F@H Web Client Dark Skin with PPD comparison by GPU / WU

[LAR_Systems]

On 11/5/2020 at 3:44 AM, SodaDog said:

Part time lurker here.
Love the interface so far, and how CPU stats are being implemented. Basically having it running to add to the database.

May I suggest a way to schedule folding tasks? We have solar and time of use rates where electricity costs basically double so I’d love the ability to have my computer go all out when the sun shines and shut down when it gets pricy.

Probably a long shot but I’d love to see a way to look up  and compare( Recorded PPD per GPU/CPU)  over (Manufacturer’s stated Wattage) in a sortable table. 

Time shifting your fold / finish folding based on time of day is something I'm working on.  F@H has fixed their service in recent updates to support "Finish" again..  So the goal is... a way for you to say... fold between X & Y time of day... if it's close to Y time of day... switch to finish WU so you don't loose the PPD you do with pause.

Otherwise some power data is currently available on the site, and the site is in redesign on my side so it can support more features as the Alpha/Beta site template was a rush job just to get up and running with a focus on the client.   

 

Would like to see it in an end state where you can go to the GPU / CPU page.    Then mess with a slider for % of increase or decrease off of manufacturer average wattage for the device for people to get a better long term cost estimate when over / underclocking.   

 

And make the rate used for cost a default that can be changed.
 

Baby steps first though... need the new site framework up and running first.

[LAR_Systems]

1 hour ago, A-J-M said:

Just noticed the new "WU reward". The "OOF" on one of my slots drew my attention...  

You have found my update lol   Got the feature in today, but did not get to do a write up yet.

 

So update to the interface.  The PPD +/- will continue to show you how your PPD is comparing to the average of all users PPD on the same project running the same GPU for the slot, this is a key performance indicator

 

But the WU Reward indicates how the currently running work unit compares to the average PPD potential overall for your card, the one found in the GPU ranks list etc. to determine if the reward of this particular WU is good or bad so you can stop wondering where your PPD went when the WUs change.

For example:   If your GPU on average gets 2MIL points... and the WU you're working on is getting 2MIL points... WU reward is "Normal" to indicate this WU is pretty close to the overall average you can expect from your GPU... this is good stuff.

 

Then the further above or below normal this work units PPD gets from your GPUs PPD average overall... you get different messages to help know it's not you, not your settings, etc. it's just a WU that is way off the average baseline good or bad.  

 

This way you can see if your PPD for a WU is good for that type of project and your GPU, but compared to all projects this single WU might not be the highest producing or could be something amazing and rare out of the norm.

 

 

WU Reward Ranks:

 

UNICORN - WTF kind of bizarre OP way epic high reward is this!  (special benchmark WUs and core tests / big changes mostly) 
GREAT - Doing well above the overall PPD average for your GPU.

GOOD - Doing a bit above the overall PPD average for your GPU.

NORMAL - +/- just about bang on the overall PPD average for your GPU 

OK -  Doing a bit below the overall PPD average for your GPU.

POOR - Doing well below the overall PPD average for your GPU.

OOF - WTF is this work units PPD problem, probably cancer, yep, it's cancer (or similar hard WU), it's ok it will be done soon.

 

[marknd59]

forget it I should read al the posts before I post.

 

[cbigfoot]

15 hours ago, marknd59 said:

forget it I should read al the posts before I post.

 

beat this  

thats new had to look up what a MYOSINS was...

[LAR_Systems]

9 hours ago, cbigfoot said:

beat this  

thats new had to look up what a MYOSINS was...

 

MYOSINS and other cause classifications is a new Algo I'm trying to ID what the causes really are when F@H just marks it "unspecified" then you start reading the info on these lessor know projects and there is some pretty interesting stuff in there.

[Gleep]

LAR_Systems,

First, I love the plugin.  Second, I have two machines folding and was wondering if there is anyway of getting the plugin to work while remotely viewing the client page? 

 

I can configure the FAH client to work from my primary desktop while viewing the second machine but the plugin doesn't load, even after tinkering with the dark mode files.  I, of course, can use dark mode while on the second machine, but I don't regularly turn on a display since it's just a folding machine and in an ideal world would prefer to view stats and contribute GPU data without having to keep a keyboard/mouse/display handy to fire up chrome. Sorry if this has been asked before.

[A-J-M]

8 hours ago, Gleep said:

LAR_Systems,

First, I love the plugin.  Second, I have two machines folding and was wondering if there is anyway of getting the plugin to work while remotely viewing the client page? 

 

I can configure the FAH client to work from my primary desktop while viewing the second machine but the plugin doesn't load, even after tinkering with the dark mode files.  I, of course, can use dark mode while on the second machine, but I don't regularly turn on a display since it's just a folding machine and in an ideal world would prefer to view stats and contribute GPU data without having to keep a keyboard/mouse/display handy to fire up chrome. Sorry if this has been asked before.

