I just completed my first NVIDIA work unit -- it took 14 hours to run and I got 480 points credit -- the SAME credit I have been getting for CPU WU's that take 9-10 hours to run!! My GPU (GT-520) ain't the fastest horse in the race, but it executes SETI and Einstein WU at least 8 times faster than similar WU's running on my AMD Dual Core CPU. So ... I will let the current GPU WU complete and not run any more!

---

There aren't any GPU applications for this project. How are you running tasks on NVIDIA?

---

Really? :))

http://asteroidsathome.net/boinc/apps.php

http://asteroidsathome.net/boinc/forum_thread.php?id=233&postid=2227#2227

---

- ALF - "Find out what you don't do well ..... then don't do it!" :)

Here is another thread to read for you:
http://asteroidsathome.net/boinc/forum_thread.php?id=131

The nVidia GPU's are not well suited for double precision, this is one reason why there is little or no gain when using them. AMD works better with dp, but there is no app at the moment.
Many people are waiting to see an openCL app.

Here is another thread to read for you:
http://asteroidsathome.net/boinc/forum_thread.php?id=131

The nVidia GPU's are not well suited for double precision, this is one reason why there is little or no gain when using them.

That's not true. The point is that the GPU lessens the time needed for the completion of a WU. Even my small GeForce GTX 650 puts it down to about 2/3.
And because both types of WUs have the same amout of computation to be done this comparison is very easy.

In my opinion this calls for higher points granted for GPU work than for FPU/SSE work.

 
Your post shows you don't understand some things:

- "the GPU lessens the time needed" - in fact the GPU do all the Computing on a task marked "for GPU"

- "both types of WUs have the same amount of computation":

1) There is no such thing as "types of WUs"

There is no difference between "CPU" and "CUDA" WUs (because "CPU" and "CUDA" WUs do not exist)
http://asteroidsathome.net/boinc/forum_thread.php?id=263&postid=3681#3681
http://asteroidsathome.net/boinc/forum_thread.php?id=263&postid=3686#3686

This is a WU:
http://asteroidsathome.net/boinc/workunit.php?wuid=22613435

The "WU" is universal (data file) and can be sent to be processed by any application that can run on the system the task is sent to.
The above WU was sent 3 times (as 3 tasks - all of which are the same file)

This is your task (data file) - which was sent to be processed by CUDA app (on NVIDIA GPU):
http://asteroidsathome.net/boinc/result.php?resultid=53605965

This is another task (same data file, copy of the same WU) - which was sent to be processed by AVX CPU app
http://asteroidsathome.net/boinc/result.php?resultid=52583477

You can see clearly that the same WU was done no-problem by both CUDA and CPU apps and the results match (validated)
And of course the credit granted to both tasks is the same as it should be (same work = same payment)


2) WUs obviously have not "the same amount of computation" as the difference in run time may be even 10 times (using the same application on the same computer)
http://asteroidsathome.net/boinc/results.php?hostid=110&offset=20&show_names=0&state=4&appid=

In my opinion this calls for higher points granted for GPU work than for FPU/SSE work.

After reading the above - do you still think so?


Terminology:
WU = data file (which will be send to different applications, the sent file is called "task")
Tasks = identical data files, copy of the same WU

Tasks are sent to any app that can do the Computing on them
Any WU can 'generate' copy of itself ("task") which is marked 'for CPU' or 'for GPU' only at the moment the task is sent
If your BOINC asks for GPU work the task will be marked 'for GPU' (CUDA)
If your BOINC asks for CPU work the task will be marked 'for CPU' (e.g. SSE3)

---

- ALF - "Find out what you don't do well ..... then don't do it!" :)

@ BilBg:

You made it pretty clear that there are no different types of WUs - sorry, that misspelling was my fault.

Background is (or better was) this: BOINC allows you to get the estimated GFLOPs for computing a WU. For this I found all the Asteroids@home WUs I got lately showed the same estimation.
And the ones done by my GPU are faster through. Or so I thought.

Guess the BOINC estimation isn't that precise after all.

In fact, I nearly always forget to use the detailed statistics pages like you mentioned (http://asteroidsathome.net/boinc/results.php?hostid=110&offset=20&show_names=0&state=4&appid=). On the other hand, A@H is the only project I know of that uses the same WU data files for both CPU and GPU. All other projects I know of use different types of files for both. This I forgot too, sorry.

In my defense, it was pretty late for me when I wrote the recent post.....

Thank you for clearing things up!
:)

Guess the BOINC estimation isn't that precise after all.

This has been a problem ever since Boinc began, it probably always will be a problem as they use formulas due to them not running an actual test unit on our pc's. This was discussed but discarded, and now the Boinc software 'learns' as we crunch more and more units. The problem is most of us don't just have rooms full of pc's doing nothing else but crunching, even checking our email, or typing in a forum, can change how much time our pc has available to crunch throwing the calculations off.

ps the reason the 'test unit' idea was discarded is that since it doesn't contribute anything to the project the projects feel we should NOT get any credits for it, but we users said 'hey I'm crunching I want some credits for my work'. So now everyone just jumps right into the pool with both feet and up to our necks. The formula IS MUCH better than it used to be, but as I said it will probably never be 100% accurate.

