Anandtech vs Tom's Hardware Folding@Home Coronavirus Race thread

[Soulkeeper]

I got the exact same error last night(first time ever). I think it was a batch of beta WUs with an issue.
nan = not a number

 

[Soulkeeper]

I think i'm starting to understand the performance/behavior of folding on the gpu.
It seems there are a lot of pcie transfers and heavy reliance on double precision math.
It uses ~1.8GB and every 5% or so there is a pause as it waits to burst more over the pcie and refill it.
The reliance on doubles means 1/4th or less of the consumer hardware gets used.
On the plus side this makes folding less likely to thermal limit, you can push the clocks/volts a lot higher.
A <2GB working set size would make sense, targeting more hardware. Not as efficient on 8/16GB card however.

This would explain why the vega vii does modest (more dp gflops). And why the newer 5700xt does well (pcie4 and higher core clocks).

 

[StefanR5R]

Day 24: Saturday, April 11
14:00 UTC stats
(10 AM EDT, 7 AM PDT)
links: official race stats, TeAm AnandTech at EOC, Tom's Hardware at EOC

Daily production


Our advantage


What's up with Tom's? Perhaps they should check out our thread and learn how to game and keep folding at the same time. ;-)

 

[Soulkeeper]

Toms wasn't ready for this race.

Info - Welcome to the Tom's Hardware F@H Team Thread (Team 40051)!

Due to a number of longstanding chronic issues with the previous threads, we present to you all a brand new thread, stickified with some extra strong adhesives, for our illustrious folding team! You can find the original thread here, which may still be useful, albeit with outdated information...

forums.tomshardware.com

Their team thread only has 3 pages, anandtech is now upto 25 pages just less activity/involvement it seems.

 

[mopardude87]

What's up with Tom's? Perhaps they should check out our thread and learn how to game and keep folding at the same time. ;-)

LOL yeah i was playing Wizard101, barely tasking the HD630 having a ball. I almost forgot i was folding cause it ran so good. Then i was able to play OG Bioshock maxed out with like 55-70fps or so at 900p. If the cpu isn't folding or has tons of threads its possible to game still but you need another gpu. Thinking i may grab a 1070 as my main when i swap out the 7700k for the 3900x, still set the 1080ti to fold. When Ampere drops, i know 3 people including myself getting upgrades in theory.

Let the 1070 game on the first 12 threads via affinity, the others will just fold with the 1080ti. All i want out of this setup.

 

[Endgame124]

I got the 1660 super some work units, but with wild fluctuations on ppd - min was 621k, max was 1.02M.

tell me there is a log somewhere that displays the final points for a WU - if so, I could do some good stuff with that

 

[StefanR5R]

08:46:49:WU00:FS01:0x22:Completed 3120000 out of 4000000 steps (78%)
08:46:58:WU00:FS01:0x22:Bad State detected... attempting to resume from last good checkpoint. Is your system overclocked?
08:46:58:WU00:FS01:0x22:Following exception occured: Particle coordinate is nan
08:49:27:WU00:FS01:0x22:Completed 3160000 out of 4000000 steps (79%)

What does 'nan' mean again btw?
What happens to the aborted WUs btw, do they get resent? (I noticed the log flagged them as 'bad work unit'! ).

nan, or more commonly NaN, means "not a number". It is a code which indicates any value which cannot be represented in standard floating point encoding anymore, i.e. a too large or too small number.

As to what happens with WUs which were returned unsuccessfully, I do think they will be sent out to another client. Just like WUs which are not returned before the "timeout" datum are being sent to another client.

tell me there is a log somewhere that displays the final points for a WU - if so, I could do some good stuff with that

In the Folding@home client's data directory, there is 'log.txt' as the current log, and 'logs/log-YYYMMDD-HHMMSS.txt' from log rotation.

