13th WCG Birthday Challenge

[Terminator_1]

Dear members of Team ANANDTECH,

For Nov 16th to Nov 22th 2017 SETI.Germany is arranging a challenge at World Community Grid on the occasion of the grids 13th birthday:

We would like to invite ANANDTECH to participate in the challenge and hope that your team and many members of your team will be joining and participting this event.
!!Be aware, only your WCG-Team-Captain can join your team to the challenge!!
https://secure.worldcommunitygrid.org/team/challenge/viewTeamChallenge.do?challengeId=8617

More information and statistics regarding the challenge can be found here.http://www.seti-germany.de/wcg/1_en_Welcome.html


Your participation is very much appreciated.

Many Greetings: Terminator

 

[petrusbroder]

TeAm AnandTech has registered,
As usually, we are in!

 

[StefanR5R]

Thanks @petrusbroder. Looking forward to the event.

 

[TennesseeTony]

What is the start time, the countdown clock is off by a month or so...

 

[StefanR5R]

UTC start and end times taken from the seti.germany forum thread.

Start: November 16, 00:00 UTC

countdown
convert time and date into local time
Sydney, Australia: Thursday, November 16, 11:00 AEDT (UTC+11)
Jakarta, Indonesia: Thursday, November 16, 07:00 WIB (UTC+7)
Stockholm, Sweden: Thursday, November 16, 01:00 CET (UTC+1)
New York, NY, USA: Wednesday, November 15, 19:00 EST (UTC-5)
New Orleans, LA, USA: Wednesday, November 15, 18:00 CST (UTC-6)
Denver, CO, USA: Wednesday, November 15, 17:00 MST (UTC-7)
Portland, OR, USA: Wednesday, November 15, 16:00 PST (UTC-8)​

End: November 22, 23:59 UTC

countdown
convert time and date into local time
Sydney, Australia: Thursday, November 23, 10:59 AEDT (UTC+11)
Jakarta, Indonesia: Thursday, November 23, 06:59 WIB (UTC+7)
Stockholm, Sweden: Thursday, November 23, 00:59 CET (UTC+1)
New York, NY, USA: Wednesday, November 22, 18:59 EST (UTC-5)
New Orleans, LA, USA: Wednesday, November 22, 17:59 CST (UTC-6)
Denver, CO, USA: Wednesday, November 22, 16:59 MST (UTC-7)
Portland, OR, USA: Wednesday, November 22, 15:59 PST (UTC-8)​

--------
Here are some technical data about work units from the various WCG subprojects:
https://www.seti-germany.de/wcg/6_en_The+WorkUnits.html

I am currently measuring a few of the subprojects, notably credits/day, and will post results here later this week. But I can only use a Linux PC for these measurements. Windows results would differ to some extent, depending on the subproject.

If you want to use a PC in this contest on which you haven't crunched WCG before, then it's beneficial to run WCG on it now and get a dozen or so tasks validated. After that, the project server will lift some per-host limits; notably validation of most WUs will then occur immediately. Exceptions are the projects MCM and FAAH1, which are quorum=2 projects, i.e. require two copies of a task to be computed independently on different hosts, and validate = get credit only after both results come in and are equal.

For optimum performance it is usually best to run only a single subproject at a time on the same host, not a mix of several subprojects. Exception: I read that MIP (Microbiome Immunity Project) is memory bandwidth intensive and also occupies more RAM compared with the other subprojects, and therefore its performance scales sub-linearly to the number of processor threads loaded with MIP. A machine fully loaded with MIP alone may therefore not perform to its full potential, depending on the particular hardware. - Edit, November 7: MIP scales linearly from 14 to 22 cores per socket for me on hosts with same memory bandwidth per socket. I.e. MIP's performance is not memory bandwidth limited in my observation.

 

[crashtech]

Is there information available on which, if any, WCG projects utilize AVX?

 

[TennesseeTony]

Run times would be beneficial info as well, and do I remember correctly they have a 1000 task limit per host?

 

[IEC]

Tagging @Markfw, since you probably have data regarding which projects run best on Ryzen/TR. I've got MIP, MCM, and SCC tasks on my boxes for the time being.

