live scraper than the benchmark app. :rolleyes:

H.
 
Is there a Linux console only version of this? Last I checked, the Linux version required X windows and I believe Wine as well.

I'd love to setup my boxen to populate the database, but I'm running Linux with a console only--no gui.
 
Working on it. Getting it console only and correctly clocking Sandy Bridge's in Linux are the big things at the moment.

H.
 
What are you doing on the forums?! You have a tool to code!

Does that help? ;)
 
firedfly said:
What are you doing on the forums?! You have a tool to code!

Does that help? ;)
Click to expand...

LOL! Yep.

Actually, I have to set up the Q6600 to be me file server - more the raid array over, then rebuild the SB as native linux so I can actually work on the code, Project for this weekend.

H.
 
If it helps you, take a look at i7z, it reads SB temps/speeds just fine


Haitch said:
Working on it. Getting it console only and correctly clocking Sandy Bridge's in Linux are the big things at the moment.

H.
Click to expand...
 
You must log in or register to reply here.
Back
Top