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

[cbigfoot]

18 hours ago, LAR_Systems said:

Don't want to surprise anyone. 

So this is the official heads up that version 1.1.2 of F@H in the dark has been submitted to Google for approval. 

This means if approved it will probably show up sometime in the next 3-5 days based on previous review rounds.  

 

This update contains UI updates, fixes, the first round of CPU support for the plugin and performance benchmark charting that can be enabled to see how a work unit is folding over time and is helpful when clocking memory / power on cards to see impacts.

 

The charting has a toggle so if not in use it does not use any GPU 3D resources and in the same line of thought animated GIFs have been removed from the GUI completely to avoid GPU 3D wasted cycles.

I will keep you posted when I know how the review goes or you will see the client in the wild before I know it's been pushed by Google.

@GOTSpectrum rest assured Team LTT skin still included.

 

Cheers,

 

Spoiler

 

 

 

Spoiler

 

 

 

was more like 12 hrs lol

[LAR_Systems]

2 hours ago, Kilrah said:

Is there a plan to also have a database for the CPU PPDPLPs?

 

Looks like CPU name comes up correctly

 

Yes, the data is collecting so I can validate and test the PPD:PLP with a larger number of samples than my 4 CPUs and when I think it's sorted out CPU DB will  be added to the website, and the CPUs in the client will link etc. like the GPUs do.

[LAR_Systems]

3 hours ago, A-J-M said:

Great job! 

Just a small blip in the non-LTT version in Slot Charts mode: the upper part of the first line (Slot Performance: ...) is slightly visible: 

 

Thanks, think it's doing it on the LTT one as well.  
 

I will give it a look.

[NetoriusNick]

really dumb Q I take it there is no way to show more than one machine per page as it runs off of the FAH service and is just a skin?

[Sequence211]

Great work.

I will have the update next time i reboot my browser.

 

A suggestion: Maybe the PPD should not be colored if the value is within +/-5% of the average?
So if your PPD is -0,62% from the database average, it should not be in red text.
This way it is more clear if the card is performing as expected, or way lower/higher than extected.

[DirectingGnu2]

Eyyyyy 3090 broke 6 million

[LAR_Systems]

3 hours ago, NetoriusNick said:

really dumb Q I take it there is no way to show more than one machine per page as it runs off of the FAH service and is just a skin?

That is correct... for now.

 

Part of the change in layout to stacked processor slots and a total estimated PPD is I'm exploring ways to have the client sync once X minutes with updates with other open clients using the same username, team and API key.

 

The stacked layout would allow for detailed summary info for lots of GPU/CPU slots to be on one screen with a machine ID etc. to tell them apart.

 

So at a glance you could see all your gear... and using name / API Key could also see them out and about with a dashboard on the DB site (the part I really want when out of the house).

I hate as much as anyone the need to open firewall ports, passwords etc. to see multiple machines using advanced client and figure this would be a much simpler more secure way for people to keep an eye on their folders.   

 

This also ties in to some python scripting I'm working on for "headless" clients so you can just run the script on folders you want to sync, then view the dashboard elsewhere.  

 

But this needs a lot of testing and math on my side for server load, caching and  bandwidth calculations to determine the best and economical way to do it from a server perspective, if that's possible.

Can't go bankrupt running a free service

[LAR_Systems]

3 hours ago, Struck said:

Great work.

I will have the update next time i reboot my browser.

 

A suggestion: Maybe the PPD should not be colored if the value is within +/-5% of the average?
So if your PPD is -0,62% from the database average, it should not be in red text.
This way it is more clear if the card is performing as expected, or way lower/higher than extected.

Makes sense... will put a +/- acceptable range on the dev list.   Does make me a little sad when I see red at -0.06 %

[NetoriusNick]

7 minutes ago, LAR_Systems said:

That is correct... for now.

 

Part of the change in layout to stacked processor slots and a total estimated PPD is I'm exploring ways to have the client sync once X minutes with updates with other open clients using the same username, team and API key.

 

The stacked layout would allow for detailed summary info for lots of GPU/CPU slots to be on one screen with a machine ID etc. to tell them apart.

 

So at a glance you could see all your gear... and using name / API Key could also see them out and about with a dashboard on the DB site (the part I really want when out of the house).

