XYMer's Home away from Home

When http://bbs.xlr8yourmac.com is down (i.e. always)
It is currently Sun Jun 25, 2017 12:40 am

All times are UTC - 8 hours




Post new topic Reply to topic  [ 936 posts ]  Go to page Previous  1 ... 34, 35, 36, 37, 38  Next
Author Message
PostPosted: Fri Mar 31, 2017 9:29 pm 
Offline
Benevolent Dictator
User avatar

Joined: Mon Apr 21, 2008 2:03 am
Posts: 14484
Ahh, that explains it, thanks.


Top
 Profile  
 
PostPosted: Tue Apr 04, 2017 12:08 pm 
Offline
User avatar

Joined: Thu May 15, 2008 8:13 pm
Posts: 9269
Location: Caught between the moon and NYC
I got the GTX1050 installed.
Code:
C:\Program Files (x86)\distributed.net>dnetc -bench rc5-72

distributed.net client for CUDA 3.1 on Win32 Copyright 1997-2011, distributed.net
Please visit http://www.distributed.net/ for up-to-date contest information.
Start the client with '-help' for a list of valid command line options.


dnetc v2.9110-519-CTR-11041423 for CUDA 3.1 on Win32 (WindowsNT 6.1).
Please provide the *entire* version descriptor when submitting bug reports.
The distributed.net bug report pages are at http://bugs.distributed.net/

[Apr 04 19:40:25 UTC] nvcuda.dll Version: 6.14.13.7892
[Apr 04 19:40:26 UTC] RC5-72: using core #0 (CUDA 1-pipe 64-thd).
[Apr 04 19:40:33 UTC] RC5-72: Benchmark for core #0 (CUDA 1-pipe 64-thd)
                      0.00:00:04.55 [956,778,921 keys/sec]
[Apr 04 19:40:33 UTC] RC5-72: using core #1 (CUDA 1-pipe 128-thd).
[Apr 04 19:40:40 UTC] RC5-72: Benchmark for core #1 (CUDA 1-pipe 128-thd)
                      0.00:00:04.46 [977,856,488 keys/sec]
[Apr 04 19:40:40 UTC] RC5-72: using core #2 (CUDA 1-pipe 256-thd).
[Apr 04 19:40:51 UTC] RC5-72: Benchmark for core #2 (CUDA 1-pipe 256-thd)
                      0.00:00:08.03 [542,017,904 keys/sec]
[Apr 04 19:40:51 UTC] RC5-72: using core #3 (CUDA 2-pipe 64-thd).
[Apr 04 19:40:57 UTC] RC5-72: Benchmark for core #3 (CUDA 2-pipe 64-thd)
                      0.00:00:04.47 [973,362,365 keys/sec]
[Apr 04 19:40:57 UTC] RC5-72: using core #4 (CUDA 2-pipe 128-thd).
[Apr 04 19:41:08 UTC] RC5-72: Benchmark for core #4 (CUDA 2-pipe 128-thd)
                      0.00:00:08.03 [550,986,142 keys/sec]
[Apr 04 19:41:08 UTC] RC5-72: using core #5 (CUDA 2-pipe 256-thd).
[Apr 04 19:41:18 UTC] RC5-72: Benchmark for core #5 (CUDA 2-pipe 256-thd)
                      0.00:00:08.00 [551,630,795 keys/sec]
[Apr 04 19:41:18 UTC] RC5-72: using core #6 (CUDA 4-pipe 64-thd).
[Apr 04 19:41:29 UTC] RC5-72: Benchmark for core #6 (CUDA 4-pipe 64-thd)
                      0.00:00:08.03 [541,692,891 keys/sec]
[Apr 04 19:41:29 UTC] RC5-72: using core #7 (CUDA 4-pipe 128-thd).
[Apr 04 19:41:39 UTC] RC5-72: Benchmark for core #7 (CUDA 4-pipe 128-thd)
                      0.00:00:08.00 [551,630,795 keys/sec]
[Apr 04 19:41:39 UTC] RC5-72: using core #8 (CUDA 4-pipe 256-thd).
[Apr 04 19:41:50 UTC] RC5-72: Benchmark for core #8 (CUDA 4-pipe 256-thd)
                      0.00:00:08.00 [558,189,217 keys/sec]
[Apr 04 19:41:50 UTC] RC5-72: using core #9 (CUDA 1-pipe 64-thd busy wait).
[Apr 04 19:41:57 UTC] RC5-72: Benchmark for core #9 (CUDA 1-pipe 64-thd busy wait)
                      0.00:00:04.54 [961,340,610 keys/sec]
[Apr 04 19:41:57 UTC] RC5-72: using core #10 (CUDA 1-pipe 64-thd sleep 100us).
[Apr 04 19:42:16 UTC] RC5-72: Benchmark for core #10 (CUDA 1-pipe 64-thd sleep 100us)
                      0.00:00:16.05 [268,709,907 keys/sec]
[Apr 04 19:42:16 UTC] RC5-72: using core #11 (CUDA 1-pipe 64-thd sleep dynamic).
[Apr 04 19:42:26 UTC] RC5-72: Benchmark for core #11 (CUDA 1-pipe 64-thd sleep dynamic)
                      0.00:00:08.06 [536,901,254 keys/sec]
[Apr 04 19:42:26 UTC] RC5-72 benchmark summary :
                      Default core : #0 (CUDA 1-pipe 64-thd)
                      Fastest core : #1 (CUDA 1-pipe 128-thd)
[Apr 04 19:42:26 UTC] Core #1 is marginally faster than the default core.
                      Testing variability might lead to pick one or the other.
So it's a big jump. Beats the 7750 by 50%. 'course the 7750 is nearly 4 years old at this point and also cost $100.