 

[Markfw]

I just run all the sub-projects, so I don't have times, sorry, isn't there a screen somewhere that shows runtime per completed task for WCG ? I think I have seen it once, but I can't find it now. As many boxes as I have running I am sure you could look and see which one does best on what. (10-12 boxes)

 

[StefanR5R]

Is there information available on which, if any, WCG projects utilize AVX?

At least SCC, OZ, OET, and MCM do not utilize AVX (or at least not to the extent that would clock down my Broadwell-EPs to their AVX turbo; they run this at their non-AVX turbo).

I didn't look very closely, but I did not notice a remarkable performance difference between Broadwell-EP and IvyBridge-E. (I can't measure my IvyBridge-E properly at the moment because it is not available for pure WCG workload at this time. Besides, this IVB-E runs Windows, and I have only Linux PCs available for comparisons at the moment. The OS could skew the results of some subprojects.)

Edit: Maybe I get an opportunity to measure at least one subproject also on BDW-EP/Windows and IVB-E/Windows on the coming weekend, for direct comparison with my BDW-EP/Linux baselines.

Run times would be beneficial info as well, and do I remember correctly they have a 1000 task limit per host?

Run times (and deadlines) are shown in the seti.ger link which I posted. I will include my own observations of run times when I report what I am measuring currently.

The limit of 1000 tasks in progress is set by the boinc-client itself, independent of projects. WCG has some more limits; I'll try to recall what they were.

isn't there a screen somewhere that shows runtime per completed task for WCG ?

On the web page, go to "My Contribution" -> "Results Status" -> then filter the results per machine, per project, and/ or per status (e.g. status = valid, to see both run time and credit).

To compute PPD, I copy and paste these tables directly into a spreadsheet (Excel or LibreOffice Calc). I didn't figure out how to split the "x.xx / y.yy" formatted columns into separate spreadsheet columns, but copy-and-paste it to a text editor which has suitable replacement functions, and then copy-paste into the spreadsheet.

 

[Markfw]

OK, is there any easy way to get like all the results, instead of one page at a time ? To put into Excel ?

 

[StefanR5R]

Not that I know of.

Each page has 15 results (except the last page which may have fewer). So you can calculate the total number of results of a given filter from 15 * (number of pages - 1) + number of results on last page.

If I want to calculate average durations or average PPD, I simply take a sample from e.g. 2 pages = 30 results. I haven't verified whether this is enough. But 1 page = 15 results may indeed give imprecise averages.

 

[Markfw]

Well, for my threadripper for mapping cancer markers, it looks like about 3.05 hours for 114 points. Sound right ? But I think that is per task, and the threadripper is doing 29 tasks at a time. Or about 26,000 ppd for that computer for that task (most are that task)

For my 28 thread E5-2683, it looks like about 5.5 hours for 27 tasks at once for 13,431 ppd.

Mt 1800X is doing 14 threads @3.25 hours or 11,785 ppd. Almost up to the 28 thread Xeons, not bad

<My 3930K at 4.5 hours for 11 threads is doing 6688 ppd

Should I keep going ?

I added everything up, and yesterday I did 136k, and it should have been 147k (after estimating the dual 6276s' at 11,297 ppd) so my estimates are a little high, but probably close.

(=26000*2+13431*4+11795*2+6688+11297)

Oh, but thats assuming every task is mapping cancer markers, which its not.. Maybe I will try one computer for all tasks, and see what I can find.

 

[crashtech]

@StefanR5R , thanks for the reply. I ask because E5-2680v2 can be had for $180 right now, and I was thinking about getting several to replace two 2660v2 and one i7-3820 CPU. I think this would give a good performance increase, but as the work moves towards utilizing AVX, the Ivy-E gets nearer obsolescence. Haswell-E still commands a price premium at this time, however.

 

[Markfw]

@StefanR5R , thanks for the reply. I ask because E5-2680v2 can be had for $180 right now, and I was thinking about getting several to replace two 2660v2 and one i7-3820 CPU. I think this would give a good performance increase, but as the work moves towards utilizing AVX, the Ivy-E gets nearer obsolescence. Haswell-E still commands a price premium at this time, however.

I see thats a 10-core socket 2011 3.6 ghz ? is that turbo ? That would make a great upgrade for my 3930K 3.2 ghz (3.5 turbo) for $180

 

[StefanR5R]

E5-2680v2 can be had for $180 right now, and I was thinking about getting several to replace two 2660v2 and one i7-3820 CPU. I think this would give a good performance increase, but as the work moves towards utilizing AVX, the Ivy-E gets nearer obsolescence. Haswell-E still commands a price premium at this time, however.

