9th Annual BOINC Pentathlon 2018 Strategy & Preparation

Yes, rosetta will consume A LOT of disk space. I can't run it on one of my 64 core rigs because they little 120GB HDD w/ ~50GB of free space gets filled so quickly and becomes unstable and it's a PITA to get it back in working order so I am leaving it on NF for now.

Anything you can bunker on Rosetta would help us. Not much we can do about Universe (as far as taking over another position), but we need to prevent ourselves from falling any more ranks so if it's running Universe I say leave it. Speaking of, I just looked at the 64 core rig to see what it was running and it needs to be running Universe so I am switching it as I type this.
 
Yeah i agree it seems likely China is going to take 3 in swim (don't count out those crazy frenchmen though, they got a univercity ffs). This means we need to do well on sprint or city run. I finally got both systems running both .
I used the host file method and confirmed this works well:

I added :
127.0.0.1 asteroidsathome.net
127.0.0.1 boinc.bakerlab.org

This seems to be working on both my rigs that are running asteroids. I couldn't seem to download much more than ~300 work units between the 3 GPU so i am prepping virtual box to run more when i finish those tasks in a day or two. That is unless we get our real GPU project and is so i will switch to that for GPU. I am hoping that will all that China is putting into swim that they haven't been bunkering in other projects too much....
 
Awesome! Just be prepared to suspend the project once we get a GPU project announced.

Also, gonna be pretty sure its gonna be limited by the PCIe bandwidth.
 
Well even if its slow you'll be okay. Its a bunker and you'll most like still complete a bunch of tasks for a decent bunker drop unless the GPU project is announced tomorrow or the day after.
 
ChristianVirtual, is the D07 (and D03, D02) should be the sum of the output from the past 7 (3 or 2) days? My D07 is only 501K.

We have 25 contributors for Universe. Fingers crossed that we can keep the 5th spot.
upload_2018-5-9_21-16-53.png
 
In less than 1 day 19hrs to the end of the race, we need to maintain that slight edge over SG in Universe, barring any big bunker drop by SG. We have being able to do this over the past 24hrs.

Keep it up. I still have 2 rigs running. One rig is expected to finish all the tasks by tomorrow lunch. The other rig will probably continue till the end of this race.

If you decided to download new tasks, there is a good chance that the validation rate is going to be low compare to older tasks that were downloaded much earlier.

So far my validation rate has improved.:)
State: All (3373) · In progress (328) · Validation pending (399) · Validation inconclusive (0) · Valid (2319) · Invalid (0) · Error (327)




upload_2018-5-9_22-7-27.png
 
Suspended NF on all systems that are connected to BoincStats and reattached Rosetta. Per the R tasks page, it says I have 9,817 WUs downloaded for 18 (I think) systems. No idea what the limit is. Was working remotely, so briefly logged into my proxy system and killed the connection; will figure out the best way to deal with that tomorrow. Soon as the Sprint is over, I'll flip NF back on.

As for Universe, here's what I have going on.

State: All (6743) · In progress (2341) · Validation pending (1496) · Validation inconclusive (0) · Valid (2903) · Invalid (0) · Error (3)

BTW, am thinking Prime Grid is best left to our GPU heavy members?
 
Kinda worried ATP might have issues with primegrid. my 1080ti is averaging 20% bus usage. really feels like they are screwing us GPU guys over in this race by only giving us 5 days of a dedicated gpu project out of a 2 week contest. And i had just finished loading up Asteroids tasks. its almost like the projects chosen and the order released were a big FU. "here we will tease you with a suboptimal project you can load up on and then just when you are getting going we will give you a proper project. Oh by the way we noticed you have miners on your team so we hand picked this one for ya"

I hope i'm wrong and he can run it properly because kicking but in GPU sprint was kind of an ace in the hole.....

Edit.. might be ok since i have a 1070 on X4 PCIE 2.0 and its hovering around 35-40% usage.
 
Team China briefly took third place by passing in universe / swimming, which put us in 4rth overall, but seems the frenchmen have turned it up a notch and took it back (y)
 
Fucking Primegrid. Podium finish overall is not going to happen. Best we can do now is hope for a 3rd place marathon.
 