I found both my displayport and hdmi to dvi cable. Unfortunately it looks like the U2211H has a problem with sleeping over displayport regardless of whether the image on screen is actively being used or not... the timer starts from the last time you touched the on screen controls and when it expires the display sleeps until you touch the on screen controls again. DVI doesn't have this issue. Dell's displayport monitors are really kind of sad, they stuck displayports on all their displays but didn't bother to test it at all. My 24" at home doesn't even work over DP at all, period. So once I get some free time I'll be flip-flopping it over to the HDMI-DVI cable. Some day I'll be able to use you, full size Displayport cable... some day...

I'm running some additional -bench tests to try and work out whether 0 or 1 is the fastest. If I let it auto-detect the core it ends up at #11 which is... bad...


Top
 Profile  
 
PostPosted: Tue Apr 04, 2017 1:26 pm 
Offline
Benevolent Dictator
User avatar

Joined: Mon Apr 21, 2008 2:03 am
Posts: 14484
Wow, quite impressive! :D


Top
 Profile  
 
PostPosted: Tue Apr 04, 2017 2:57 pm 
Offline
User avatar

Joined: Thu May 15, 2008 8:13 pm
Posts: 9269
Location: Caught between the moon and NYC
It looks good for $100, I was just hoping for a bit more given the 7750. However when used as a 3D card it should be substantially better - NVidia seems to regard GPGPU as an afterthought.

So far I've had 2 instances where #0 was the fastest core, 5 instances where #1 was the fastest core, and one instance where #3 was the fastest core. I'm thinking #1 is the way to go.

I'll have to fire up the OpenCL client and see if it wants to play ball. I installed the latest drivers last week after I ordered this card (and verified both cards used the same driver installer), so installation was a breeze. Aside from displayport.

Edit: Dammit, since I wrote this I hooked up the 2nd monitor via HDMI/DVI and #3 got 4 wins in a row, so #1 and #3 are tied.


