GT 520 GPU vs. AMD CPU


Message boards : Number crunching : GT 520 GPU vs. AMD CPU

Message board moderation

To post messages, you must log in.
AuthorMessage
w1hue

Send message
Joined: 19 Jun 12
Posts: 20
Credit: 18,012,737
RAC: 2,696
Message 3911 - Posted: 3 Jan 2015, 4:50:33 UTC
I recently completed four Astroids@Home WUs using a NVIDIA GT 520 GPU (overclocked 15%) and the average run time was 38,353 seconds. The previous four WUs were executed in the AMD Athion 64 X2 4600+ CPU (overclocked 10%) in the same machine and had an average run time of 30,878 seconds. The GPU executes SETI, Einstein and Milkyway WUs between 6 and 8 times faster than equivalent WU in the CPU -- so I will continue running those WUs and not this project's WUs in the GPU and this project's in the CPU.

My GT 520 ain't the fastest horse in the race, but gee. . .
ID: 3911 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
noderaser
Avatar

Send message
Joined: 26 Jun 13
Posts: 22
Credit: 665,693
RAC: 508
Message 3912 - Posted: 3 Jan 2015, 5:29:48 UTC

Last modified: 3 Jan 2015, 5:30:31 UTC
The massively parallel capabilities of many GPUs can greatly accelerate some (but not all) types of computation.

Most of us found that out 6-7 years ago :P

ID: 3912 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
w1hue

Send message
Joined: 19 Jun 12
Posts: 20
Credit: 18,012,737
RAC: 2,696
Message 3916 - Posted: 4 Jan 2015, 3:15:03 UTC - in response to Message 3912.  
Most of us found that out 6-7 years ago :P

Good for you. . .

This is obviously an example where using a GPU to do the calculations is of no particular advantage. So why bother. . .
ID: 3916 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
noderaser
Avatar

Send message
Joined: 26 Jun 13
Posts: 22
Credit: 665,693
RAC: 508
Message 3917 - Posted: 4 Jan 2015, 4:45:40 UTC
To be honest, I haven't really compared my GPU and CPU tasks on this project too closely. With some projects the size of each task can vary a lot, and many give GPUs larger tasks since they are able to process normal ones much quicker. However, I'm just speaking generally--don't have enough experience with this particular project to say whether that is the case or not, since it seems a lot of GPU tasks get validated against CPU ones.

If you're interested in increasing your GPU's output, you might have a look at some of the other GPU projects.

ID: 3917 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
w1hue

Send message
Joined: 19 Jun 12
Posts: 20
Credit: 18,012,737
RAC: 2,696
Message 3919 - Posted: 5 Jan 2015, 3:40:39 UTC - in response to Message 3917.  

Last modified: 5 Jan 2015, 3:41:28 UTC
To be honest, I haven't really compared my GPU and CPU tasks on this project too closely. With some projects the size of each task can vary a lot, and many give GPUs larger tasks since they are able to process normal ones much quicker. However, I'm just speaking generally--don't have enough experience with this particular project to say whether that is the case or not, since it seems a lot of GPU tasks get validated against CPU ones.

Since the GPU tasks are given the same credits as the CPU tasks, I assume that they are both doing the same sort of computations.

If you're interested in increasing your GPU's output, you might have a look at some of the other GPU projects.

I am running SETI@HOME, Einstein@Home and Milkyway@Home tasks in the GPU because they run considerably faster than in the CPU. Except for some of the Einstein tasks -- which I run (CPU versions only) in a couple of machines that don't have BOINC compatible GPUs.
ID: 3919 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
noderaser
Avatar

Send message
Joined: 26 Jun 13
Posts: 22
Credit: 665,693
RAC: 508
Message 3920 - Posted: 5 Jan 2015, 5:38:04 UTC - in response to Message 3919.  

Last modified: 5 Jan 2015, 5:38:15 UTC
Since the GPU tasks are given the same credits as the CPU tasks, I assume that they are both doing the same sort of computations.

All tasks get the same credit at this project, when they are completed and verified successfully, regardless of their computational length.

ID: 3920 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 1 Jan 14
Posts: 300
Credit: 32,051,276
RAC: 14,682
Message 3921 - Posted: 5 Jan 2015, 12:35:26 UTC - in response to Message 3919.  

Last modified: 5 Jan 2015, 12:37:31 UTC
To be honest, I haven't really compared my GPU and CPU tasks on this project too closely. With some projects the size of each task can vary a lot, and many give GPUs larger tasks since they are able to process normal ones much quicker. However, I'm just speaking generally--don't have enough experience with this particular project to say whether that is the case or not, since it seems a lot of GPU tasks get validated against CPU ones.


Since the GPU tasks are given the same credits as the CPU tasks, I assume that they are both doing the same sort of computations.


If you're interested in increasing your GPU's output, you might have a look at some of the other GPU projects.


I am running SETI@HOME, Einstein@Home and Milkyway@Home tasks in the GPU because they run considerably faster than in the CPU. Except for some of the Einstein tasks -- which I run (CPU versions only) in a couple of machines that don't have BOINC compatible GPUs.


