[franz]

Lol good luck!

 

[tictoc]

Lol good luck!

Slowly, but surely. Maybe 20% reported so far.

 

[tictoc]

This Pent has been a bit crazy for me. I finally finished the build on my new home server, just in time for the Pent to start. This means that I have a mess of rigs sitting around on benches to try and keep my network, file server, meida PC, and lab up and running, since I didn't want to be down for the Pent. While dealing with that mess, I have been finding some random bugs on the latest stable and rc kernels + ROCm 5.1. I might have to just roll with some headless Ubuntu 20.04 installs, to get some additional AMD GPUs up and running.

I'm currently building a new router to go along with my downsized home server, so it has been a bit of an adventure keeping as much gear up and running as possible. Hopefully I can get a few more machines up and running over the weekend. Currently I have a handful of GPUs and a few 2P systems that are not operational.

 

[franz]

The next Javelin throw was announced, 2 days from now.

@tictoc nice job getting it up and running. That is on my to do list, well it has been since I bought the house 5 years ago lol.

 

[tictoc]

I've had multiple servers running at my house for a long time, but now I am getting rid of my big 42U rack with multiple 2U/4U chasis, disc shelves, switches, and UPS's. I am moving everything into a single watercooled Threadripper + 2 GPU system, and a 12U rack for networking and power gear.

 

[franz]

Running some benchmarks during this pent. Fun fact, Numberfields GPU app sucks in windows. Half the speed of Ubuntu. RTX2060 in both rigs. Will keep the windows rig on Primegrid until that is over.

 

[franz]

Well I just woke up to silent rigs. Power went out briefly sometime last night so I have no idea how many hours were lost.

 

[bfromcolo]

Well I just woke up to silent rigs. Power went out briefly sometime last night so I have no idea how many hours were lost.

I hate it when that happens. When we were in CO we could go many months without power issues. Since moving to the PNW power it is very flaky, brief outages are common and we have had a few significant ones of a day or longer. We have a large generator that kicks in automatically after some number of seconds. But my systems riding out these power hits is interesting, I have an old Intel server (2xL5640) and my Raspberry Pi just survive anything, when the power is back they will come back, and they handle even dirty power and very brief interruptions fine. I have a 1700x running Ubuntu though that will auto-restart after power outage more than a few seconds but not the shorter ones, and Windows is useless. Not sure how the fiber gets powered, but even when I am on generator that usually works.

It will be nice to unload these Prime and Numberfields bunkers later today and rearrange some things. Running out of Universe on one system thats getting no tasks available, although the server status says there is plenty.

 

[tictoc]

It took a few tries for me to load up on more Uni tasks. It is very random. A few clients couldn't get anything no matter what, and the others just picked up work right away. 🤔

 

[franz]

Im having the same issues with Universe, Im sure other teams are as well. I will be dumping my bunkers soon and keep all but one rig on Primegrid until the scores stabilize.

 

[tictoc]

My PrimeGrid output has been much less than I would like. Having some driver/kernel issues on the 4x Radeonm VII machine, and I haven't had a chance to get the 6900XT up and running. Hopefully I can sort that out tonight, and put up a few more points for the last few days of the Cross Country.

 

[franz]

I now have zero input/output on Universe and Numberfields, so thats fun.

Some notes for Nvidia and Primegrid
RTX2060 PPS aboout 10% better PPD than WW
RTX3060 WW about 20% better PPD than PPS, RTX 3060 is a LHR gpu, but I havent seen it perform worse than the 2060 in anything other than Primegrid.

 

[bfromcolo]

I have about 1183 Universe split between 2 systems, maybe a days worth. Hopefully another bonus day comes tomorrow or I have choose if I release these to get more, assuming it will give me any.

We should know what the City Run is going to be any time now. 5 day discipline announced 3 days in advance.

 

[franz]

The sprint was announced, SR base. Not sure how Im going to split up my GPUs now.

 

[bfromcolo]

A third project with GPU support! Wasn't expecting that, not sure they have done this in the past.

 

[tictoc]

This is the first time there have been this many disciplines with GPU apps. Until a few years ago, it was actually limited to only one GPU app... The SRBase GPU app is a little nutty, so if you have multiple GPUs, the easiest way to make it play nicely is multiple clients and some cc_config edits.
The CPU app is a real cooker on my Threadrippers, since there is no AVX offset on AMD processors, and SRBase actually puts a higher heat load on my processors than Prime95 small FFTs.

 

[franz]

Yeah my experience is I can only run SRbase on my primary GPU and another project on the other GPUs. This only works if the two projects are running in the same client if I move one out then it creates issues trying to use the primary for both projects. Im just not sure I have the time to test different configs to let SRbase bunker while trying to keep us getting pionts in Primegrid.

EDIT: FYI I wont be putting much into the next Javelin throw. So I will be bunkering until the 4th throw. Let me know if you guys have decent bunkers for the next throw, I was going to wait, but realized the projects may expire before the 4th throw. I have about 600 gpu tasks. If we get a decent 3rd throw we wouldnt have to worry about that project anymore.

 

[tictoc]

I just realized that I didn't post the link to the individual stats. I didn't break out the Javelin, so that will just be the total for the whole Pent, but the other disciplines should match up to our team totals.

I just need to clean up the Google sheet, and then I'll post the link if you guys are interested. Stats are set to update every hour, except for the times when Google decides to do some random caching of the prior updates, in which case I just have to give the sheet a manual refresh.

-Edit- @franz you are probably best to just keep cranking out the points in PrimeGrid. I didn't have a chance to get any other GPUs on it yesterday, but I hope to get that sorted this evening.

 

[bfromcolo]

I only have 152 Numberfields, all completed. They all expire on the 14th.

 

[bfromcolo]

80 in progress task limit for TF on SR Base.