Top
 Profile  
 
PostPosted: Tue Apr 04, 2017 3:09 pm 
Offline
User avatar

Joined: Thu May 15, 2008 8:13 pm
Posts: 9269
Location: Caught between the moon and NYC
Code:
C:\Projects\dnetc>dnetc.com -bench rc5-72

distributed.net client for OpenCL on Win32 Copyright 1997-2016, distributed.net
Please visit http://www.distributed.net/ for up-to-date contest information.
Start the client with '-help' for a list of valid command line options.

dnetc v2.9112-521-CTR-16021318 for OpenCL on Win32 (WindowsNT 6.1).
Please provide the *entire* version descriptor when submitting bug reports.
The distributed.net bug report pages are at http://bugs.distributed.net/

[Apr 04 23:07:19 UTC] RC5-72: using core #0 (CL ANSI 1-pipe).
[Apr 04 23:07:25 UTC] RC5-72: Benchmark for core #0 (CL ANSI 1-pipe)
                      0.00:00:03.27 [1,337,897,665 keys/sec]
[Apr 04 23:07:25 UTC] RC5-72: using core #1 (CL 1-pipe).
[Apr 04 23:07:31 UTC] RC5-72: Benchmark for core #1 (CL 1-pipe)
                      0.00:00:02.94 [1,499,093,773 keys/sec]
[Apr 04 23:07:31 UTC] RC5-72: using core #2 (CL 2-pipe).
[Apr 04 23:07:37 UTC] RC5-72: Benchmark for core #2 (CL 2-pipe)
                      0.00:00:03.07 [1,443,160,475 keys/sec]
[Apr 04 23:07:37 UTC] RC5-72: using core #3 (CL 4-pipe).
[Apr 04 23:07:42 UTC] RC5-72: Benchmark for core #3 (CL 4-pipe)
                      0.00:00:03.02 [1,462,426,654 keys/sec]
[Apr 04 23:07:42 UTC] RC5-72 benchmark summary :
                      Default core : #-1 (undefined) 0 keys/sec
                      Fastest core : #1 (CL 1-pipe) 1,499,093,773 keys/sec
[Apr 04 23:07:43 UTC] Compare and share your rates in the speeds database at
                      http://www.distributed.net/speed/
                      (benchmark rates are for a single processor core)
Well that was unexpected.


Top
 Profile  
 
PostPosted: Tue Apr 04, 2017 3:52 pm 
Offline
User avatar

Joined: Thu May 15, 2008 8:13 pm
Posts: 9269
Location: Caught between the moon and NYC
Code:
            0                1                2                3
Result 1    1,355,379,804    1,443,698,228    1,449,158,204    1,454,476,251
Result 2    1,337,513,193    1,429,161,857    1,436,142,433    1,421,233,337
Result 3    1,340,472,317    1,434,707,298    1,440,482,744    1,434,291,010
Result 4    1,331,597,903    1,413,895,873    1,435,820,462    1,409,175,539
Result 5    1,355,854,326    1,441,630,933    1,447,072,278    1,588,342,179
Result 6    1,340,485,961    1,439,648,559    1,426,497,406    1,417,767,294
Result 7    1,325,760,130    1,428,437,000    1,426,254,935    1,440,231,418
Result 8    1,349,328,422    1,443,908,662    1,450,453,641    1,420,705,203
Result 9    1,337,897,665    1,499,093,773    1,443,160,475    1,462,426,654
Result 10    1,358,402,267    1,438,476,490    1,466,884,049    1,443,107,025

Average    1,343,269,199    1,441,265,867    1,442,192,663    1,449,175,591
God bless spreadsheets and averaging functions.


Top
 Profile  
 
PostPosted: Tue Apr 04, 2017 4:31 pm 
Offline
Benevolent Dictator
User avatar

Joined: Mon Apr 21, 2008 2:03 am
Posts: 14484
Ah, much better. :)


Top
 Profile  
 
PostPosted: Tue Apr 04, 2017 4:37 pm 
Offline
User avatar

Joined: Thu May 15, 2008 8:13 pm
Posts: 9269
Location: Caught between the moon and NYC
That's actually in line with what I was expecting.

In two years OpenCL went from being about 75% the speed of CUDA to 150% the speed of CUDA.

And all I had to sacrifice was $100 and a DVI port.