I know you didn't mean it as 'all inclusive' but there are several other projects too that can use a gpu to help them out too. A short list being Collatz, Moo, GpuGrid, PrimeGrid, Bitcoin Utopia, Poem but the units are hard to get as they are limited, and even World Community Grid has some sometimes although not at the moment. This is NOT meant as a 'which project is best list' it is just a list of additional places that can use a gpu if you want to expand your horizons a little bit. Some projects pay alot of credits for gpu crunching, some not so much, Asteroids is in the later category. Some projects require certain gpu features which not all gpu's have, ie 'dual precision', MilkyWay is one that does this. Some can use both Nvidia and AMD/ATI cards, ie PrimeGrid, some can not and instead use one brand or the other, ie GpuGrid.
ID: 3921 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
w1hue

Send message
Joined: 19 Jun 12
Posts: 20
Credit: 18,012,737
RAC: 2,696
Message 3922 - Posted: 5 Jan 2015, 18:46:19 UTC - in response to Message 3921.  
I am currently running projects that are of most interest to me. The ones that you mentioned are of no or little interest. I'm not just doing it for the "points". . .

Thanks anyway.
ID: 3922 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 1 Jan 14
Posts: 300
Credit: 32,051,276
RAC: 14,682
Message 3923 - Posted: 6 Jan 2015, 12:45:47 UTC - in response to Message 3922.  
I am currently running projects that are of most interest to me. The ones that you mentioned are of no or little interest. I'm not just doing it for the "points". . .

Thanks anyway.


But yet your very first post in this thread included this:
"My GT 520 ain't the fastest horse in the race, but gee. . .", I am confused! If it's about the Science then who cares how long it takes, or how many points we get, as long as GOOD Science is getting done. Science is a LIFETIME endeavor, not an overnight thing, some of these projects will NEVER end!!
ID: 3923 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
w1hue

Send message
Joined: 19 Jun 12
Posts: 20
Credit: 18,012,737
RAC: 2,696
Message 3924 - Posted: 6 Jan 2015, 22:28:02 UTC - in response to Message 3923.  

But yet your very first post in this thread included this:
"My GT 520 ain't the fastest horse in the race, but gee. . .", I am confused! If it's about the Science then who cares how long it takes, or how many points we get, as long as GOOD Science is getting done. Science is a LIFETIME endeavor, not an overnight thing, some of these projects will NEVER end!!

I don't really care how long it takes -- if I did, I'd invest in faster hardware! I was just surprised that there seemed to be no advantage in using the GPU over the CPU for this project.
ID: 3924 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 1 Jan 14
Posts: 300
Credit: 32,051,276
RAC: 14,682
Message 3927 - Posted: 7 Jan 2015, 11:52:02 UTC - in response to Message 3924.  

But yet your very first post in this thread included this:
"My GT 520 ain't the fastest horse in the race, but gee. . .", I am confused! If it's about the Science then who cares how long it takes, or how many points we get, as long as GOOD Science is getting done. Science is a LIFETIME endeavor, not an overnight thing, some of these projects will NEVER end!!


I don't really care how long it takes -- if I did, I'd invest in faster hardware! I was just surprised that there seemed to be no advantage in using the GPU over the CPU for this project.


There is alot that goes into the making of an app, the skill of the programer is only a part of it and each project has their own programers. Trying to get relevant data out of a gpu app can be difficult as the way the project is used to doing it works just fine for a cpu, which is much different. Rosetta has said MANY times that they won't make a gpu app as their current data crunching methods just doesn't translate over that well, and there is no advantage for them or us. I'm guessing Asteroids has a similar problem.
ID: 3927 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
MarkJ
Avatar

Send message
Joined: 27 Jun 12
Posts: 129
Credit: 62,714,553
RAC: 0
Message 3929 - Posted: 7 Jan 2015, 12:13:43 UTC

Last modified: 7 Jan 2015, 12:18:32 UTC
Part of the problem is Asteroids do floating point maths. Most GPU's except the super expensive ones are deliberately crippled by the manufacturers. Typically they cripple them to the order of 1/24th or 1/32nd of their single precision performance. It is much more noticeable on lower end cards because they have less processing units than a high end card.

Most projects do single precision maths in the GPU apps. Milkyway do floating point, but insist on a card that can do it. I am sure if single precision or integer maths was good enough Asteroids would have used it in their apps.
BOINC blog
ID: 3929 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
w1hue

Send message
Joined: 19 Jun 12
Posts: 20
Credit: 18,012,737
RAC: 2,696
Message 3932 - Posted: 8 Jan 2015, 3:22:32 UTC - in response to Message 3929.  

Last modified: 8 Jan 2015, 3:24:10 UTC
Milkyway do floating point, but insist on a card that can do it. I am sure if single precision or integer maths was good enough Asteroids would have used it in their apps.

My GT 520 does a good job on Milkyway tasks (they use DOUBLE PRECISION FP, BTW). Check my stats below -- all Milkyway credits were earned with the GPU.

Edit -- well forget looking at my stats in the signature line; BOINC Stats must be screwing around with user IDs again!!
ID: 3932 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : GT 520 GPU vs. AMD CPU