LAR_Systems

My router 12v power supply died causing me to wake up to cold folding machines last Sunday.

You're on the list now.

Don’t buy a 4090 until they figure out the power draw problem that’s melting cables. There are people with melted native cables on new PSUs with the 12v cord so it’s not just the nvidia adapter, it’s the card pulling more power than it should.

It's 21C where I am right now. Feels like it should be light for another 2 hours and the baseball game would be on. Instead, nope sweating in the dark on a Saturday night with you lot!

You're in the list now!

My desk is beside a server rack in the basement, shorts year round in Canada gets you looks when you step outside to get the mail, take out the bins etc. This is the tech geek life.

Yes, most of my systems operate on spite and coffee Or I used the Events results file to get peoples names, which I think are Forum names, not always folding names, I corrected a few that I knew from previous years myself, but you may be one of those lost in translation. Give me your folding name and I can update the script.

Looks like everything is trending on track for that lol.

Looks like they have been just assigning more BioFilm work units in the past week and not as much variety in work unit types than in previous weeks. https://folding.lar.systems/gpu_ppd/brands/amd/folding_profile/ellesmere_xt_radeon_rx_470480570580590

PROJECT #18454 RESEARCH FOR BIOFILM https://folding.lar.systems/projects/folding_profile/18454 Looks like other 480 users have successfully folded the project before, wondering if you got a bad work unit or hardware settings like overclocks etc. causing issues.

Okay, so now that my internet is back and stable... Built out a bit of a tool for the event so you can see participants rolling 24 hour PPD including your own in case you want to try tweaking your rigs and see if it's making a difference at a glance faster than EOC reports or waiting for the end of day event report. NOTE: This does not show the event TOTAL standings, this does not replace the blog post or official daily results in any way. This page shows you your PPD for the past 24 based on when you're actually looking at the page... so if you look at 3PM it's your PPD logged by F@H between 3PM today and 3PM yesterday... you look at 9AM... it's your PPD between 9AM today and 9AM yesterday etc. These values may seem a little high, or a little low hour to hour as some hours F@H is going to record your hardware dumping in a bunch of WUs and others maybe only a few, so depending on what hours those are in the 24 hour cycle it can make the numbers look a little better or worse than you might expect based on the "daily" total that's provided from EOC or the event based on using the same time of day to calculate each day. I have been using this tool so I can more easily see if changes I make to hardware are having a net positive or negative impact on 24 points aka PPD and the rolling window let's me make corrections or try new stuff every hour or so. So this might be useful to you, it might not, it's SUPER WTF ALPHA so may not be perfect, but here ya go. https://folding.lar.systems/league/

So, had a hardware casualty today, woke up to find all GPU systems idle and only CPUs slots with long timeframes still running. My rack routers power supply died in the middle of the night of course while I was sleeping, not while I was AWAKE AND USING THE INTERNET like I am the majority of the time and could have swapped PSUs right away. I imagine that's going to cost me some drops in rank.

I always love how F@H WU servers and calculation seem reliable and consistent until we're in an event and then I'm like... WTF when looking at EOC saying that out of 11 devices only 1 recorded a work unit in a 3 hour window which is impossible. #JustFoldingEventThings

This is the way it's supposed to look!

Each time I enter in to these events I'm like... this year I have a pretty good setup... then someone doubles my PPD lol. I might have the database information, but I don't have 2X the cards, power and money to run it all double equipment each year. Oh well, hope I stay in the top 10 by the end.

Sometimes these events come down to f@h server issues, power outages and unplanned folding machine crashes. it ain’t over until it’s over.

Overall PPD Ranks by GPU - https://folding.lar.systems/gpu_ppd/overall_ranks Overall PPD Ranks by GPU power usage to PPD - https://folding.lar.systems/gpu_ppd/overall_ranks_power_to_ppd Scores / Ranks are calculated based on real user average PPD data as the ranks, points and efficiency changes with the types of work units that F@H is assigning... Example, if you look at the 12 week chart for windows folding on 3090 it's trending up this week because F@H is giving out higher PPD work units (less % low PPD cancer WUs) more often than previous days, weeks etc. https://folding.lar.systems/gpu_ppd/brands/nvidia/folding_profile/ga102_geforce_rtx_3090 4090 is almost up 6million points from 14mil to 21mil PPD since launch 4 weeks ago... Looks due to F@H seems likely sending that card very specific work units now vs when they were likely gathering what it does best when it was new and they had no data on it. https://folding.lar.systems/gpu_ppd/brands/nvidia/folding_profile/ad102_geforce_rtx_4090 bottom of the page highlights this as some Cancer work units only got 3.5mil PPD average vs Alzheimer's work units at 22.7 mil PPD average. Folding hardware performance is always a moving target, that's why I built the "real time" database and plugin that feeds it because any static forum post or list is just a set of results from that point in time and may not be correct just days later. @gorgon has the info you need to understand what your card can do, tune it, optimize it for power to performance etc. My database tells you what the general trend is for your card and work units so you don't pull your hair out once you have your card dialed in and suddenly the PPD shoots up or down week to week as the F@H trends change, not your configuration.

I would give em a section on my site if interested lol.

Only a handful of data samples but one of you guys has a RTX 4090 logging work units. Not surprising the PPD looks nuts, and the power usage as well. https://folding.lar.systems/gpu_ppd/brands/nvidia/folding_profile/ad102_geforce_rtx_4090 Now we wait and see to what this normalizes out to over a couple days of data.

You are above average on the current work unit, wait until you start computing some Cancer WUs

Hey Everyone, The linked list might be of some help as the dropping GPU prices are putting more and more options out there for us to get, but you might want to know which ones are not going to cost you a fortune in power consumption over time. Results are ranked to show the GPUs with best KWh cost per 1 million F@H points. You can click on the GPUs name to see more details about the GPUs folding / power estimates and efficiency across different types of research. Cheers https://folding.lar.systems/gpu_ppd/overall_ranks_power_to_ppd

CPU reporting to the DB site would be disabled, because identifying the CPU comes from the browser, not F@H API... they pass the model of the GPUs, but not the CPUs in the web client, just a slot and the number of cores. Which is too bad, since they know the CPU cause they show it in the log, but not the APIs.

You already have my head spinning on that with your post. Would mean running the client logic a bit different, but not too different and just disabling CPU reporting if you were using this approach. I was working on a python client (so it runs on everything) that would send data headless to the cloud and then report back to page kind of like how crypto mining systems work... the issue then becomes the traffic, data, overhead, cost on my end aka... but this could be a nice middle ground cause I could support multiple ports in a config screen, the user just has to get them on localhost where I'm allow to talk to them.

You have found a bug in the daily use estimate that is making the form more confusing that it needs to be. To start folding at 5:00 PM each afternoon and stop folding at 1PM the next day to avoid the hot afternoon you will set it like this. Thanks to your post I have updated the code to include little hints in the start/finish info to say daily, next day, same day etc. depending on the hours selected and fixed the bug in the Daily Use estimate. You can set the client to these hours right now and it should do what you're looking for, but the Daily use estimate hours will be small / wrong for now. I have submitted the bug fix to the web store for Google review and you should see what I'm showing here in a few days once they approve the change. In the meantime if you set these hours the system should start at 5PM each day, fold 20 hours and switch to finish the next day at 1PM. Cheers,

Thanks for the report this has bene fixed.