Top
 Profile  
 
PostPosted: Wed Apr 05, 2017 3:34 pm 
Offline
User avatar

Joined: Thu May 15, 2008 8:13 pm
Posts: 9269
Location: Caught between the moon and NYC
My 480 is definitely faster "cold" and slows down once it gets cranking. When cooled down and running -bench it averages 3800Mkeys/s. I did a similar spreadsheet to verify with current drivers my core selection was still good, and it was, but the key rate was GT740 higher in bench vs. full load.

Which is why its a shame I can't leave the fans pegged at 100%, it might be able to run full speed and stay at cooler temps. Right now I have it configured for a target of 45 degrees with a max of 50 degrees and it floats between 45 & 50 while running dnetc. When the temp hits 40 the fans shut off.

Sometimes I really &*(#$&*(#$ people and their noise goblins. Its like a fan killed their parents, so after they grew up they roam the city at night, seeking vengeance and retribution on noisy fans.


Last edited by MonkeyBoy on Wed Apr 05, 2017 3:37 pm, edited 1 time in total.

Top
 Profile  
 
PostPosted: Wed Apr 05, 2017 3:37 pm 
Offline
Benevolent Dictator
User avatar

Joined: Mon Apr 21, 2008 2:03 am
Posts: 14484
Why can't you run the fans faster?


Top
 Profile  
 
PostPosted: Wed Apr 05, 2017 3:38 pm 
Offline
User avatar

Joined: Thu May 15, 2008 8:13 pm
Posts: 9269
Location: Caught between the moon and NYC
I might be able to if I had the time to figure out how to flash a new image onto the card. With the stock image when the temp hits 40 the fans shut off. So I try to keep temps close to 40 but without actually hitting 40.

When I'm playing a game or doing anything 3D related the fans are running so I'm just not quite sure why the fans ever need to shut off, except to exist as a checkmark on someone's annoying checklist.


Top
 Profile  
 
PostPosted: Wed Apr 05, 2017 4:04 pm 
Offline
Benevolent Dictator
User avatar

Joined: Mon Apr 21, 2008 2:03 am
Posts: 14484
Do the fans only have 2 pins & not adjustable?

MSI Afterburner can't control the fan(s)?

https://www.msi.com/page/afterburner


Top
 Profile  
 
PostPosted: Wed Apr 05, 2017 4:57 pm 
Offline
User avatar

Joined: Thu May 15, 2008 8:13 pm
Posts: 9269
Location: Caught between the moon and NYC
Yeah I might be able to use MSI afterburner, but I've been burned by it in the past. It's like Norton Antivirus... after you install Norton your system exists to run Norton, any use you may get out of it is incidental. Most of what afterburner does was basically rolled into AMD's drivers ages ago, but they depend on the card's firmware to a great extent. Afterburner may operate the same way, it's been a long time since I tried it (ever since overclocking was placed in the OEM drivers there's been little need - my fans on the 7750, 6950, and 280 all ran/run at whatever speed I set them to run at).

I'm really loathe to install more software, I'm starting to trim back installed software to stop the hangover at startup from growing longer. One of these days I need to wander through the non-PnP drivers list and pare them back by hand, which is dangerous business and best done after a backup.

There's a certain amount of software I need for work, I can't just nuke it all and optimize the box for games and GPGPU, no matter how much better that solution might be for games & GPGPU.


Top
 Profile  
 
PostPosted: Tue Apr 25, 2017 1:52 pm 
Offline
User avatar

Joined: Thu May 15, 2008 8:13 pm
Posts: 9269
Location: Caught between the moon and NYC
The RX 580 is out. Basically just the RX 480 with heat issues resolved so it can run at faster clocks. Some people have been flashing the RX 580 firmware on top of the RX 480, but since the 580 has a faster clock it could brick the 480 (if that particular 480 can't run at the faster speed, which is true of many early 480s). Some of the changes were implemented in silicon so it's not entirely clear what, if anything, is gained.