I use RDP: https://docs.microsoft.com/en-us/windows-server/remote/remote-desktop-services/clients/remote-desktop-clients

This way, you have total control of the distant machine(s). 

[LAR_Systems]

14 hours ago, Gleep said:

LAR_Systems,

First, I love the plugin.  Second, I have two machines folding and was wondering if there is anyway of getting the plugin to work while remotely viewing the client page? 

 

I can configure the FAH client to work from my primary desktop while viewing the second machine but the plugin doesn't load, even after tinkering with the dark mode files.  I, of course, can use dark mode while on the second machine, but I don't regularly turn on a display since it's just a folding machine and in an ideal world would prefer to view stats and contribute GPU data without having to keep a keyboard/mouse/display handy to fire up chrome. Sorry if this has been asked before.

Various browser security layers prevent running the extension connecting to clients on machines other than the local machine.

 

To get around this, a python script version of the "client" is in development  for running on headless or remote servers that would then send data to a centralized server where you could view / control it via a webpage.   

 

But this will not be ready for awhile at it requires a lot of testing cross platform.

 

[Gleep]

On 11/9/2020 at 10:34 PM, A-J-M said:

I use RDP: https://docs.microsoft.com/en-us/windows-server/remote/remote-desktop-services/clients/remote-desktop-clients

This way, you have total control of the distant machine(s). 

The second machine is running linux and thus RDP isn't going to work for me.

 

On 11/10/2020 at 4:15 AM, LAR_Systems said:

Various browser security layers prevent running the extension connecting to clients on machines other than the local machine.

 

To get around this, a python script version of the "client" is in development  for running on headless or remote servers that would then send data to a centralized server where you could view / control it via a webpage.   

 

But this will not be ready for awhile at it requires a lot of testing cross platform.

 

Thanks for the reply, good to hear there is something in the works.

[A-J-M]

10 hours ago, Gleep said:

The second machine is running linux and thus RDP isn't going to work for me.

 

Thanks for the reply, good to hear there is something in the works.

In the meantime, you might find HFM.net interesting: https://github.com/harlam357/hfm-net

You can install it on the Windows machine and add your Linux client. It gives you a good level of control on the remote (and local) client and builds a database of all "your" WUs. Not as fancy and fun as the Dark Extension, but nice, too. 

[TVwazhere]

On 11/6/2020 at 2:17 PM, LAR_Systems said:

OOF - WTF is this work units PPD problem, probably cancer, yep, it's cancer (or similar hard WU), it's ok it will be done soon.

This made me laugh

[Kilrah]

On 11/13/2020 at 1:22 AM, Gleep said:

The second machine is running linux and thus RDP isn't going to work for me.

With linux you use VNC as equivalent.

[Letgomyleghoe]

nvm

[LazyDev]

I don't know if this is a bug, but with my 6 GPU's installed an running, I only get to see three graphs on the web page.

 

[cbigfoot]

18 hours ago, LazyDev said:

I don't know if this is a bug, but with my 6 GPU's installed an running, I only get to see three graphs on the web page.

 

you are using it wrong you shouldn't have but 3 per computer  Right @LAR_Systems

[LAR_Systems]

21 hours ago, LazyDev said:

I don't know if this is a bug, but with my 6 GPU's installed an running, I only get to see three graphs on the web page.

 

I can say you have pushed this beyond my testing setup for sure.

That said, looking at the slot IDs I think that would be the issue... they are not in order or sequential suggesting you have been adding and removing slots running up the ID number.   

Think in the live version of the client I only account for 8 slots with charting as 8 GPUs is pretty much the max config for a pretty baller system running 4X GPU lanes per card.

Would suggest in the short term that you finish the slots... remove them all.   Reset, add them back.

You should get slots 1-6 which the charting will show for.

[LAR_Systems]

Hi All

 

Version 2.0 of the https://folding.lar.systems/ PPD database site has just gone live.

 

Along with some navigation, UX and overall template updates the 2.0 version launches the public version of the CPU PPD database and provides stats on the number of samples etc. available for devices when showing averages.

GPU Database: https://folding.lar.systems/gpu_ppd/overall_ranks
CPU Database: https://folding.lar.systems/cpu_ppd/overall_ranks

These is also a simplified database for viewing active folding projects

https://folding.lar.systems/projects/

 

and keep a look out for large data tables now having search / sort enabled to help you view data the way you want to.

 

Cheers!

 

[shorttack]

Found your sight last weekend and joined immediately. Giving you 3090, 3080, and 2080 super clients.

 

What I can't explain is why my clients are doing so much better than your average clients? The 3090 is doing 51.5% better than the database average on project 13428, yet my machine is an Ivy Bridge_E 4-core from maybe eight years ago. However, almost all my clients run Linux Mint 19. My Windows testing never came near to a 50% Windows penalty.

