RC-5/72 -- Two more days - Team Anandtech moves up

I might throw a little power at this and see what my new rx460 can do. I think I tried to attach to moowrap on boinc and it immediately would error. I know moowrap is using an older version, so will see if a new client will work for me.

 

Is this app able to use double precision on old ATI cards like the 5870?

 

Single precision only.
They have a client speed table on their site if you are wondering how fast your card is. Note that there are not many newer cards listed.
http://cgi.distributed.net/speed/query.php?cputype=all&arch=4&contest=all&multi=3

 

Such a shame, I came into a 5870 and a 5970 which is a real beast at DP, but there does not seem to be a way to put them to use.

 

For what it's worth, AMD FirePro W7000, a GCN 1.0 card, works without any apparent problem. I tried Moo!Wrapper and auto-selected client application (distributed.net client 1.04, opencl_ati_101). This is on Windows 7 with Radeon Pro 16.12.1 Crimson Relive driver.

A 6144-points task takes 34 minutes on this card, which works out to 260 000 PPD if working fulltime. I have no idea how this translates into blocks per day.
(W7000 = Pitcairn XT GL, 1280 shaders @950 MHz, very similar to R9 270, or downclocked HD 7870)

 

I believe that Milkyway@home and Primegrid GFN21/GFN22 are the only gpu applications from BOINC projects that currently require double precision. And the last time I checked, Folding@home leaned much more heavily on single precision hardware.

 

Einstein isn't dependent on double precision? One way to find out is...I'm tossing some Einstein to the Nvidia's and see how they compare to the R9-280X fleet....

 

Thanks, I was having trouble getting the old 5870 to work on M@H, but with the right driver, it is finally detected and crunching away!

 

Shocked! 90% done with some Einstein tasks on GTX 1070 Nvidia cards in only 11 minutes!!!!!

Definitely single precision! (versus 30-ish minutes on R9-280X, which has an extremely high 1/4 Double Precision to Single Precision rate).

And now 100% done, in only 11.75 minutes.

EDIT: Crap, that was one task at a time versus 3 at a time for the AMD card, so scratch my conclusion above for a few more minutes....

 

Getting the number of concurrent tasks right is something that isn't always intuitive. My Fury was not always saturated with two tasks, so I gave it three, and it slowed to a crawl! I got it before rediscovering DC, had I known, hell, I'd have stuck with my old 7970.

 

(Sorry to hi-jack the thread LANMAN)

GTX1080 running three concurrent Einstein in 25.8 minutes each, versus 27.5 minutes for 3 concurrent tasks on a R9-280X. So yeppers, Einstein is single precision. I guess I thought it was double precision due to how well the ancient 280X can hold it's own on that project.

Back to the subject: Looks like a few more easy targets have popped up for an easy overtake within a week or two.

 

In case anybody else joined RC5-72 via Moo!Wrapper recently: Joining the wrapper's TeAm AnandTech does not automatically join you with distributed.net's team AnandTech 10635. To do so,

• complete one or more WUs,
  
• go to to http://moowrap.net/ -> "Your account" -> "Moo! Wrapper preferences", and there look up your Distributed.net ID,
  
• go to the dnet RC5-72 stats page at http://stats.distributed.net/projects.php?project_id=8 and enter your Distributed.net ID into the participant search box: this should bring up your personal stats page,
  
• at your personal stats page, click "Please email me my password",
• wait for the mail to arrive, then check your profile at http://stats.distributed.net/participant/pedit.php,
  
• go to the AnandTech 10635 stats page http://stats.distributed.net/team/tmsummary.php?project_id=8&team=10635, click "I want to join this team".

Edit, May 13: inserted 1st and 2nd step

 

I asked in another thread about you systems before reading this. So the 1080's must have just come online for Einstein and that's why they appear to be running behind the 280's

 

How are these cards doing, I am unfamiliar with them?

 

Where should the mail be going? My moowrap.net email? Because it seems like it didn't.

 

Yes, the e-mail address of your account at moowrap.net is being used by the wrapper to identify your completed work at distributed.net. After you completed at least one WU, account creation at distributed.net happens transparently in the same way as if you would use the original distributed.net client instead of the wrapper: distributed.net auto-generates a password, and you can get this password mailed to the address that the client used in reports of complete work.

Anywhere from 0 to 24 hours after your client first flushes some work units back to us, our stats server will detect that you are a new participant and start publishing stats about your work as well as your ranking on its web site. [...] If you enter your full email address or if the search based on your query only returns one result, you will be directly directed to your personal stats page. [...] Finally, you will need to request that your stats password be sent to you by email to the address that you specified in your email client configuration, which is also the one under which you appear in the stats server. To do this, you must be on your (or one of your) personal stats page(s) [at distributed.net] and click on the link at the very bottom of that page that reads: "I cannot remember my password. Please email [insert your email address here]'s password.". Once you click on that link, you should receive an email with your stats password shortly thereafter. (source)​

Maybe a spam filter caught it.

 

Could it be using your old e-mail if you were previously using the dnet client?
Maybe a new user was created and you are now on two accounts?

 

Both the accounts I can find for myself haven't submitted work since 2011. So I guess I can't find myself at distributed.net. Can you?

 

On third thought, I think I did find myself, but I seem to have a problem in the stats: https://stats.distributed.net/participant/psummary.php?project_id=28&id=342597

 

Did you first check your current e-mail address at moowrap.net, then searched for this e-mail address (not for the nick name) at stats.distributed.net?

Anyway, when you click "RC5-72" at the bottom of this page, you get to https://stats.distributed.net/participant/psummary.php?project_id=8&id=342597 which looks better from here.

 

Yep, but I don't run moowrap.

No I am not finding you. Am at a loss as to where your points would be going.

 

Waving to @Ken g6 as he rushes past my humble 30 k blocks/day.

 

Oh, hey, I stopped because I didn't think my stats were registering! I guess I'll start up again soon.