Looks like they dropped the 580 at the original 480 prices, so $200 for 4GB and $230 for 8GB card. The RX 480 has been steadily moving down in price, though you can get some variants of the 480 8GB for more than the cheapest 580 8GB card.

I'm currently in the throes of moving at work so the 1050 is down. I have no idea when I'll have enough time to get my office up and running again, I have an unholy amount of equipment to get installed and working before next Monday. And some of it they won't let me touch until Friday morning! Whee!


Top
 Profile  
 
PostPosted: Wed Apr 26, 2017 4:24 pm 
Offline
User avatar

Joined: Thu May 15, 2008 8:20 pm
Posts: 2247
MonkeyBoy wrote:
I'm currently in the throes of moving at work so the 1050 is down. I have no idea when I'll have enough time to get my office up and running again, I have an unholy amount of equipment to get installed and working before next Monday. And some of it they won't let me touch until Friday morning! Whee!
thanks for keeping us up to date on the new RX cards!

are you moving your office in the same building or moving buildings?


Top
 Profile  
 
PostPosted: Wed Apr 26, 2017 4:46 pm 
Offline
User avatar

Joined: Thu May 15, 2008 8:13 pm
Posts: 9269
Location: Caught between the moon and NYC
We have two sites, and essentially we're consolidating into a single building.

I get a window office this time, whee.

That also means I need to invest in sweaters for the days when its cold outside and fans for the days when its hot outside, because window offices have the worst heating/cooling.

As you may imagine, the people who tend to gravitate towards window offices are the least willing to just buck up and deal with it. And they call me, because someone apparently slapped an HVAC tag on all my shirts while I wasn't looking. :lol:


Top
 Profile  
 
PostPosted: Thu Apr 27, 2017 6:04 am 
Offline
User avatar

Joined: Thu May 15, 2008 8:20 pm
Posts: 2247
congrats on the window office! that can be nice to get some natural light, especially in the winter. you should get a plant :P


Top
 Profile  
 
PostPosted: Tue May 09, 2017 10:45 am 
Offline

Joined: Thu Dec 13, 2012 8:01 pm
Posts: 778
Just posting over here because it seems like I wasn't getting the latest posts from this thread . . . but, also, as I have been getting some attention from the forum lately I tried to run some dnetc time for the team . . . which should have been uploaded/posted Sunday evening west coast 6pm . . . checking back this am (Tues) not only are the recent ~173 stats not posted, but there should have been another time set posted from earlier in 2017, but it shows only back to Dec 3 '16 . . . . ???

It seemed to have the record of each of the 24 stats that it took my PowerMac many hours to kick out listed in 2016, and then once or twice with what should be the new MPro late in 2016 . . . of somewhere in the 170 stats area that takes the same amount of time to do as the PowerMac . . . but, then after that . . . nothing. I'm pretty sure I ran a set sometime in Jan/FebMarch?? . . . and then, over the recent week a few hours at a time . . . not received??

Nothing new as far as that goes??

e.e.e.


Top
 Profile  
 
PostPosted: Tue May 09, 2017 10:52 am 
Offline
Benevolent Dictator
User avatar

Joined: Mon Apr 21, 2008 2:03 am
Posts: 14484
Check the dnetc.ini on the one you think didn't submit, is the account listed right? If it's blank, they just go in, but nowhere in particular.


Top
 Profile  
 
PostPosted: Tue May 09, 2017 11:43 am 
Offline

Joined: Thu Dec 13, 2012 8:01 pm
Posts: 778
BDAqua wrote:
Check the dnetc.ini on the one you think didn't submit, is the account listed right? If it's blank, they just go in, but nowhere in particular.


OK, thanks . . . away from that computer right now, so I'll have to check the file later . . . when I'm hot-wiring the PM HD and fiddling with that I'll check and see what's showing on the list . . .

PS: So, if they "just go in" would they just go into the team's stats . . . or just into the great universe of stats??

e..


