BOINC Pentathlon 2025

[TennesseeTony]

May or may not have been the same storm, but....

Tennesseetony was affected by this (new story link below, not spamming) Could have affected others here also, 2 that I know of.

437 Reaktionen · 26 Kommentare | At least 14 people were killed as severe storms swept through Kentucky overnight and the number is expected to grow, Governor Andy Beshear said. | CBS News

At least 14 people were killed as severe storms swept through Kentucky overnight and the number is expected to grow, Governor Andy Beshear said..

www.facebook.com

Fortunately all my family is safe, a bit north of the most affected area.

 

[IEC]

No power outages here (*coughs*, this week) though ambient temps are insufferably warm.

I look forward to taking a summer hiatus.

 

[StefanR5R]

Hmm, the only thing mentioned in original message was 1024 daily quota so no idea what type of "quirk" you're talking about.

I don't want to harp on, but I put it in other words once more:

The daily quota was exactly what I was talking about. Consider a 24h period. The host receives 1024 tasks after the start of this period, and returns results of all of them before the 24h period is over. Those results which can be validated right away because a wingman result exists turn out valid. Yet no new tasks are assigned to the host anymore before the 24h period is over. Or only one new task after each additionally returned result, I don't remember how it worked exactly.

IIRC the quota is increased over time based on production history of the host, but this process takes time.

 

[Rattledagger]

The host receives 1024 tasks after the start of this period, and returns results of all of them before the 24h period is over.

While a 1024 daily quota can be a problem with a 64+ thread server, I still don't really see the big problem then it comes to "normal" home computers.

To me a "normal" home computer include at most one "usable" GPU + up to 32-thread CPU. While some CPU's also include GPU, I would guess this built-in GPU is so weak compared to the other GPU that where's not much point to use it.

For numbers, going by Einstein@home's top-50 computers, one of them at least according to given information include 6 Geforce 5090 and this is AFAIK at the moment the fastest GPU you'll find in a "normal" home computer. Among the close to 17k "pending" + "valid" results, 6 have run-time below 100 seconds, less than 134 for 100 - 199 seconds, less than 80 for 200 - 299 seconds, less than 6440 for 300 - 399 seconds and more than 10139 took 400 seconds or longer, all the way up to 655 seconds.
Taking a low-end average means on average it takes at least 358 seconds run-time per task.

Meaning, if limits to a single GPU, this means at most 241 tasks/day.

If computer in addition use CPU to crunch on, now if I read application-page correctly, only the FGRP5 is not using GPU and currently send-out work. Among top-50 computers most uses GPU and not CPU, but a few also use CPU. It seems the fastest CPU result took 3893 seconds run-time. Using this "best-case" result and assuming max 32 threads means 710 tasks/day.

CPU + GPU is wherefore 951 tasks/day < 1024 daily quota.

 

[Skillz]

That host with "6" 5090s does not have all 5090s. It's also got 4090s on that host as well.

What we don't know is how many of those tasks are running simultaneously on the host. It's very easy to configure Einstein@home to run multiple tasks per GPU. Assuming it's just one task completing in ~150 seconds that's 576 tasks per day on GPU only.

We can also see that it's not using the custom app for those tasks which is a good bit faster than the stock app.

Additionally running a mix of 6 5090s or 4090s will take one hell of a PSU and will consume more than a single 15A 110V breaker can handle. Not knowing what kind of power it's consuming plays a huge role on those numbers as well. They could very well be power limiting those GPUs to keep the whole system within' ~1600W. Doesn't seem like it based on the ~100 second run times, but it's something we do not know.

I'm not running E@H hard at all this year, but the last year (2 years ago) that Einstein@home was in the Pent I was running into that 1024 limitation on nearly all of my computers the whole time. I was forced to create new instances on each one of them every day to keep them going.

It's not as big a problem this year because the higher PPD app is the O3AS and it's run times are significantly longer.

But the biggest problem this occurs is during bunkering with a new host. You download ~1000 tasks. Crunch all 1000 of them in the 1 - 3 days leading to the event. Drop the bunker and then you wont be able to get any new tasks on that host since the server see's all 1000 of those tasks completed within' a 24 hour period. Even though majority of those tasks will be validated by that time. You'll still be in the quota jail. This is possible with nearly any computer with a GPU.

 

[mmonnin03]

Rattle, Your data is bogus. Multiple tasks at once means the per time per task is much less. Running Windows means a lot of missing performance. My own 3080Ti in Linux can run 400-500 tasks per day on a single GPU. A 5090 could do much, much more.