A move to Haswell-E and -EP would also require different boards and RAM, and RAM is costly now and in the foreseeable time.

I see thats a 10-core socket 2011 3.6 ghz ? is that turbo ? That would make a great upgrade for my 3930K 3.2 ghz (3.5 turbo) for $180

3.6 GHz is the single-core turbo which is virtually never reached.
Base clock is 2.8 GHz:
https://ark.intel.com/products/75277/Intel-Xeon-Processor-E5-2680-v2-25M-Cache-2_80-GHz
All-core turbo is 3.1 GHz:
https://www.intel.com/content/dam/w...cification-updates/xeon-e5-v2-spec-update.pdf

 

[Markfw]

A move to Haswell-E and -EP would also require different boards and RAM, and RAM is costly now and in the foreseeable time.


3.6 GHz is the single-core turbo which is virtually never reached.
Base clock is 2.8 GHz:
https://ark.intel.com/products/75277/Intel-Xeon-Processor-E5-2680-v2-25M-Cache-2_80-GHz
All-core turbo is 3.1 GHz:
https://www.intel.com/content/dam/w...cification-updates/xeon-e5-v2-spec-update.pdf

So 3.1 ghz@10 cores vs 3.5 ghz@6 cores... Hmm still might be a 50% improvement with no other costs for $180 (mhz@cores is 21 vs 31)

 

[crashtech]

I see thats a 10-core socket 2011 3.6 ghz ? is that turbo ? That would make a great upgrade for my 3930K 3.2 ghz (3.5 turbo) for $180

Yes, they are a great value as long as the motherboard will accept them.

 

[Markfw]

Yes, they are a great value as long as the motherboard will accept them.

I can't find anything that says that my motherboard (ASRock X79 Extreme7)will take ANY Xeon, but I am sure it will.

What do you think ?

 

[StefanR5R]

Index into the community-maintained WCG FAQ:

https://www.worldcommunitygrid.org/forums/wcg/viewthread_thread,21501​

FAQ on the results status page:

https://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=6105#86841
Results get cleared out of this table 4 days after validation.​

FAQ on limits of tasks in progress:

https://www.worldcommunitygrid.org/forums/wcg/viewthread_thread,28636
New clients get only 5 tasks per logical CPU, until they complete (and validate?) their first task.
Then the quota is raised to 80 tasks per logical CPU and 2560 per client. (The latter limit is ineffective because the client limits itself to 1000 runnable tasks.)​

--------

@Markfw, http://www.asrock.com/mb/Intel/X79 Extreme7/#CPU

 

[crashtech]

I can't find anything that says that my motherboard (ASRock X79 Extreme7)will take ANY Xeon, but I am sure it will.

What do you think ?

http://www.asrock.com/mb/Intel/X79 Extreme7/#CPU

 

[Markfw]

http://www.asrock.com/mb/Intel/X79 Extreme7/#CPU

You make me feel like an idiot sometimes... I was there and did not see that page. So it does support it, and now I am getting one !

Thanks !!!!!!!!!!!!!!!!! :beer::beer::beer::beer::beer::beer::beer:

 

[StefanR5R]

@crashtech, @Markfw, if you want to use your socket 2011 hosts primarily for GPU computing, then stick with i7 for higher single-threaded performance. (Wait, where am I, in the I don't need another system thread, or WCG b-day thread?)

 

[crashtech]

You make me feel like an idiot sometimes... I was there and did not see that page. So it does support it, and now I am getting one !

Thanks !!!!!!!!!!!!!!!!!

Sorry, Mark! I have been in the same boat as far as missing something like that. I'm just glad that your board supports just about every LGA2011 CPU ever made! It's a nice one.

 

[Markfw]

@crashtech, @Markfw, if you want to use your socket 2011 hosts primarily for GPU computing, then stick with i7 for higher single-threaded performance. (Wait, where am I, in the I don't need another system thread, or WCG b-day thread?)

I just got a second threadripper, but for 4 MORE cores for $180. I am going for it !

I just got one for $178, and updated my bios already, I can't wait !!