DHEP Challenge

[Skillz]

Does TAAT plan to sign up for this?

https://boincstats.com/en/stats/challenge/team/chat/1060

 

[crashtech]

Hmm, I'm game if everyone else is.

 

[StefanR5R]

It's a Tetrathlon.*

                             Th Fr Sa Su Mo Tu We Th
                             23 24 25 26 27 28 29 30
      Formula Boinc Sprint   __ ▄▄ ▄▄ ▄▄
  Project "X" marathon leg          ▄ __ __
         PrimeGrid TRP-LLR      ▄▄ ▄▄ ▄▄ ▄▄ ▄▄ ▄▄ ▄▄
       DHEP Island Hopping   __ __ __ ▄▄ ▄▄ ▄▄ ▄▄ ▄▄

Edit:
Formula Boinc Sprint:

• project announcement on Wednesday, May 22, 22:00 UTC
• return results after Thursday, May 23, 22:00 UTC and before Sunday, May 26, 21:59 UTC

PrimeGrid TRP-LLR:

• begin downloading tasks after Friday, May 24, 00:00 UTC
• return results before Thursday, May 30, 24:00 UTC

DHEP Island Hopping:

• report results from Sunday, May 26, 00:00 UTC to Thursday, May 30, 24:00 UTC
• the client reports results to the DHEP project every 15 minutes, given network access
• the DHEP project gives credit every hour

________
*) without a naughty shoutbox, but also without a daily bulletin. :-(

Edit 2, May 22: fixed FB Sprint start time

 

[crashtech]

I thought that at least the FB Sprint ended just as DHEP was beginning? But the overlap with PG might make it a no go for me, did not initially realize that.

 

[Skillz]

Isn't PG usually GPU?

Edit
Nevermind, I see that it's a CPU subproject.

 

[StefanR5R]

I thought that at least the FB Sprint ended just as DHEP was beginning?

I added the precise times now in my post above. (Somebody shout if I made any mistakes.)

But the overlap with PG might make it a no go for me, did not initially realize that.

I am currently thinking: I have run TRP-LLR, and more so LLR in general, before. I have not run DHEP yet. I am beginning to get curious.

PS,
if we want to dip a toe into the Island Hopping challenge, TACommittee needs to sign us up.

Edit:
Apparently, it is possible to bunker DHEP to some unknown extent. I edited post #3 once more.

 

[biodoc]

Sure I'm game. The sprint will be a priority for me though.

 

[crashtech]

Won't DHEP run okay on old stuff? I could put some Sandy and Ivy cores on it, and even 8 Harpertown cores, lol. I keep that one around because it's the only board left with a floppy controller. Though only about 1 out of 10 floppies that I have are still readable...

 

[StefanR5R]

Won't DHEP run okay on old stuff? I could put some Sandy and Ivy cores on it, and even 8 Harpertown cores

There are quite a few Ivy Bridge-EP hosts among those with top RAC.
If you are not power constrained and have both old hosts and new hosts, you could run DHEP and PG in parallel on the respective hardware, indeed. I on the other hand am power constrained and have only new CPUs. (OK, I still have the Merom, the Phenom II, and the Ivy-Bridge-E. But they are still sitting in moving boxes, as I am first and foremost power constrained, or rather, power bill constrained.)

 

[crashtech]

Well, I'm more restrained than constrained, and probably ought to be more restrained instead of less. I tell myself that some of this hardware is only for "Sprints," but that definition has been, hmm, a bit overly elastic for me.

 

[Ken g6]

if we want to dip a toe into the Island Hopping challenge, TACommittee needs to sign us up.

I'm on the committee, but I find I have a conflict of interest here. I guess people want me to sign us up anyway? Edit: OK, fine, we're signed up.

Project "X" marathon leg

What's that?

 

[StefanR5R]

What's that?

It's where people are fishing for the last few Cod from hopelessly overexploited stocks.

 

[Howdy]

Well I made an account and am running this on 1 machine to see what happens. So far it runs cool and seems to not require much memory. Actually a very interesting project IMO......

 

[Assimilator1]

Yea DHEP ran much cooler than SETI, Asteroids or LHC on my i7 4930k (Ivybridge E).

So we're meant to run PG & DHEP in parallel?
I'll pass on running PG , can I still join in just running DHEP?

 

[Ken g6]

So we're meant to run PG & DHEP in parallel?

Probably not. Edit: Well, not on the same machine, anyway. Hence my "conflict of interest".

I'll pass on running PG , can I still join in just running DHEP?

Go for it!

 

[StefanR5R]

Yea DHEP ran much cooler than SETI, Asteroids or LHC on my i7 4930k (Ivybridge E).

So we're meant to run PG & DHEP in parallel?

Probably not. Edit: Well, not on the same machine, anyway. Hence my "conflict of interest".

I wonder what sort of code DHEP's application is; including, why it runs cooler.

Is it accessing RAM a lot? In that case it would not go well together with PG's LLR based application, which itself accesses RAM extensively if there is not sufficient cache. (And even if the CPU has enough cache for one or a few LLR processes, if there is an additional load which wildly jumps around in RAM, that load may cause the CPU to evict hot data of the LLR processes from the caches, I am afraid.)

Or is DHEP simply just branch-heavy code, instead of data intensive? In that case, a certain amount of DHEP could perhaps be put onto those HyperThreads which are of little to no use anyway when LLR is doing its thing.

So far it runs cool and seems to not require much memory.

If it doesn't allocate much memory, then perhaps it also doesn't access memory very intensively.

Of course, instead of idly wondering, I could as well simply test it with the very same method with which I tested for LLR throughput in the past (back then without any concurrent loads). But not today anymore.

@Assimilator1, as noted by @crashtech, PG's LLR based projects love Haswell and later, and conversely don't shine on Ivy Bridge and former. Hence, the Ivy Bridge-E should have more of a momentum at DHEP.

(Though I for one will probably run neither during the Formula Boinc sprint.)

 

[Howdy]

Well I actually spoke too soon on the memory usage. Apparently at the start DHEP didn't use much, started @ ~38% usage and has went up from there to 93%. It's in "waves" so to speak, peaks @ 93% and backs off to the 38% usage- rinse and repeat- 16gb of RAM on a "Skeletor" type machine. Nothing stating "waiting for memory" from what I have observed.

 

[StefanR5R]

If I wanted to take measurements over e.g. three or more of such cycles, would that mean an hour, or several hours, or a day of test duration, roughly?

 

[Howdy]

It looks to me on this machine it was over a 30-45 min timeline. I paused it to see if I could time it.

CORRECTION- 15 minute timeline- This is over a 19hr total run time. I restarted the machine and it has since made a dive to 27% ever so slowly. I should have screen shot it before the restart...............

OBSERVATION:
So I started DHEP on my DD (Win10), less cores with more memory (24gb/ was 32gb 1stick FUBAR'd vs 16gb) available compared to the other machine (Linux Mint). DD does not exhibit the same memory "wave" as the Linux machine. So is it possible it backs off when the memory is at a certain remaining usable percentage? (Linux machine)

@StefanR5R

I believe I figured it out possibly (amazing if I did) BUT according to the web site
"Remember the DHE BOINC app connects to our servers every 15 minutes to upload its progress. There is little value in having a percentage completion figure for a task with a non-defined processing time. "
Thus the memory "wave" I am seeing??

 

[ao_ika_red]

Yea DHEP ran much cooler than SETI, Asteroids or LHC on my i7 4930k (Ivybridge E).

So we're meant to run PG & DHEP in parallel?
I'll pass on running PG , can I still join in just running DHEP?

I'm running DHEP right now and I can tell that its CPU utilisation is quite lower than other BOINC project. That probably explains why it runs cooler.

 

[crashtech]

I wonder if this is a project where a bit of over-subscription can be beneficial.

 

[Assimilator1]

Over subscription?

I've noticed that DHEP uses quite a bit of RAM, although I don't recall how much. However when I play Elite Dangerous I normally back off BOINC to 50% (6 threads), and that normally works fine (excluding with VR). But with DHEP I was getting really bad stuttering even with BOINC completely paused, turned out RAM usage was the issue! Not just BOINC & ED, but DHEP was a major factor.

Anyway, I've stopped new task d/ls & will be switching to DHEP soon.

Btw, are we signed up for this race now?

 

[crashtech]

Over subscription?...

Yeah, like if you have 8 logical cpus like an i7, use cc_config.xml to say you have more, something like this:

<cc_config>
     <options>       
       <allow_multiple_clients>1</allow_multiple_clients>
       <ncpus>12</ncpus>
       <skip_cpu_benchmarks>1</skip_cpu_benchmarks>
     </options>
</cc_config>

Or maybe just 9, or 10. Whatever gives the best throughput or points.

 

[StefanR5R]

Btw, are signed up for this race now?

We are. But first things first: In two hours from now, we should learn about the Formula Boinc sprint project that'll be on on Thursday to Sunday. Don't miss to tune in to Orange Kid's FB sprint thread in time. :-)

 

[TennesseeTony]

Bump.

But mostly I just wanted to post something off topic that is not worthy of a thread.

I got my ticket! Get yours!



The details are fuzzy, but I think they are sending 'a chip' that perhaps an electron scanning microscope is needed to read the names engraved upon it.