I hate as much as anyone the need to open firewall ports, passwords etc. to see multiple machines using advanced client and figure this would be a much simpler more secure way for people to keep an eye on their folders.   

 

This also ties in to some python scripting I'm working on for "headless" clients so you can just run the script on folders you want to sync, then view the dashboard elsewhere.  

 

But this needs a lot of testing and math on my side for server load, caching and  bandwidth calculations to determine the best and economical way to do it from a server perspective, if that's possible.

Can't go bankrupt running a free service

oh yeah dont go BC on something free

just trying to get the web page working for my headless unit and its annoying its not enabled from the start

[NetoriusNick]

can this not be run on more than one web control page at once?

 

[Favebook]

21 minutes ago, NetoriusNick said:

can this not be run on more than one web control page at once?

 

Why would you even do that?

[LAR_Systems]

21 minutes ago, NetoriusNick said:

can this not be run on more than one web control page at once?

 

You can't open it twice in the same browser on the same system.   

The F@H will yell at you and the client stops the old browser in favor of the new one.

If you mean, can this reference a web client on a different system, no.   Web browser security policies will not let the F@H extension or my version of it point to anything but the local machines host address to read a F@H service.

This is where the advanced client has it's uses as it's got admin / system permissions installed as software and does not have to obey web browser security policies. 

[NetoriusNick]

10 minutes ago, LAR_Systems said:

You can't open it twice in the same browser on the same system.   

The F@H will yell at you and the client stops the old browser in favor of the new one.

If you mean, can this reference a web client on a different system, no.   Web browser security policies will not let the F@H extension or my version of it point to anything but the local machines host address to read a F@H service.

This is where the advanced client has it's uses as it's got admin / system permissions installed as software and does not have to obey web browser security policies. 

ah okay fair enough urg wanted to see how the headless was doing on points in comparison to your records

[NetoriusNick]

14 minutes ago, Favebook said:

Why would you even do that?

many folding rigs wanted to see how they are doing in comparison to LAR records

[LAR_Systems]

25 minutes ago, NetoriusNick said:

ah okay fair enough urg wanted to see how the headless was doing on points in comparison to your records

You’re looking for the python script I’m messing with for headless systems.  Coming soon lol, but need to finish cpu addition to the db site first.

[NetoriusNick]

9 hours ago, LAR_Systems said:

You’re looking for the python script I’m messing with for headless systems.  Coming soon lol, but need to finish cpu addition to the db site first.

please tell me you are going to include babysitter of some kind and something HFM like?

[NetoriusNick]

9 hours ago, LAR_Systems said:

You’re looking for the python script I’m messing with for headless systems.  Coming soon lol, but need to finish cpu addition to the db site first.

if you want to look/ check temps bpytop is a nice project I have used on a few machines

[LAR_Systems]

On 10/23/2020 at 4:26 AM, NetoriusNick said:

please tell me you are going to include babysitter of some kind and something HFM like?

The goal is babysitter, that sends data to site, site that let's you see what's going on,  remotely send commands to babysitter to execute basic commands like finish, pause, stop.

 

Goals: without a chrome extension / web client. 

• Babysitting
• Remote review of all your folders 
• Remote manual control of all your folders
• Remote scripted control of your folders aka (if after X time of day fold while power is cheap (it's cold, etc.), set projects to finish X time before power gets expensive, it gets hot out again, or pause / stop)

 

[LAR_Systems]

HI everyone,  Quick update. 

 

There was some server updates today to deal with server resource issues so sorry if you  were impacted by some strangeness like WU averages etc. being a little strange while this occurred.

 

In addition to the cleanup you will now find that for GPU projects the Atoms and Cause is reported on the slots.

Atoms and Cause coming to CPU slots soon.

 

 

The PPD indicator has been updated so there is a +/- 10% tolerance for the red or green indicators coming on for when you GPU slot is operating within this "Average" range making it a more useful indication as to when PPD is out of the ordinary.

 

Project pages on the database site have been updated to show the Atoms, Core and project branch status of projects.



And CPU based project pages (link found in the CPU slots in the web client) now contain the BETA charts for PPD-PLC (PPD per logical processor) as that system builds data and is under review.