Spoiler: example

$ grep Final /var/lib/fahclient/log.txt
18:12:24:WU01:FS01:Final credit estimate, 77361.00 points
19:41:06:WU00:FS00:Final credit estimate, 167665.00 points
19:44:26:WU02:FS02:Final credit estimate, 174291.00 points
21:09:22:WU03:FS01:Final credit estimate, 170233.00 points
21:54:36:WU04:FS02:Final credit estimate, 113989.00 points
22:39:09:WU01:FS00:Final credit estimate, 176701.00 points
23:12:55:WU00:FS01:Final credit estimate, 122679.00 points
23:43:42:WU02:FS02:Final credit estimate, 76759.00 points
00:43:45:WU03:FS00:Final credit estimate, 123456.00 points
01:58:13:WU00:FS02:Final credit estimate, 125837.00 points
02:10:06:WU01:FS01:Final credit estimate, 174472.00 points
03:38:30:WU02:FS00:Final credit estimate, 174988.00 points
04:08:58:WU03:FS02:Final credit estimate, 122993.00 points
04:11:42:WU00:FS01:Final credit estimate, 123386.00 points
05:32:34:WU04:FS01:Final credit estimate, 78220.00 points
05:34:29:WU02:FS02:Final credit estimate, 76005.00 points
05:41:03:WU01:FS00:Final credit estimate, 123596.00 points
08:04:35:WU04:FS00:Final credit estimate, 127857.00 points
08:28:25:WU03:FS01:Final credit estimate, 175741.00 points
08:36:42:WU00:FS02:Final credit estimate, 175475.00 points
10:08:09:WU01:FS00:Final credit estimate, 125839.00 points
10:31:01:WU02:FS01:Final credit estimate, 123472.00 points
11:36:41:WU04:FS02:Final credit estimate, 171548.00 points
12:34:40:WU01:FS01:Final credit estimate, 124678.00 points
13:04:55:WU00:FS00:Final credit estimate, 175620.00 points
13:47:35:WU02:FS02:Final credit estimate, 120686.00 points
14:09:51:WU03:FS01:Final credit estimate, 91818.00 points
15:03:17:WU02:FS03:Final credit estimate, 14035.00 points

 

[StefanR5R]

Toms wasn't ready for this race.

Info - Welcome to the Tom's Hardware F@H Team Thread (Team 40051)!

Due to a number of longstanding chronic issues with the previous threads, we present to you all a brand new thread, stickified with some extra strong adhesives, for our illustrious folding team! You can find the original thread here, which may still be useful, albeit with outdated information...

forums.tomshardware.com

Their team thread only has 3 pages, anandtech is now upto 25 pages just less activity/involvement it seems.

Oh, they created a "Distributed Computing" subforum now! They didn't have this until recently. Entering the search terms "folding@home" or "boinc" in their forum shows discussions all over the place, buried between lots of other topics. (They don't have an actual boinc team anyway, and you have to admire their folding team for their past and current activity even without an own place in Tom's forum, until now.)

 

[Endgame124]

nan, or more commonly NaN, means "not a number". It is a code which indicates any value which cannot be represented in standard floating point encoding anymore, i.e. a too large or too small number.

As to what happens with WUs which were returned unsuccessfully, I do think they will be sent out to another client. Just like WUs which are not returned before the "timeout" datum are being sent to another client.


In the Folding@home client's data directory, there is 'log.txt' as the current log, and 'logs/log-YYYMMDD-HHMMSS.txt' from log rotation.

