News:

If you are a member of the Team on BOINC you still need to register on this forum to see the member posts.  The posts available for visitors are not posted to much by members.
 Remember to answer the questions when Registering and also you must be a active member of Team BOINC@AUSTRALIA on BOINC.

Main Menu

Project Update

Started by Dingo, January 30, 2016, 12:48:05 AM

Previous topic - Next topic

Dataman

Quote from: tazzduke on June 08, 2017, 06:58:03 AM
Hey DM

I gather that is running two tasks at a time??

Regards
No the GPU utilization is 1.00 on all of them. I tried 0.50 on the 1080 this morning but did not think I gained much and I lose another whole CPU thread that could be doing POGS. I use 0.50 at Milkyway and that works great as as MW does not use much CPU.


kashi

#586
Good that the new one installed without problem and runs cool. v:

Was a bit surprised at the great variability in Einstein runtimes for the same cards and the sometimes smallish difference between 970s, 980s 1070s and 1080s.

At first thought it was sometimes due to configurations with multiple cards slowing GPU processing down a bit but I've seen some slower times with single cards. And no, not that slow that it is likely 2 or more tasks run concurrently.

Then thought perhaps the speed of the CPU can have an affect because there is a portion of CPU processing after the semicoherent output stage is completed by GPU.

Then realised that various models/brands of the same card type can run at different speeds or boost differently. This could also perhaps be affected by cooling solution used on card, case cooling, room temperature.

Then there's potential CPU starvation issues, some may get a GPU speed up by leaving a CPU core free in addition to the CPU core automatically allocated to GPU tasks.

In some projects the higher model cards don't get to more fully utilise their performance potential unless they run multiple concurrent tasks but Dataman's experience with his 1080 and 2 concurrent means possibly not on Einstein.

So don't know, perhaps it could be a number of those issues. Strange though, I'd expect a much greater difference considering the differences in hardware specification. In other words, would have thought a 1080 or even a 1070 should absolutely blitz my 970 but doesn't seem to be the case on Einstein.

Do know it's hard to research many different configurations now in Einstein as the website has been horridly vandalised so only shows 50 hosts.

Noticed the task list is almost unusable now as well as no task numbers are shown and also no listing of different task types for those who crunch more than 1 type or do Einstein CPU as well.

What is it with these projects completely mucking up their websites lately? Is it some misguided and poorly implemented attempt to make the websites "mobile friendly" or what? Next they'll be hiding wingman details as well on quorum tasks like the horror, the horror of WCG. Bah, humbug.

Running multiple cards can be a pain sometimes and is something I resolved not to do again. I too would choose a single higher model GPU if I was getting stuck right into GPU crunching again. However this all makes me wonder if a couple of lower models such as 1050 Ti or 1060 would give better bang per watt on Einstein than a single top model. Because the electricity bills keep coming in after the cost of GPUs is forgotten.

chooka03

When I had my R9 280x I could run more than 1 WU with Einstein but the models after that could only run 1 WU at a time. It was well talked about on the forum.
I wasn't worried really. 1 WU was enough to occupy most of the card.

tazzduke

Greetings

Running 2 workunits on my GTX 980, 960 and 950, the 980 blitzs through 4 workunits per hour, whereas the 960 and 950 do 2 per hour.

I think I should put the 960 and 950 back to 1 task, but I cant be stuffed lol.

So in other words the 980 is doing the same work as the 960 and 950 combined, go figure.

Regards



 AA 24 - 53 participant

Dataman

#589
Yep, my 1070's and 1080 are barely outperforming the 970's on Einstein. After the AA I will try MW. I once ran 0.33 at MW and got 3 to run but the card got hot quickly. Think I only ran 6 or 9 wu's that way.
Why does everything have to have a light on it these days? The 10x0's shine like there is a flashlight (torch) inside the case. There should be a "turn off the $#@%'ing light" option.  :rofl: At least they don't blink.  :wink
The cooling is fantastic! I cannot remember a time, when fully loaded, the GPU temps were cooler than the CPU temp's.
I think the 10 series should be awesome at MW, PG and Amicable Numbers. Should be fun.
:greet


JugNut

#590
Yea they should end up being great at Amicable Numbers & Prime Grid as well but milkyway err probably not.  The Nvidia 10 series double precision math capabilities are even more terrible than usual.  I was going to look up the numbers for you but thought nah.. it would only prove what I already know, and that's they're crap at double precision math. 

Of course they are deliberately crippled by Nvidia so they can sell more multi thousand dollar Tesla's & Quatro's, but somehow that doesn't make me feel any better about it.

You see a month or so ago I bought 1080 ti FE off a friend who had bought one to many and couldn't squeeze em in his flash micro chassis. Soo.. I did him a solid and bought one off him at mates rates.  Well, we are certainly not talking cheap,  cheaper might be a better way to describe it.
 
Anyway as I was getting to,  my 5 year old AMD HD 7970 runs rings around the 1080 ti with Milkyway tasks.  The 7970 is at least 3 times faster,  maybe even more.  Now I wasn't expecting miracles but I didn't expect just how poorly it end up doing.  It's still worth a try though,  you have nothing to lose after all.   

Single precision is what these cards excel in which luckily for us is what most other projects use when using GPU's. So if you really want to stretch it's legs then crunch some collatz & then see the credits fly. The numbers these cards put out is amazing,  OK not anywhere near BU amazing but still it's as close as you'll get post BU anyway.  GPUgrid is another that should be good with the 10 series cards too.

Anyhoo that's enough yabber for me for the night(it's 3.45am here) it's time for bed, all the best  :wavey2:

Dataman

#591
Latest news (?) from Henri. I am going to pass for now. Henri seems to be up to something?

http://www.bitcoinutopia.net/bitcoinutopia/forum_thread.php?id=1143#12098




chooka03

LOL  :rofl:
LOVE the picture.

JugNut

#593
Mining makes a less than glorious return. So far just for GPU's(ATI only, cuda soon)  But this time you get to keep most the profit...

http://boinc.gridcomputingcenter.org/forum_thread.php?id=226&postid=1907#1907


PS:  Epic pic DM..  :rofl: