LTT Folding Team's Emergency Response to Covid-19

[HaRdLy007]

6 minutes ago, Alex Atkin UK said:

See a few pages back.

Any idea how far back?

 

I have an all regions quota of 4, but can only run 1 GPU per instance, not sure why.

[biohazard911]

Got all excited to upgrade my dedicated folding rig, Ryzen 5 1600AF, got a great deal on an MSI B450 Tomahawk from newegg.com, 16 gigs of Gskill Aegis DDR4 3200 CL16, and a nice little Silicon Power NVME SSD. Gonna pair it with my Strix Vega 56 and my RX 470 to give me a nice little Performance Per Dollar setup.....But I got almost done with the build, went to put on the wraith stealth.....and well lets see if you can spot whats missing on the board.  

 

Newegg wont replace it as its out of stock, only way to return it is to print a shipping label....on the printer I don't have.....they refuse to mail me one knowing I have no way to get the label printed. 

 

So much for getting it running before the end of the competition, ordered a backplate off amazon which wont come till next Monday.  

 

[Ayren]

1 minute ago, HaRdLy007 said:

Any idea how far back?

 

I have an all regions quota of 4, but can only run 1 GPU per instance, not sure why.

if youre talking about aws G instances only the 12xlarge have multiple GPUs iirc?

[Ayren]

1 minute ago, biohazard911 said:

Got all excited to upgrade my dedicated folding rig, Ryzen 5 1600AF, got a great deal on an MSI B450 Tomahawk from newegg.com, 16 gigs of Gskill Aegis DDR4 3200 CL16, and a nice little Silicon Power NVME SSD. Gonna pair it with my Strix Vega 56 and my RX 470 to give me a nice little Performance Per Dollar setup.....But I got almost done with the build, went to put on the wraith stealth.....and well lets see if you can spot whats missing on the board.  

 

Newegg wont replace it as its out of stock, only way to return it is to print a shipping label....on the printer I don't have.....they refuse to mail me one knowing I have no way to get the label printed. 

 

So much for getting it running before the end of the competition, ordered a backplate off amazon which wont come till next Monday.  

 

just use zipties

[ObeliskAG]

9 hours ago, Duffbro said:

More than 2 ExaFLOPS baby ....whoooohooooo

Supercomputer performance is measured doing double-precision LINPACK benchmarks, not single-precision claculations, so comparing Folding@Home to supercomputers isn't exactly apples-to-apples. It would be really interesting to see a real apples-to-apples comparison.

[HaRdLy007]

1 minute ago, Ayren said:

if youre talking about aws G instances only the 12xlarge have multiple GPUs iirc?

I'm using Google.

[Gorgon]

2 hours ago, ShortRouter said:

I just have the web terminal window always open. Same for 2 other cloud providers I'm testing. You can do

watch -n 0.5 nvidia-smi

to refresh the nvidia-smi output every 0.5 seconds. (exit with ctrl+c)

The remote control thing from FAH uses Telnet which is very insecure. Use SSH whenever possible. (Google Cloud VM + SSH client requires advanced configuration of iptables, using static ip and maybe more).

nvidia-smi has a cycle stats built in - no need to use watch

 

Another option is install nvtop

 

My personal favorite is:

nvidia-smi -i 0 -l 1 --format=csv,noheader --query-gpu=temperature.gpu,power.draw,clocks.current.sm,fan.speed

[benlinf]

Anyone else having problems uploading finished WUs?

Have lots of the followingin the log:

 

21:02:58:WARNING:WU00:FS01:Exception: Failed to send results to work server: Failed to connect to 140.163.4.231:80: A connection attempt failed because the connected party did not properly respond after a period of time, or established connection failed because connected host has failed to respond.

 

It's been stuck for several hours now.

Have quit and restarted the client also rebooted computer.

 

Any help much appreciated!

[Alex Atkin UK]

17 minutes ago, HaRdLy007 said:

Any idea how far back?

 

I have an all regions quota of 4, but can only run 1 GPU per instance, not sure why.

I believe you have to request a quota increase for the specific GPU you want.

[ShortRouter]

13 minutes ago, ObeliskAG said:

Supercomputer performance is measured doing double-precision LINPACK benchmarks, not single-precision claculations, so comparing Folding@Home to supercomputers isn't exactly apples-to-apples. It would be really interesting to see a real apples-to-apples comparison.

you mean single precision?

[ShortRouter]

18 minutes ago, HaRdLy007 said:

Any idea how far back?

 

I have an all regions quota of 4, but can only run 1 GPU per instance, not sure why.

i posted this exact solution but I'll post again: https://stackoverflow.com/questions/45227064/how-to-request-gpu-quota-increase-in-google-cloud

[HaRdLy007]

3 minutes ago, Alex Atkin UK said:

I believe you have to request a quota increase for the specific GPU you want.

Thanks!

 

2 minutes ago, ShortRouter said:

i posted this exact solution but I'll post again: https://stackoverflow.com/questions/45227064/how-to-request-gpu-quota-increase-in-google-cloud

Thank you!

[wcorbe]

We are #strongler when we work together.

[greblos]

Well this is interesting.  The OS stats screen shows 6 AMD GPUs under Mac OS now.  Considering there has been no Mac OS GPU support thus far, I wonder if there's a new development, or if this is an anomaly.

[Metallus97]

5 minutes ago, greblos said:

Well this is interesting.  The OS stats screen shows 6 AMD GPUs under Mac OS now.  Considering there has been no Mac OS GPU support thus far, I wonder if there's a new development, or if this is an anomaly.

-snip-

Interesting! I guess anomaly but would love to get my 5500M to work!

[Tills_Monocle]

11 hours ago, Zonker said:

It's crazy competitive at that level. Be lucky not to lose spots, let alone pull ahead. Gotta be a time when you hit your hardware peak.

I'm getting close to it, but I'm still spending a few hours most days with no work units. I might lose spots today, i had two work units assigned to a collection server last night that isn't taking any WUs. They are still waiting to send.

[Alex Atkin UK]

8 minutes ago, greblos said:

Well this is interesting.  The OS stats screen shows 6 AMD GPUs under Mac OS now.  Considering there has been no Mac OS GPU support thus far, I wonder if there's a new development, or if this is an anomaly.

Hackintosh machines using https://amd-osx.com/?

[Metallus97]

2 minutes ago, Alex Atkin UK said:

Hackintosh machines using https://amd-osx.com/

And they will run GPU? Why would a normal mac not run GPU then?
I think there is no C21 or C22 available for macOS... well maybe they got something in the works.

But this amd-osx will be my next quarantine project UH YEHA

[Alex Atkin UK]

Just now, Metallus97 said:

And they will run GPU? Why would a normal mac not run GPU then?
I think there is no C21 or C22 available for macOS... well maybe they got something in the works

Oh GPUs, duh, don't mind me, I've only had 5 hours sleep.

[ObeliskAG]

42 minutes ago, ShortRouter said:

you mean single precision?

Sure. From the Wikipedia page for Summit (my bolding):

"Summit is the first supercomputer to reach exaflop (a quintillion operations per second) speed, achieving 1.88 exaops during a genomic analysis and is expected to reach 3.3 exaops using mixed precision calculations.^"

[Zonker]

15 minutes ago, Tills_Monocle said:

I'm getting close to it, but I'm still spending a few hours most days with no work units. I might lose spots today, i had two work units assigned to a collection server last night that isn't taking any WUs. They are still waiting to send.

As long as we make a difference. Team effort.

[Tills_Monocle]

17 hours ago, Pasi123 said:

GTX 770 is still more powerful than any of the graphics cards that I have now

 

I use GTX 960, GTX 760, GTX 660, GTX 470, GT 640 and GT 430. And CPUs that I use are Xeon X5670 6c/12t 4.2GHz, i5-3450 4c/4t 3.7GHz, X5677 4c/8t 3.6GHz, i3-4150 2c/4t and i5-3210M 2c/4t.

Right now I'm not folding with the X5670/GTX960 and i5-3450/GTX760/GT430 because I'm using them for something else, but I'll be folding with them soon.

 

I've been folding since March 15th and I have a bit over 3.6 million points

I would offer to send you the GTX 770 but it had a small fire in 2017  I dust my PC much more regularly now.

[ShortRouter]

28 minutes ago, ObeliskAG said:

Sure. From the Wikipedia page for Summit (my bolding):

"Summit is the first supercomputer to reach exaflop (a quintillion operations per second) speed, achieving 1.88 exaops during a genomic analysis and is expected to reach 3.3 exaops using mixed precision calculations.^"

From my understanding it's a mix of FP16 (half) and FP32 (single) precision then. Maybe that's why the FAH stats page lists both.

"Mixed precision is the use of both 16-bit and 32-bit floating-point types in a model during training to make it run faster and use less memory."

[ObeliskAG]

15 minutes ago, ShortRouter said:

From my understanding it's a mix of FP16 (half) and FP32 (double) precision then. Maybe that's why the FAH stats page lists both.

"Mixed precision is the use of both 16-bit and 32-bit floating-point types in a model during training to make it run faster and use less memory."

FP32 is not double precision:

16-bit: Half
32-bit: Single
64-bit: Double
40-bit or 80-bit Extended-precision formats

https://en.wikipedia.org/wiki/IEEE_754#2019

 

If Folding@Home used any GPU double-precision, my AMD R9 390X would be killing it, instead of being the slowest card.

[ShortRouter]

@ObeliskAGtypo, I meant to write single lol. T'is getting late over here in Europe.