Spoiler: example

$ grep Final /var/lib/fahclient/log.txt
18:12:24:WU01:FS01:Final credit estimate, 77361.00 points
19:41:06:WU00:FS00:Final credit estimate, 167665.00 points
19:44:26:WU02:FS02:Final credit estimate, 174291.00 points
21:09:22:WU03:FS01:Final credit estimate, 170233.00 points
21:54:36:WU04:FS02:Final credit estimate, 113989.00 points
22:39:09:WU01:FS00:Final credit estimate, 176701.00 points
23:12:55:WU00:FS01:Final credit estimate, 122679.00 points
23:43:42:WU02:FS02:Final credit estimate, 76759.00 points
00:43:45:WU03:FS00:Final credit estimate, 123456.00 points
01:58:13:WU00:FS02:Final credit estimate, 125837.00 points
02:10:06:WU01:FS01:Final credit estimate, 174472.00 points
03:38:30:WU02:FS00:Final credit estimate, 174988.00 points
04:08:58:WU03:FS02:Final credit estimate, 122993.00 points
04:11:42:WU00:FS01:Final credit estimate, 123386.00 points
05:32:34:WU04:FS01:Final credit estimate, 78220.00 points
05:34:29:WU02:FS02:Final credit estimate, 76005.00 points
05:41:03:WU01:FS00:Final credit estimate, 123596.00 points
08:04:35:WU04:FS00:Final credit estimate, 127857.00 points
08:28:25:WU03:FS01:Final credit estimate, 175741.00 points
08:36:42:WU00:FS02:Final credit estimate, 175475.00 points
10:08:09:WU01:FS00:Final credit estimate, 125839.00 points
10:31:01:WU02:FS01:Final credit estimate, 123472.00 points
11:36:41:WU04:FS02:Final credit estimate, 171548.00 points
12:34:40:WU01:FS01:Final credit estimate, 124678.00 points
13:04:55:WU00:FS00:Final credit estimate, 175620.00 points
13:47:35:WU02:FS02:Final credit estimate, 120686.00 points
14:09:51:WU03:FS01:Final credit estimate, 91818.00 points
15:03:17:WU02:FS03:Final credit estimate, 14035.00 points

Fantastic!

This means that it would be easy to write a little perl script to parse the logs and get the points per hour for times the host was actually folding. Even if you only fold 15 hours over a day, you could calculate the points and leave out the gaps.

 

[TennesseeTony]

That is good news for Tom's. I had suggested Ryan Smith contact PURCH and tell them to get Tom's a proper forum. Perhaps this is their response. EDIT: Nope, my message was from the 29th of March...long after that thread was started (17th of March).

In other news...NOOOO! I finally caught my 2080Ti without a WU, first time since enabling 'beta' flag (instead of 'advanced'). Fortunately it only took two 'pause-fold' cycles to pick up a new task.

 

[StefanR5R]

Their team thread only has 3 pages, anandtech is now upto 25 pages just less activity/involvement it seems.

BTW, current "Active users" count according to EOC:

TeAm AnandTech ...... 1,427 (+69▲)
Tom's Hardware ........ 2,455 (-204▼)​

"Active users" are those which received points during the past 7 days.
The change of active users refers to the past 7 days too.

This means that it would be easy to write a little perl script to parse the logs and get the points per hour for times the host was actually folding. Even if you only fold 15 hours over a day, you could calculate the points and leave out the gaps.

That's right. There is a small inconvenience: The timestamps at each line hold only hh:mm:ss (in UTC). The script may need to add in the dates from the "************ Date: yyyy-mm-dd ************" lines also.

 

[Endgame124]

BTW, current "Active users" count according to EOC:

TeAm AnandTech ...... 1,427 (+69▲)​

Tom's Hardware ........ 2,455 (-204▼)​

"Active users" are those which received points during the past 7 days.
The change of active users refers to the past 7 days too.


That's right. There is a small inconvenience: The timestamps at each line hold only hh:mm:ss (in UTC). The script may need to add in the dates from the "************ Date: yyyy-mm-dd ************" lines also.

Not much of an issue at all - I’ve been parsing and filtering log files with Perl for (ugh) 20 years now. I might have some time tonight for a little scripting.

 

[mopardude87]

Starting to have some issues picking up a WU for gpu, hmm. Was doing pretty good here.

 

[StefanR5R]

Eww, there are now project 14378 WUs clogging my 64- and 128-thread slots which do not use more than 40 threads, and give only 6 or 7 % of the PPD of normal WUs.

 

[borandi]

That is good news for Tom's. I had suggested Ryan Smith contact PURCH and tell them to get Tom's a proper forum. Perhaps this is their response. EDIT: Nope, my message was from the 29th of March...long after that thread was started (17th of March).