I posted a very large app_config I made for PrimeGrid in the open forums. That way you can set multi threading accordingly if anyone intends on running it. It is good to have even if not just so that it is ready for future events. I can't say I'm surprised due to it being in the Pentathlon a few years back. After all the whining pshoefer did in the shoutbox about how nobody wanted the events to be the same old thing each year....sure seems to be pretty close.
 
speaking of primegrid. I tried a bunch of things to block the site last night and didn't have any luck.

I tried all of these host names:
127.0.0.1 primegrid.com
127.0.0.1 prpnet.primegrid.com
127.0.0.1 prpnet.mine.nu
127.0.0.1 psp-project.de

I also tried an outbound firewall rule for IP: 25.253.170.196 ( windows firewall as i don't have direct access to my router (landlord))

None of these stopped work units from uploading. Anyone else have better luck?
 
I'm going to toss a small bunker into Rosetta and Asteroids, but probably continue most of my focus on Numberfields. That doesn't mean everyone needs to follow suit. Just where I'm going to be. I will have a couple barely capable devices maybe helping with PrimeGrid, but don't expect much from me there.
 
Perhaps i shoudn't have blown the doors off Amicable so soon. JK. I do what i want. If any of the babies want to cough up ~$1105 a month in electric to apply points to their projects, they can be my guest. :p

PCIe 5.0 specifications will be out in Q2 2019. PCIe 4.0 specifications are RTM. These projects that are lane-limited will not be lane-limited for that much longer. I suspect most folks running mining setups like mine are not going to scramble to join folding efforts, but it certainly makes these types of setups far more applicable to cross-over efforts.

In regards to PrimeGrid, i need to pull some tasks in and see what the complete times look like.

Asteriods appears fine.


Edit 1: Found a PrimeGrid NVidia app that appears to take serious advantage of GPU compute.

Proth Prime Search Sieve (PPS-Sieve)

Supported platforms:
  • Windows: 32 bit, 64 bit (+CUDA234, AMD OpenCL1)
  • Linux: 32 bit, 64 bit (+CUDA23, AMD OpenCL1)
  • Mac: 32 bit, 64 bit (+CUDA32, AMD OpenCL1,2,3 - 64 bit only)
1 Requires AMD Accelerated Parallel Processing (APP) drivers.If APP driver not available for your card, then the ATI Stream SDK is needed.
2 Minimum required BOINC client: 7.0.0.
3 Minimum required OS: Mac OS 10.7 (Darwin 11.0.0).
4 Nvidia Windows drivers 295.xx and 296.xx should not be used.

Deadline: 6 days

Recent average CPU time: 20:39:51
Recent average GPU time: 0:20:17

Top host is this fellow with a set of 1070 Ti's .. If i can hit this fellow's times, i may be golden on PrimeGrid too.


Edit2: Looks like i am matching his times (and slightly better) with 1070s. I wish our opponents the best of luck.

Edit3: PCIe bus on 2.0 x1 is ~28% for a single 1070.
 
Last edited:
Oh Primegrid. Maybe we need to get a few folks from AggieThePew...

Since we already know the 5 projects, maybe this weekend discuss final strategies and resources allocation in this forum? Or at least tell us what each of you are running here?

Currently running:
Universe - 2 rigs, 40T total (one finish by lunch time and the other by end of the event)
Rosetta - 2 rigs, 48T, bunkering
Asteroid - 1/2 rig 4T, bunkering
PG - 0
NF - 0

1/2 rig because asteroid runs on AVX and it is the temp and affecting this computer that I'm currently writing now.
 
Also, posted on PrimeGrid's news feed

Do not use Nvidia driver 397.31
It has come to my attention that Nvidia recently released a new video driver that is causing widespread problems. 397.31 seems to cause the GPU to become inaccessible to both CUDA and OpenCL programs after a while, and can only be reset by rebooting the computer. It's likely this affects all BOINC projects and not just PrimeGrid. Indeed, it probably affects all GPU programs, even those that don't run under BOINC. There's a newer, hotfix driver 397.55 which seems to correct the problem. I strongly recommend either upgrading to 397.55 or rolling back to a driver earlier than 397.31.
Click to expand...
 
Bunkering is a serious pain in the @$$, ain't it.

Edit: Kinda definitely not up for switching around configurations like this on a consistent basis.
 