Top
 Profile  
 
PostPosted: Tue May 09, 2017 3:29 pm 
Offline
User avatar

Joined: Thu May 15, 2008 8:13 pm
Posts: 9269
Location: Caught between the moon and NYC
It gets credited to whatever email address is in dnetc.ini. If it was left at default then it's credited to whatever groups that default address is in (typically none).


Top
 Profile  
 
PostPosted: Tue May 09, 2017 3:41 pm 
Offline
User avatar

Joined: Thu May 15, 2008 8:20 pm
Posts: 2247
eep, if you take a long break between doing work, i recommend deleting the buff-in.r72 file. sometimes the work you downloaded will have got completed by someone else and won't count. by deleting the buff-in.r72 file, the client will be forced to recreate the the buff-in.r72 file, and you will be working with fresh work units from that point.


Top
 Profile  
 
PostPosted: Wed May 10, 2017 2:23 pm 
Offline

Joined: Thu Dec 13, 2012 8:01 pm
Posts: 778
mc68k wrote:
eep, if you take a long break between doing work, i recommend deleting the buff-in.r72 file. sometimes the work you downloaded will have got completed by someone else and won't count. by deleting the buff-in.r72 file, the client will be forced to recreate the the buff-in.r72 file, and you will be working with fresh work units from that point.



@MB & mc68k:

Interestingly the notifications aren't picking up on this thread still . . . anyway, @MB . . . I checked yesterday by firing dnetc and it showed my email address . . . but possibly this problem relates to what "mc68k" is referring to . . . as often indeed days or weeks goes by between running sessions . . . . So the genius of this approach is that the buff-out file remains to be counted . . . but it then has to recalc the "remaining" stats in buff-in after it is deleted . . . ??? Something to try, because my habit is to only run a few hours . . . to build up to the 172 stats or so that seems to be the "flush" point . . . .

e.e.p.


Top
 Profile  
 
PostPosted: Wed May 10, 2017 3:12 pm 
Offline
User avatar

Joined: Thu May 15, 2008 8:13 pm
Posts: 9269
Location: Caught between the moon and NYC
If you wait too long what's in buff-out won't get counted either. The servers have no way of knowing what blocks you've crunched vs. haven't crunched, all they know is you've checked out a set of blocks. Until you check those blocks in they're assigned to you. Periodically they go through and see blocks that have been assigned but not worked on and just re-assign those blocks as available to be assigned to others. I imagine, given the workspace of RC5-72, they only have a set number of blocks that can be checked out at any given time, and keeping those blocks checked out in perpetuity would yield an unworkable amount of data, since the checked out blocks would grow and grow and grow... So they have, say, 100,000 blocks that can be checked out at any given time. When someone tries to request block 100,001 they look at the oldest block that was checked out and invalidate it.

This is all in theory of course, I haven't read a .plan that explains what their process is. I think that's by design, because there are a crazy amount of people trying to rig the system, basically submitting fake blocks to increase their stats count. By giving them more information the miscreants will adjust tactics to try and game the system in new ways. By keeping it unwritten they can change tactics as they see fit to play cat & mouse with the miscreants.

Personally I have my systems run a second -update process once an hour so the running dnetc process always has blocks to work on. The pause while the active process is fetching & flushing causes a measurable dip, at least when your GPU is crunching 3.6 billion keys a second.


Top
 Profile  
 
PostPosted: Wed May 10, 2017 4:15 pm 
Offline

Joined: Thu Dec 13, 2012 8:01 pm
Posts: 778
MonkeyBoy wrote:

Personally I have my systems run a second -update process once an hour so the running dnetc process always has blocks to work on. The pause while the active process is fetching & flushing causes a measurable dip, at least when your GPU is crunching 3.6 billion keys a second.


OK . . . I wish that my GPU would be kicking out stats big time, but, apparently I'm sloughing along with my cpu . . . one processor at a time. How is that "second- update" time set up?

e.e.p


Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 936 posts ]  Go to page Previous  1 ... 34, 35, 36, 37, 38  Next

All times are UTC - 8 hours


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
cron
Powered by phpBB® Forum Software © phpBB Group