FB 2017 strategy to overtake Gridcoin

Yeah, but at 50M PPD that's only 20 days. 50M PPD is easily possible. phoenicis and I were doing close to 90M before the heat shut me down. The team was right around 100M PPD. We did 2B+ points in about 2 months.
 
Hopefully weather will be cooperating when fall starts next month. I can power up one 1080 and one 970 if we want to take #1 spot in Collatz. By then we should be at number spot for most of the low hanging fruit projects.

I updated my first post to reflect our current status. If we take #1 in yoyo, dBM and Numberfield, we will gain + 27 points which will then match Gridcoin current tally of 1034 (1007 + 27). We can then pick up a few projects which will require more crunching power to take the #1 spot. Suggestions?
 
You’re right, it’s getting difficult to choose cpu targets following our sterling progress on the low hanging fruit projects. Ramblings on the remaining active cpu projects from my perspective are:
  • Asteroids – Effectively CPU only as it’s not a particularly good return on gpu investment. Achieving 1st place will be difficult as we initially struggled during the sprint just to exceed the leaders ppd for a short period.
  • Climate Prediction – not a particularly popular project with incredibly long task run times.
  • Gerasim – OK with short run times but low points and quite a gap between us and 1st place.
  • Leiden – sign up limitations mean that some team members would have to use the HardOCPtest account to contribute. Seems to significantly favour Linux over Windows with respect to run times and credit accrual (I’m currently Windows only).
  • LHC – Multithreaded applications are not particularly multiprocessor friendly. Big gap to number 1 handicapped by the CreditNew point award death spiral.
  • Rosetta – big gap to number 2 with little chance of achieving 1st place given that we failed to bag gold in the sprint even with fastgeek’s support. Very much favours intel over amd which is bad for me as my cpus are now all amd.
  • WCG – our recent move to 2nd place gives me a great excuse to move to a favoured project but it may be unnecessary as we will probably retake 1st place during the competitions later in the year.
So, in summary, I really don’t have a bloody clue, lol. I guess I’d be OK crunching any of these. The preference would be that whatever we do grind, we do it en masse as a team so as to prevent any boredom setting in.
 
Last edited:
Skillz said:
See that MN Scout ? One 1080 Ti can almost do 10M PPD on Collatz.
Click to expand...

I ran Collatz a while back. I think it was for one of our sprints. Have 20 million points there. Would have been with a 1080.

Tried adding my Nvidia 710 into my system. My $20 backup card. I like this notice on one of the MilkyWay results:

Estimated Nvidia GPU GFLOP/s: 61 SP GFLOP/s, 8 DP FLOP/s
Warning: Bizarrely low flops (8). Defaulting to 10

I couldn't figure out how to have 3 tasks going on my main card, and only 1 on this sucky card. I know it is pointless, but was curious what would happen. Obviously running 3 tasks at once isn't the best on the 710. Took 48 minutes instead of 4 on the 1080 :p (running 3 tasks at once on both cards).
 
Numberfields needs help. We lost bunch of FB points on it recently.
 
Skillz said:
See that MN Scout ? One 1080 Ti can almost do 10M PPD on Collatz.
Click to expand...

Did one task right now, 260 seconds for 27,900 points. Looks like 9million a day. Maybe my optimizations are not the same as yours. Below is what I'm running from sometime in the past when I helped out.

collatz_sieve_1.21_windows_x86_64__opencl_nvidia_gpu.config
verbose=1
kernels_per_reduction=48
sleep=1
threads=9
lut_size=17
reduce_CPU=0
sieve_size=30
cache_sieve=1

app_config.xml
<app_config>
<app>
<name>collatz_sieve</name>
<gpu_versions>
<gpu_usage>1</gpu_usage>
<cpu_usage>0.5</cpu_usage>
</gpu_versions>
</app>
</app_config>
 
You're pretty much optimised. I use similar configs for Ti except lut_size is set to 18 and I run 2 tasks per gpu to sqeeze out another half million or so ppd. So settings are:

verbose=1
kernels_per_reduction=48
sleep=1
threads=9
lut_size=18
reduce_CPU=0
sieve_size=30
cache_sieve=1

and

<app_config>
<app>
<name>collatz_sieve</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.25</cpu_usage>
</gpu_versions>
</app>
</app_config>
 
OK that is what I just found out in testing.
Lut of 19 appears to triple the time. lut of 18 may shave off 10 seconds, but I don't have enough samples. Upping the kernels to the max 64 doesn't appear to do anything noticeable. Threads at 11 errored out the work units, 10 appears OK, but no benefit detected. Sieve size of 31 errored.
 
Lol, yes, I remember trying out various settings with a similar experience before I settled on the above.
 
phoenicis said:
Lol, yes, I remember trying out various settings with a similar experience before I settled on the above.
Click to expand...

Found out threads 10, actually slows it down. Appears consistently by 10+ seconds. I'm going with the settings you listed.

Think I'll average out right around 250 seconds.
 