Last edited:
atp1916 PPS Sieve normally benefits from 2 tasks per GPU for max points. In app_config.xml in the primegrid data folder add this block:

Code: 
<app_config>
....snip... remove me if you have no other subproject spcs
<app>
<name>pps_sr2sieve</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>
....snip... remove me if you have no other subproject spcs
</app_config>

PPS Sieve needs a full thread per task. A full core of 2 threads is even better. Watch your completion times, and make sure they aren't slowed by the CPU projects you have running next to it. In BOINC manager I set my "usage limits" to 78% for my CPU, so that some of the CPUs are kept free for PPS Sieve GPU tasks.

I'm not sure the point of bunkering with PPS Sieve. The tasks are done in a bit more than 4 minutes on my cards. PrimeGrid is a well managed project, and the work generator is tuned, and it considers how many tasks they've assigned you in the recent past, and the length of time you've taken to return the WUs. They even adjust the deadlines around per user (note they always give you full points for the WUs even if you return late.) I guess what I'm getting at, is I wonder if they'll even assign multiple days of tasks.

I'm just going to start up my cards 1 day early. Should be about 4.5 million PPD.
 
Last edited:
They sent me around 200 tasks, with 2 gpus. I could finish them in 12 hours or less. My list isn't getting any longer, even when telling BOINC to pull 2 or 3 days.

I remember that my PPS Sieve tasks needing validation used to sit around 1000, so it might be better to just start returning tons of tasks now, and get that number up there, so that those points come in during the challenge.
 
If you think that'll work go for it. I will switch over when I get home.

Also, we should probably drop universe now. We lost our position and it doesn't look like we are in bad shape of losing another. It was also only 3 overall points.
 
atp1916 said:
MN Scout Gonna try that config out now.
Click to expand...

You should see the GPU usage go up to 100%. The task times will be longer, but it'll be less than twice the length, so you'll be doing more work.

Skillz I don't know if switching now is good or not. I haven't been following the challenge closely. Maybe bunker for Asteroids for another day and switch tomorrow morning to PrimeGrid? idk.
 
Mike Goetz one of the PrimeGrid managers just increased the number of tasks per GPU allocated to 1000, after I posed the question to him on the PrimeGrid discord channel. Maybe you could bunker now if you figured out what to block..
 
I'm most likely sitting idle on asteroids by now. Will have to check when I get home.
 
MN Scout said:
Mike Goetz one of the PrimeGrid managers just increased the number of tasks per GPU allocated to 1000, after I posed the question to him on the PrimeGrid discord channel. Maybe you could bunker now if you figured out what to block..
Click to expand...

That works for me. I am going to pull out of swim and asteroids and do Rosetta and prime. Just gonna suspend network I guess to get a bit of bunker.
 
So, as a really simplistic summary (correct where wrong).

CPUs good on: Numberfields, Rosetta, (Universe), Asteroids
NVIDIA GPUs good on: Primegrid, meh on Asteroids
AMD GPUs: might as well just run Moo!

That about right?
How much are you guys leaving on Numberfields? I'm trying to decide whether to leave my 4p on it for the long haul or to switch it over to Rosetta and Asteroids.
 
AMD GPU's can be ran on PrimeGrid as well. They just don't do as well as nVidia on PPS sieve. However, they tend to outshine nVidia on the GFN work units. I have no idea which is king on the AP27 work units.

For the sake of Pentathlon...Moo! doesn't matter. For the sake of formula-boinc, Moo! is all that matters...lol
 
EXT64 put all your CPUs on Rosetta right now. Don't worry with Numberfields, we got a decent lead.

Toss the GPUs (all of them) on PrimeGrid.
 
Just an FYI. I don't have the time or desire to be bouncing all over the place right now so my contribution to Numberfields will remain the same until the marathon is over. Then I'll be switching back to WCG. At the end of the month I'll be shutting things down as I get ready for the 1,000 mile drive to my summer home in the Colorado Rockies. I plan on bringing a few computers with me to leave in CO as I plan to build some new TR rigs when I get back to TX for the winter.
 
You're good man. We dont need to drop NF 100% anyway. It may be the only medal we get.

Otherwise I'd be blowing up your phone or PM with moving your stuff. :p
 
So:
My 55 Nvidia GPUs seem to do pretty good on PrimeGrid.

Asteroids also seems normal and unimpeded by lane bandwidth constraints.

How should I allocate these fellers?
 
You must log in or register to reply here.
Back
Top