LTT Official Folding Month VI

[TheLANguy]

Strange 1/30 days folding? It looks like the active days folding reset.

[dogwitch]

aiming at end of event to have hit 750 million points total.

[podkall]

8 hours ago, ObeliskAG said:

Quite surprising to see just how much more Folding grunt a RTX 3080 Ti (10,240 shaders) has over a RTX 3080 (8,704 shaders), given how you don't see a comparable uplift in gaming FPS. However, the 4070 Ti seems to be the big winner for Folding efficiency.

actually you can often see slight fps change but huge productivity uplift in hardware,

also 3080 vs 3080ti isn't just, 1000+ more shaders.

 

the most I can quickly gather is,

3080ti has:

1000+ more shading units

20+ more ROPS

16 more TMUS

2GB more VRAM

12 more SM units

48 more Tensor cores

12 more RT cores

1MB more L2 cache

64bit higher memobry bus width

 

with that done I'll leave you with another meme:

[HoldSquat]

9 hours ago, dogwitch said:

aiming at end of event to have hit 750 million points total.

Drop the 0 and that's where I hope to be Just tossed the RPi5 in the other room and running headless to help

[LAR_Systems]

12 hours ago, TheLANguy said:

Strange 1/30 days folding? It looks like the active days folding reset.

It's a Monthly Active day count, and it's December, 1st   The GUI is probably cached on the word November and will reset shortly based on your location / timezone.

 

Edit: Pushed so much new stuff this year, I will be bug squashing most of next year.

[LAR_Systems]

15 hours ago, soccermomisqueen said:

Looking at the details page for the 12600k, it is either the servers (and/or clients) had a brain fart and used the GPU and CPU combined numbers, or the CPUs that were used for the numbers just hit a massive batch of workloads that were perfectly optimized for CPU folding since the top 4 in the ranking chart are the only ones that are above the "average performance of the CPU"

 

 

Another thing that is interesting as well is that the CPU folding times are very low and seem to be very GPU-like (I am basing that off of a i7-10870H CPU @ 2.20GHz taking about 8 hours on 8 cores to complete 1 WU).

This is a known pain in my ass when one of two things happen with the Folding@Home V7 client.

 

2 things can cause this.

 

The first is unicorn work units, if you go to the bottom of the page you will see some work units / projects might report some massive anomaly of PPD for a project that drags the average up on a device well beyond how it would perform on the majority of projects.

 

The other is some people will port forward the web client to a remote machine so they can monitor that machine... but the V7 Client does not report back the real CPU running on the slot as part of the remote data, it ends up reflecting the local machine the client is viewed on CPUs.

So if you have someone with a Xeon / Epic nasty CPU server, but they view it on some old i3 POS laptop... the I3 might get recorded as having awesome stats... there is no way for me to tell that's happening automatically, all I can do is sweep moderate records that don't make sense from time to time.

The good news... the new V8 F@H Client Beta reports the real CPU on a local or remote machine removing the bulk of these issues... as I see I more users start using the V8 Beta Client for my system, I will remove CPU reporting completely from the V7 version to have better data overall.

 

Thank you for attending my Ted Talk.

 

Edit:  Investigated, this particular CPU seems to be someone remoting in to something bigger, I just searched by the irregularities and they are all grouped.  Thanks for the report.

 

[justpoet]

1 hour ago, LAR_Systems said:

This is a known pain in my ass when one of two things happen with the Folding@Home V7 client.

 

2 things can cause this.

 

The first is unicorn work units, if you go to the bottom of the page you will see some work units / projects might report some massive anomaly of PPD for a project that drags the average up on a device well beyond how it would perform on the majority of projects.

 

The other is some people will port forward the web client to a remote machine so they can monitor that machine... but the V7 Client does not report back the real CPU running on the slot as part of the remote data, it ends up reflecting the local machine the client is viewed on CPUs.

So if you have someone with a Xeon / Epic nasty CPU server, but they view it on some old i3 POS laptop... the I3 might get recorded as having awesome stats... there is no way for me to tell that's happening automatically, all I can do is sweep moderate records that don't make sense from time to time.

The good news... the new V8 F@H Client Beta reports the real CPU on a local or remote machine removing the bulk of these issues... as I see I more users start using the V8 Beta Client for my system, I will remove CPU reporting completely from the V7 version to have better data overall.