In other news...NOOOO! I finally caught my 2080Ti without a WU, first time since enabling 'beta' flag (instead of 'advanced'). Fortunately it only took two 'pause-fold' cycles to pick up a new task.

Tom's / AT hasn't been part of Purch for over a year now. Most of Purch was acquired by Future plc in 2018.

 

[mopardude87]

Eww, there are now project 14378 WUs clogging my 64- and 128-thread slots which do not use more than 40 threads, and give only 6 or 7 % of the PPD of normal WUs.

That is weird, hopefully not a issue i run into with the 3900x. I assume such a WU wouldn't like idk load only 20 of the 24 threads for example?

 

[Assimilator1]

I think i'm starting to understand the performance/behavior of folding on the gpu.
It seems there are a lot of pcie transfers and heavy reliance on double precision math.
It uses ~1.8GB and every 5% or so there is a pause as it waits to burst more over the pcie and refill it.
The reliance on doubles means 1/4th or less of the consumer hardware gets used.
On the plus side this makes folding less likely to thermal limit, you can push the clocks/volts a lot higher.
A <2GB working set size would make sense, targeting more hardware. Not as efficient on 8/16GB card however.

This would explain why the vega vii does modest (more dp gflops). And why the newer 5700xt does well (pcie4 and higher core clocks).

Traditionally F@H had been predominantly SP, which in the past (& still?) allowed Nvidia cards to dominate AMD cards (other way around in MW@H & Einstein@H). I believe F@H still mostly uses SP, but DP somewhat more now, people in the F@H bench thread were telling me that anyway .

 

[StefanR5R]

Eww, there are now project 14378 WUs clogging my 64- and 128-thread slots which do not use more than 40 threads, and give only 6 or 7 % of the PPD of normal WUs.

That is weird, hopefully not a issue i run into with the 3900x. I assume such a WU wouldn't like idk load only 20 of the 24 threads for example?

On my 88-thread computers on which I am using two CPU slots, I actually switched the 14378-infected 64-thread slot down to 24 threads (and the previous 24-thread slot with a good project to 64 threads) but I forgot to check whether or not the 14378 WU used all of the 24 threads. Now I don't have such WUs anymore, thankfully.

 

[Markfw]

Only 100 million more difference to make the 1 billion point lead ! 2 days at the current rate, and we have 4, so it should happen.

 

[TennesseeTony]

Early on in this contest, the TeAm was ranked 17th in the world. Then along came a team called PCMR and blew by us at hyper-sonic speed as if we were sitting still.

So, the TeAm now ranks 18th in the world. But that will change in about 12 hours, as we overtake team 'Pleasuredome Community.' Therefore we will soon return to our long held 17th place....which we stole from Tom's a few Decembers ago.

 

[Assimilator1]

Good to hear .

Who are PCMR? (wondering where their firepower comes from).

 

[sswingle]

Early on in this contest, the TeAm was ranked 17th in the world. Then along came a team called PCMR and blew by us at hyper-sonic speed as if we were sitting still.

PC Master Race from Reddit

 

[sswingle]

Good to hear .

Who are PCMR? (wondering where their firepower comes from).

PC Master Race from Reddit

 

[Markfw]

Only 100 million more difference to make the 1 billion point lead ! 2 days at the current rate, and we have 4, so it should happen.

only 81 million to go.... 19 million in 5 hours ? That is gain, not total points.

 

[Ionstream]

How

That is good news for Tom's. I had suggested Ryan Smith contact PURCH and tell them to get Tom's a proper forum. Perhaps this is their response. EDIT: Nope, my message was from the 29th of March...long after that thread was started (17th of March).

In other news...NOOOO! I finally caught my 2080Ti without a WU, first time since enabling 'beta' flag (instead of 'advanced'). Fortunately it only took two 'pause-fold' cycles to pick up a new task.

How'd you guys keep getting GPU WUs? My GPU has been idle for the longest time :/ The CPUs on the other hand are happily cranking away, but their return is honestly rather meh.