Guess the BOINC estimation isn't that precise after all.

This has been a problem ever since Boinc began, it probably always will be a problem as they use formulas due to them not running an actual test unit on our pc's. This was discussed but discarded, and now the Boinc software 'learns' as we crunch more and more units. The problem is most of us don't just have rooms full of pc's doing nothing else but crunching, even checking our email, or typing in a forum, can change how much time our pc has available to crunch throwing the calculations off.

ps the reason the 'test unit' idea was discarded is that since it doesn't contribute anything to the project the projects feel we should NOT get any credits for it, but we users said 'hey I'm crunching I want some credits for my work'. So now everyone just jumps right into the pool with both feet and up to our necks. The formula IS MUCH better than it used to be, but as I said it will probably never be 100% accurate.

Thanks! I didn't know that.
:)

Behold: The question for credits on a test/calibration WU skipped the better method? Man, sometimes reasons are just as stupid as can be.... Seems to me that this was greed. I mean, it would have been only one WU. Ah come on.......

Guess the BOINC estimation isn't that precise after all.

This has been a problem ever since Boinc began, it probably always will be a problem as they use formulas due to them not running an actual test unit on our pc's. This was discussed but discarded, and now the Boinc software 'learns' as we crunch more and more units. The problem is most of us don't just have rooms full of pc's doing nothing else but crunching, even checking our email, or typing in a forum, can change how much time our pc has available to crunch throwing the calculations off.

ps the reason the 'test unit' idea was discarded is that since it doesn't contribute anything to the project the projects feel we should NOT get any credits for it, but we users said 'hey I'm crunching I want some credits for my work'. So now everyone just jumps right into the pool with both feet and up to our necks. The formula IS MUCH better than it used to be, but as I said it will probably never be 100% accurate.

Thanks! I didn't know that.
:)

Behold: The question for credits on a test/calibration WU skipped the better method? Man, sometimes reasons are just as stupid as can be.... Seems to me that this was greed. I mean, it would have been only one WU. Ah come on.......

The problem is that even with a test unit, using your pc while crunching will affect your crunching speed and times, so even running a non points paying test unit wouldn't solve all the problems.

And it's one unit PER project, so every project you sign up for you would have to run the test unit, each being different as each project is different from each other. It all got to be very complicated.

All other projects I know of use different types of files for both.

Recently (14 Oct 2014) you did some work for SETI@home
http://setiathome.berkeley.edu/hosts_user.php?sort=rpc_time&rev=0&show_all=1&userid=9721985

Look in this table under 'SETI@home v7':
http://setiathome.berkeley.edu/apps.php

Any SETI@home v7 WU can be done by any of >20 apps listed:
On CPU (Intel, AMD, ARM)
On GPU (NVIDIA, ATI AMD, Intel iGPU, Neon)
Under different OSes (Windows, Mac OS, Linux, Android)

---

- ALF - "Find out what you don't do well ..... then don't do it!" :)

ASTEROIDS GPU is not only slowing the CPU very seriously,
but today I tried to run it on CPU + GPU.

All CPU WU's are in "waiting mode".

Waiting for what, I don't know.

As far as I know, it's the only GPU project having such a negative impact on the CPU while it is not supposed to.

Even if you dedicate a CPU core to ASTEROIDS GPU.

It looks like no improvement was made on these GPU apps since launch,
regardless the numerous crunchers remarks.

It's "sad".

Best,

Ph.

ASTEROIDS GPU is not only slowing the CPU very seriously,
but today I tried to run it on CPU + GPU.

All CPU WU's are in "waiting mode".

Waiting for what, I don't know.

As far as I know, it's the only GPU project having such a negative impact on the CPU while it is not supposed to.

Even if you dedicate a CPU core to ASTEROIDS GPU.

It looks like no improvement was made on these GPU apps since launch,
regardless the numerous crunchers remarks.

It's "sad".

Best, Ph.

Asteroids gpu units NEED a cpu core too, they crunch on both depending on how far along the unit is. This is becoming more the norm rather than the exception across Boinc as more and more projects figure out it is faster and more efficient, meaning the units are being packed with more calculations so more work is being done.

Hi Mikey,

Thank you for your message.

Sorry, but ...

a) even if a dedicate a cpu core / GPU WU, (app_config) it slows down and sometimes freezes the CPU WU's ???

b) CPU time shown in the stats don't confirm what you are writting.

Less than 10 seconds CPU time per GPU WU (CUDA55)

c) ie on GPUGRID, I know how to "oblige" BM to use the core CPU it is supposed to use, and there is a real gain. (CPU time = GPU time)

d) don't understand why, but when running Asteroids on CPU + GPU, my machines are crashing

e) What's the point wasting GPU time when the gain is only 50 % ag AVX WU's ?
If optimzed, these CUDA app should/could run in about 10 minutes ...

Nevertherless, I like the project, and I try to understand how to improve crunching on it :)

Best

Phil1966