 

 

These client based updates will trickle in to the wild over the next couple hours and take effect with a refresh of the page when they do.

[Starman57]

Just a heads up.  The folding in the dark page doesn't display quite correctly when using the FAH Client 7.6.21.

 

Steve

[LAR_Systems]

9 hours ago, Starman57 said:

Just a heads up.  The folding in the dark page doesn't display quite correctly when using the FAH Client 7.6.21.

 

Steve

An update has been pushed which should trickle out over the next hour or so on refresh of the client.

 

The new project for this GPU messaging was not launching correctly.

When a GPU loads a project that does not have any or has very low WU samples in the database it's supposed to simply say that in the performance window until enough data has collected to start showing averages.

 

What you're seeing is the we go data display screens... and what they do when they don't have data which makes your running machine = to the average because your machine is comparing itself to itself.

 

Anyways... hopefully it's patched now... hard one to tackle cause you only see the case of a GPU / Project combination not already in the DB... and my GPUs for testing are the most common in the DB.

[Starman57]

Confirmed.  Looks good.

 

On another note, are the CPU PPD:PLP numbers exclusive?  Based on what I'm seeing here, when used in conjunction with a thirsty graphics card, BOTH CPU and graphics card numbers are affected.  The trick is to find the right balance to max out the "effective" total system PPD.  This of course skews your database because individually their numbers are higher.

 

Or to put another way, I have to sacrifice CPU threads to keep the graphics card "fed", and for the purposes of your database, will skew the CPU numbers lower because I'm not using the full potential of the CPU for folding.  In fact, I'm also skewing the graphics card lower too, just so long as the drop in performance is less than the gain on the CPU side.

 

This is why I don't bother to CPU fold on anything less than 6 cores (6 or 12T) or so.  The drop in performance of the graphics card is more than offset by the marginal increase in PPD of the CPU, bottlenecks being what they are.  I'm not sure if the FAH client will ever alleviate this, or it's just something we have to live with.

 

 

[LAR_Systems]

2 minutes ago, Starman57 said:

Confirmed.  Looks good.

 

On another note, are the CPU PPD:PLP numbers exclusive?  Based on what I'm seeing here, when used in conjunction with a thirsty graphics card, BOTH CPU and graphics card numbers are affected.  The trick is to find the right balance to max out the "effective" total PPD.  This of course skews your database because individually their numbers are higher.

 

 

It's why the database is based on averages.

Some people run stuff overclocked max voltages... others run underclocked and power limited to reduce heat.   Some choke a GPU with folding on all threads of the CPU at the sametime... etc. etc.

 

So as WU samples build values are all based on averages of all samples to take that spectrum and try to find a middle baseline with a margin of error that is still more accurate then the napkin math done for this stuff based solely on clocks and cores cause it takes how the different WUs also hammer the hardware differently.

 

As for finding the balance... the addition of CPU PPD-PLC is completely new to the system so I'm not here yet.   

 

But the end goal is to to have the average PPD-PLC, CPU and Logical Cores assigned to the folding machine do some math against the GPUs assigned to the folding machine to actually warn / recommend users on how to adjust the number of logical cores they are using to not choke the GPUs and get max PPD IN general or from the most efficient device they have.

Rather my GPU churn out an extra 100PPD on less CPU threads than my CPU on more threads if the CPU is using 3X the power etc.

[Starman57]

Ah, lovely clarity.

 

This work is good stuff.  I applaud your "end goal" as it finally gives an appropriate feedback loop to the end user as to HOW they should be using the FAH application based on system specs.  I would surmise that many will find this information very useful, most especially those with dedicated folding rigs.

 

As I can now actually see, what I've been hand tuning in the past, where and when the performance was affected by two compute units fighting over fixed resources.  It was problematic in the past as the compute LOADS for various WUs created all sorts of variability in observations.  Not good for nailing things down.

 

I can now see the light.

Thanks.

 

[LAR_Systems]

Think this will be the last code update of the day.

 

Couple more little patches checked off the list, of note that you would see is the visual glitches when the slots are loading and might display a sliver the of the performance window profile.

 

As there are a few other bug fixes in the mix I would suggest refreshing the window to make sure these updates take effect on your system.

Cheers,