There are two files the settings need to be placed in. The files only appear when an associated task is downloaded. The settings do not kick in for the currently running task but rather when the following task begins. They are:

collatz_sieve_1.21_windows_x86_64__opencl_nvidia_gpu.config

and

collatz_sieve_1.21_windows_intelx86__opencl_nvidia_gpu.config

One thing to bear in mind is that they will be overwritten whenever a new version of the app causes a re-download of the files.
 
Skillz said:
Numberfields needs help. We lost bunch of FB points on it recently.
Click to expand...
Numberfields is one of these low hanging fruit projects. I'll crunch this once we hit our targets for yoyo and dDM soon. CV and phoenicis are helping out on these two. Thanks guys.
 
yeah, dDM ... soon #1

Screen Shot 2017-08-28 at 23.14.10.png
didn't expected we show up as rank #3, #5 and #7 (pututu) ; but hey, no complains. I wish we would have more chances in WCG or Rosetta instead. But for that we need fastgeek ...
 
Last edited:
ChristianVirtual said:
more chances in WCG
Click to expand...
I think consensus here is that we will go all out during the end of the year with Xmas/birthday challenge. That's for CPU.

For GPU, it seems Collatz has more interest? I'm sure these two will keep us very busy till the end of the year.

What about Leiden Classical? Anyone want to run this? I would go with team consensus like a wolf pack;)
 
pututu said:
I think consensus here is that we will go all out during the end of the year with Xmas/birthday challenge. That's for CPU.

For GPU, it seems Collatz has more interest? I'm sure these two will keep us very busy till the end of the year.
Click to expand...
switched to Collatz, Milkyway and Primegrid; just hope that end of year we get enough CPU power to catch up ... sentimental I am on bio/health projects. too much attention on math-stuff and space-stuff

I hope the next sprint is GPUGrid, ah, that would feel good ...
 
I have a few borgs randomly helping on Leiden, but keep in mind that they limit the number of work units per day per USER account. So, even if multiple people want to hop on the HardOCPtest account, it will be limited.

Leiden Classical allows a maximum of 16 WUs per request and a maximum of 128 WUs per day per user. 10/29/15 via PM from admin
Click to expand...

https://hardforum.com/threads/all-inclusive-dc-project-list.1801908/#post-1040955850

The admin still has not replied to my inquiry about how to find the weak account key...
 
GPUGrid has a terrible time with having enough work units. I certainly hope that it is NOT chosen for a sprint. It would piss a lot of people off.
 
Viewing the sprint history, I suspect the next one will be a cpu one so hopefully we should be safe from keyboard rage for a couple of weeks.

Gilthanis said:
I have a few borgs randomly helping on Leiden, but keep in mind that they limit the number of work units per day per USER account. So, even if multiple people want to hop on the HardOCPtest account, it will be limited.
Click to expand...

After the sprint and LHFruit are taken care of, it sounds like we may be able to have our cake and eat it. I've already got a Leiden account and, at the same time, I fancy a bit of WCG for a while. I've not done any badge hunting all year!

So, aside from sprint, order of cpu attack as follows?

dDM/Yoyo
Numberfields
Leiden/WCG

Any thoughts?
 
I've got some borgs helping with YoYo badge hunting right now as well. In your case phoenicis 3 of those 5 are all badge projects. Sounds like you will have the opportunity to raise your bling. :jimlad:
 
It seems they've increased the max tasks per day on Leiden to 1024. I'm not sure whether it's per account or computer.
 
That is awesome news. Is that what you are seeing via your account or was that posted somewhere?
 
There's posts on their forum complaining that 1024 is just not enough.
 
phoenicis said:
So, aside from sprint, order of cpu attack as follows?

dDM/Yoyo
Numberfields
Click to expand...

So here is my summary:

If my math is correct, if we get top spot in these three LHF projects, we will get +21 points which will bring us on par with GC with a total of 1013 + 21 = 1034 points.

Then, we need to defend all other positions and perhaps take top spot in our favorite projects like WCG? Leiden would be an easy one if we can get 1024 WUs per day. Need to build some buffer.

So I vote WCG since we have more interest in this project.

What about GPU? I vote Collatz since PG and MW is a bit harder to get top spot unless consensus says otherwise.

I think this is the strategy that we all have to do for the rest of the year.

If we don't act like a wolf pack, it will be harder to get top spot in the non-LHF projects.

I think we still have time to decide after we match the #1 spot with GC.
 
I think PG is doable, but it would take a few GPUs with optimizations dedicated to it for awhile. I was calculating out a potential for 1.5+ million ppd with PPS Sieve tasks running 2up per gpu on my Nvidia 1080TI. All the other tasks are lower scoring by 1/3 or much more. The next worthy PG ppd project is the Genefer 22 tasks but they take a day to calculate per unit, and I saw approximately 500k ppd for them. I never got to fully run it alone for a day to confirm the PPS Sieve numbers, but the handful of tasks I did run calculated in that high direction.

I'm going to run out the Collatz and MilkyWay tasks on my computer then run PG for awhile. I want to see that high number, before helping with whatever other GPU project the team decides to focus on.