 

Any ideas? Any thoughts for new features that allowed you access to folding client platform data?

DX79SI config

[Favebook]

11 minutes ago, shorttack said:

Found your sight last weekend and joined immediately. Giving you 3090, 3080, and 2080 super clients.

 

What I can't explain is why my clients are doing so much better than your average clients? The 3090 is doing 51.5% better than the database average on project 13428, yet my machine is an Ivy Bridge_E 4-core from maybe eight years ago. However, almost all my clients run Linux Mint 19. My Windows testing never came near to a 50% Windows penalty.

 

Any ideas? Any thoughts for new features that allowed you access to folding client platform data?

DX79SI config 4.57 kB · 0 downloads

Probably due to linux. Also, those are new cards and there isn't a ton of data on them yet, that is why they might vary a lot in PPD.

Linux does give you better PPD, that is for sure, but I wouldn't say every unit is 50%+ better than on windows.

[A-J-M]

I guess many people tune down their big cards if they have to live and/or work with them. If you have your rigs in the basement, no-one is disturbed by the noise (and the heat in summer). If you work with them all day...

 

I do "live" with my rigs and I have made this experience with two 2080ti. One is now under water and can work 100%, most of the time above 5M, I'm still waiting for a water block for the second one and I had to limit her to 75%. And in the summer, I limited all GPUs because of the heat. 

 

Further, I would guess that people who actually work with their folding rigs will rather use Windows, whereas dedicated folding rigs in a separate room will rather run on Linux. That would account for the better results reported with Linux. I have tested this last spring with a one dedicated machine (X299, 5700XT, 1080ti) and I could not see any sustained difference. 

 

EDIT: (re test of last spring: there was a clear difference, 20-25%, for CPU folding, though)

[Kilrah]

6 hours ago, shorttack said:

What I can't explain is why my clients are doing so much better than your average clients? 

Make sure to refresh the page, with fast cards sometimes the page doesn't update with the new project on a new WU, and the % comparison bases itself on the project of the previous WU.

[terkhen]

Hello,

 

Is it possible to use the extension with URLs other than 127.0.0.1 and https://client.foldingathome.org/ ? I usually check the status of my folding computer from a laptop in the same local network and in that case the theme of the page is not changing.

[A-J-M]

52 minutes ago, terkhen said:

Hello,

 

Is it possible to use the extension with URLs other than 127.0.0.1 and https://client.foldingathome.org/ ? I usually check the status of my folding computer from a laptop in the same local network and in that case the theme of the page is not changing.

Not yet: 

 

[LAR_Systems]

13 hours ago, shorttack said:

Found your sight last weekend and joined immediately. Giving you 3090, 3080, and 2080 super clients.

 

What I can't explain is why my clients are doing so much better than your average clients? The 3090 is doing 51.5% better than the database average on project 13428, yet my machine is an Ivy Bridge_E 4-core from maybe eight years ago. However, almost all my clients run Linux Mint 19. My Windows testing never came near to a 50% Windows penalty.

 

Any ideas? Any thoughts for new features that allowed you access to folding client platform data?

DX79SI config 4.57 kB · 0 downloads

Hi, thanks for installing!

 

Aside from what the others have said, Linux is better PPD, faster cards people tend to downclock... etc.

Project 13428 has historically been a poor performing project on 30 series systems.   What may have happened is F@H / Project team has modified the project WUs or bonuses as I can see these kinds of fluctuations in the data that are bigger than just normal daily adjustments.

For the newer 30 series cards I know from what I'm running F@H is leaving a lot of GPU power on the table right now, so hopefully they are further optimizing work units for the newer cards. 

All that said... I have been collecting WU data from the client by OS for a couple months now and with the most recent version of the website deployed I can start working out the best way to convey that data on the site and in the client.   

Right now you are seeing the GPU / Project average, but I want to get it so you are seeing the GPU / Project average by your OS as well for better comparisons.   

But then I need that info on the site side by side as well for when people freak-out someone says they are getting way higher / lower PPD on the same cards not taking OS and drivers in to account.

For now, just keep folding, you might have gotten some project test WUs that are just faster running and higher reward and they will be washed out by the average of all work, or you're setting the new trend for the project if it's higher reward or more productive now on your GPU and that will start to creep up in the "Average PPD" over time until it levels off to around what your system is producing.

[LAR_Systems]

FYI: Google has changed the way they review the extension sometime around 12AM EST today and therefore I have to resubmit it updated.

 

It has to do with the CPU tracking feature, they are suggesting the app does not use the CPU permission, and you can't have an app that asks for permissions it does not use... meanwhile it showing CPU names, logical processors and their PPD suggests otherwise.

I assume it's a matter of review automation not being able to tell how it's used, and it's hopefully resolved quickly.

Kinds feeling like Luc with floatplane app fun right now.  

I will update once back online.