 

Thank you for attending my Ted Talk.

 

Edit:  Investigated, this particular CPU seems to be someone remoting in to something bigger, I just searched by the irregularities and they are all grouped.  Thanks for the report.

 

I seem to remember somebody did a couple WU on LNG, just to show the difference at one point (no clue which CPU or generation).  So things like cryo/AC cooling systems I'm sure will be other outliers as well.

[LAR_Systems]

1 hour ago, justpoet said:

I seem to remember somebody did a couple WU on LNG, just to show the difference at one point (no clue which CPU or generation).  So things like cryo/AC cooling systems I'm sure will be other outliers as well.

The system working on averages smooths out that kind of thing... like unless you run on LNG for weeks you should not taint the record sets, as there would be tons of regular entries and the few strange outliers.   Same is on GPU PPD when some change clocks, bios, water cool etc.

The big killer is when they attach to something remotely and forward the ports and then log hundreds of data samples over the course of weeks as that will start to move the needle if it's a server CPU with massive number of cores, but the machine they observe on is some low core regular chip... that will mis read the number of threads vs the PPD (lower threads, higher PPD) and juice the scores.  

The records in question on the reported case where in the thousands... someone left that running some strange way for weeks on end.

[Alex Atkin UK]

3 minutes ago, LAR_Systems said:

The big killer is when they attach to something remotely and forward the ports and then log hundreds of data samples over the course of weeks as that will start to move the needle.

I don't understand, how is it able to get the CPU wrong if forwarding?

 

The whole web page is fed from port 7396 so even if you forwarded that the client should report the remote CPU data.

[LAR_Systems]

19 minutes ago, Alex Atkin UK said:

I don't understand, how is it able to get the CPU wrong if forwarding?

 

The whole web page is fed from port 7396 so even if you forwarded that the client should report the remote CPU data.

In Version 7 the JSON with the CPU slot PPD does not identify it other than it's a CPU and number of threads.  The work work around was needing to use the plugin/browser extension framework to get the actual CPU name, make, min/max threads etc.

In Version 8 they handle this correctly where all hardware is identified over the socket connection of the remote machine by the folding client and passed in the early connection data.

 

In Version 7, the default web interface and it's underlying code / APIs appeared to be an after thought with the focus on the trey application which reports all the correct full remote information via the logs etc.

[Alex Atkin UK]

2 minutes ago, LAR_Systems said:

In Version 7 the JSON with the CPU slot PPD does not identify it other than it's a CPU and number of threads.  The work work around was needing to use the plugin/browser extension framework to get the actual CPU name, make, min/max threads etc.

In Version 8 they handle this correctly where all hardware is identified over the socket connection of the remote machine by the folding client and passed in the early connection data.

 

In Version 7, the default web interface and it's underlying code / APIs appeared to be an after thought with the focus on the trey application which reports all the correct full remote information via the logs etc.

Come to think of it, v7 only tells you the CPU as I recall in System Info.  I couldn't be bothered to scrape that so hard-code the CPU info in my v7 to v8 json translation script.

[LAR_Systems]

1 minute ago, Alex Atkin UK said:

Come to think of it, v7 only tells you the CPU as I recall in System Info.  I couldn't be bothered to scrape that so hard-code the CPU info in my v7 to v8 json translation script.

Which I don't have access to in a web extension, for the remote system etc.  Hence the issue.

 

Best that could be done on 7, know the vast majority are not port forwarding, get the info the extension is allowed to get and have to sweep moderate stuff that is whacky... but that's manual... and I only got so many hours in the day.

 

If V8 becomes the main app, problem solved, but been sitting on the same beta version for a long time, so who knows.  

[Alex Atkin UK]

4 minutes ago, LAR_Systems said:

Which I don't have access to in a web extension, for the remote system etc.  Hence the issue.

 

Best that could be done on 7, know the vast majority are not port forwarding, get the info the extension is allowed to get and have to sweep moderate stuff that is whacky... but that's manual... and I only got so many hours in the day.

 

If V8 becomes the main app, problem solved, but been sitting on the same beta version for a long time, so who knows.  

Yeah I was dabbling with v8 last event, but it stopped working on the Mac Mini so went full v7 this time.

[SpaceRazor]

5 hours ago, LAR_Systems said:

This is a known pain in my ass when one of two things happen with the Folding@Home V7 client.