(Grrr... why do I have MW cpu tasks... I disabled that.... must have snuck in between setting changes. What a pointless waste of power there when the GPU is so much faster at them. Ohhh I enable them for the sprint.:playful: Time to abort some tasks.)
 
Last edited:
pututu said:
So I vote WCG since we have more interest in this project.

What about GPU? I vote Collatz since PG and MW is a bit harder to get top spot unless consensus says otherwise.
Click to expand...

Sounds good. After LHF projects, defend + wolf pack WCG and Collatz then move on eg PG and Leiden.
 
pututu said:
Collatz as explained in earlier post will give you a high number like 10M PPD per 1080Ti. PG can't beat that.
Click to expand...

Oh I know that. I'm trying to get to the max from each project for a day out of my equipment. So far I haven't ran Prime Grid for a full day. Been jumping between too many sprints and challenges.

Edit: I'm good with focusing on Collatz and WCG for a good bit to get to number one. I'd move my GPU over to Collatz after Labor Day and leave it there. WCG already crunching.
 
Last edited:
Breaking news, my Nvidia 710 is delivering an eye watering 644ppd for the team on Asteroids!:wideyed: So crazy. I put the lock down, no cpu, no 1080 for you Asteroids. :p
 
Well guys, just remember my boxen is offline during the heat. Once it cools back off and I don't need to run my AC anymore I will be back full tilt. I only run one 4P off and on through the week. It's cool(ish) outside right now so I've left it running after the recent sprint. My GPUs are offline exep the rigs that I leave running for other reason (HTPC, NAS, main rig)

Priller seems to be traveling a lot for work lately, but once we get him back he is big into WCG so he'll be able to add at least 1 2P rig to that, possibly 2x 2P rigs.

I have no idea what happened to fastgeek. He hasn't been around much since he last posted about his stuff going offline for a couple weeks then asking about gridcoin and the other cryptomining DC teams. lol
 
That's fine buddy, anything you are comfortable throwing at the targets would be great.

Having just sampled the heat you guys are experiencing over there, I can quite understand why AC costs are outrageous right now.

I'm looking forward to temps cooling down enough for you to fire up the full farm again.
 
MN Scout said:
I couldn't figure out how to have 3 tasks going on my main card, and only 1 on this sucky card. I know it is pointless, but was curious what would happen. Obviously running 3 tasks at once isn't the best on the 710. Took 48 minutes instead of 4 on the 1080 :p (running 3 tasks at once on both cards).
Click to expand...

MN Scout , BOINC does not let you get THAT advanced with the mixing of cards. However, if you were supporting 2 GPU projects at the same time, you could exclude one card from project A and let card 2 work on project A & B or just B. That way if you had one card that could perform tons better by running multiple work units, you could still maximize it and run the other one maximized on a different project.

I had a complex setup once with 4 or so different nVidia cards in it and some of them were as old as the 8400 GS all the way up to 750Ti in generations....
8400 GS, GT430, GT650, and 750TI IIRC
 
Gilthanis said:
MN Scout , BOINC does not let you get THAT advanced with the mixing of cards. However, if you were supporting 2 GPU projects at the same time, you could exclude one card from project A and let card 2 work on project A & B or just B. That way if you had one card that could perform tons better by running multiple work units, you could still maximize it and run the other one maximized on a different project.

I had a complex setup once with 4 or so different nVidia cards in it and some of them were as old as the 8400 GS all the way up to 750Ti in generations....
8400 GS, GT430, GT650, and 750TI IIRC
Click to expand...

Yup, I excluded each GPU project I didn't want to run on the 710. Would really be nice if they had an easy interface for configuring these things right in the application.

Milkyway actually doesn't run all that bad on that little 710 card. 3 tasks in 45 minutes is actually OK sort of. Compared to the 5 hours a task for Asteroids, which is only 50% slower than a CPU task, so really not all that bad. Practically no points though, but that is a low scoring project. Asteroids kind of feels like a complete waste to run on either my CPU or the good GPU, so that is why I have it running on the itzy bitzy teeny weeny GPUeenie.:wacky:

Here is my cc_config.xml that I have now:
<cc_config>
<log_flags>
</log_flags>
<options>
<use_all_gpus>1</use_all_gpus>
<client_download_url>http://www.worldcommunitygrid.org/reg/ms/viewDownloadAgain.do</client_download_url>
<client_version_check_url>http://www.worldcommunitygrid.org/download_all.php?xml=1</client_version_check_url>
<network_test_url>http://www.ibm.com</network_test_url>
<exclude_gpu>
<url>https://boinc.thesonntags.com/collatz/</url>
<device_num>1</device_num>
</exclude_gpu>
<exclude_gpu>
<url>http://www.primegrid.com/</url>
<device_num>1</device_num>
</exclude_gpu>
<exclude_gpu>
<url>http://milkyway.cs.rpi.edu/milkyway/</url>
<device_num>1</device_num>
</exclude_gpu>
<exclude_gpu>
<url>http://asteroidsathome.net/boinc/</url>
<device_num>0</device_num>
</exclude_gpu>
</options>
</cc_config>
 
Last edited:
You must log in or register to reply here.
Back
Top