 

2 things can cause this.

 

The first is unicorn work units, if you go to the bottom of the page you will see some work units / projects might report some massive anomaly of PPD for a project that drags the average up on a device well beyond how it would perform on the majority of projects.

 

The other is some people will port forward the web client to a remote machine so they can monitor that machine... but the V7 Client does not report back the real CPU running on the slot as part of the remote data, it ends up reflecting the local machine the client is viewed on CPUs.

So if you have someone with a Xeon / Epic nasty CPU server, but they view it on some old i3 POS laptop... the I3 might get recorded as having awesome stats... there is no way for me to tell that's happening automatically, all I can do is sweep moderate records that don't make sense from time to time.

The good news... the new V8 F@H Client Beta reports the real CPU on a local or remote machine removing the bulk of these issues... as I see I more users start using the V8 Beta Client for my system, I will remove CPU reporting completely from the V7 version to have better data overall.

 

Thank you for attending my Ted Talk.

 

Edit:  Investigated, this particular CPU seems to be someone remoting in to something bigger, I just searched by the irregularities and they are all grouped.  Thanks for the report.

 

Thanks for following up on this. I have been shopping for some new hardware for a plex server and I was thinking about going with the 12600k because it was on sale for $153 USD. I checked the folding performance out of curiosity and was shocked to see the discounted 2 gen old CPU topped the list. I figured it was an anomaly in the data but I had no idea why. It has gone back up to around $190 at this point so I am holding off for a bit hoping it goes on sale again before Christmas. I would much rather build an AMD system again, but Plex uses the Intel iGPU for transcoding and so I'm forced to buy something from Intel regardless.

[justpoet]

11 minutes ago, SpaceRazor said:

Thanks for following up on this. I have been shopping for some new hardware for a plex server and I was thinking about going with the 12600k because it was on sale for $153 USD. I checked the folding performance out of curiosity and was shocked to see the discounted 2 gen old CPU topped the list. I figured it was an anomaly in the data but I had no idea why. It has gone back up to around $190 at this point so I am holding off for a bit hoping it goes on sale again before Christmas. I would much rather build an AMD system again, but Plex uses the Intel iGPU for transcoding and so I'm forced to buy something from Intel regardless.

Cheap ARC card?

[SpaceRazor]

I certainly considered that route. One of the big benefits of using onboard graphics is that you get a ton of transcoding performance with very little power draw or heat. Also, it leaves my PCIe slots for GPUs more suited to folding

 

I can't say that Intel onboard graphics are any better than AMDs, just that Plex has chosen to support and tailor their software to Intel graphics. It is a decision they made before AMD put themselves on back on the map in the CPU space so I get why they backed that horse, but now they are on a path that would require big changes if they wanted to add support for AMD onboard graphics and I don't know if the will is there.

[Windows7ge]

Does this mark the end of folding month? I just got my electric bill and it shot up $110...

[GOTSpectrum]

3 minutes ago, Windows7ge said:

Does this mark the end of folding month? I just got my electric bill and it shot up $110...

The end... a few more days yet 

[will0hlep]

3 minutes ago, Windows7ge said:

Does this mark the end of folding month? I just got my electric bill and it shot up $110...

folding month runs until the end of the 6th of december

[GOTSpectrum]

curious if it can hold 3ghz and not kill work units 

 

Pretty good temp for a two slot 4070ti 

[cbigfoot]

the math checks out points look good.

[Windows7ge]

15 minutes ago, GOTSpectrum said:

The end... a few more days yet 

RIP wallet.

[GOTSpectrum]

1 minute ago, Windows7ge said:

RIP wallet.

I'm burning £12 a day...

[Alex Atkin UK]

21 minutes ago, GOTSpectrum said:

curious if it can hold 3ghz and not kill work units 

 

Pretty good temp for a two slot 4070ti 

That's interesting, as I seem to get voltage limited at ~1040mV on the 4090 FE.  Always puzzled me why the voltage curve goes to 1250mV but it wont actually go that high.

 

I probably lose a bit on the 4070 Tis as they're on Linux, but then they perform so much better on Linux it would probably be a downgrade to move to Windows even if they clocked higher.

[Windows7ge]

3 minutes ago, GOTSpectrum said:

I'm burning £12 a day...

 

...a lot...