3/19/2014 10:51:04 PM | Asteroids@home | Giving up on download of input_20770_25: permanent HTTP error
uh-oh!

Gong show? Lol, don't know what you mean, but anyway MW doesn't have a scrn saver.
MW could be a useful benchmark for A@H because it uses DP, but like you said it could still give different results nonetheless.

---

Team AnandTech - SETI@H, Muon1 DPAD, Folding@H, MilkyWay@H, Asteroids@H, LHC@H, POGS, Rosetta@H, Einstein@H,DHPE & CPDN

Main rig - Ryzen 3600, 32GB DDR4 3200, RX 580 8GB, Win10
2nd rig - i7 4930k @4.1 GHz, 16GB DDR3 1866, HD 7870 XT 3GB(DS), Win7

Milky Way uses OpenCL. Would it be an accurate comparison between cards if they use different platforms?

Probably not then.

---

Team AnandTech - SETI@H, Muon1 DPAD, Folding@H, MilkyWay@H, Asteroids@H, LHC@H, POGS, Rosetta@H, Einstein@H,DHPE & CPDN

Main rig - Ryzen 3600, 32GB DDR4 3200, RX 580 8GB, Win10
2nd rig - i7 4930k @4.1 GHz, 16GB DDR3 1866, HD 7870 XT 3GB(DS), Win7

I was wondering if users who run A@H on GPUs could give a few examples of how fast an A@H task completes on a given model of GPU. Also, how many A@H tasks are completed in parallel. Or simply put: How many A@H tasks can your GPU complete in 24 hours?

Thanks!

---

http://iqjar.com

GTX 780Ti @ 1124MHz completes in about 1430 seconds.

Thanks, Matt!

So, about 24 minutes for a task on the GTX 780Ti. And only one task at a time?
Wow, it seems GPUs are not as fast as I thought. The GTX 780Ti costs about 800$ and it has a whole lot of CUDA cores. So a more affordable model (say around 150$), with less cores and less clock speed would probably take hours to chew a task, no faster than a mid-range CPU core...

---

http://iqjar.com

Ok. So let's be honest... If we're porting asteroids to gpu, the only way to go is opencl! That will of course take into account that all nvidia gpus except the titan suck at DP...

So no need to waste energy in porting the app to cuda when we already know that asteroids will be dominated by ATI/AMD gpus because those are less crippled regarding DP performance.

To face the fact, Asteroids will be a "second" Milkyway@home regarding GPU dominance.

AMD/ATI GPUs will dominate and nvidia will be left biting the dust.

If one takes a look into the top hosts list@Einstein, he will see that the top hosts are mainly AMD gpu systems, also the project is single precision (except the new gw-wu's).
For a DP-project AMD should be the gpu to be choosen. Even the new AMD A10-7xxx APU's do a very good job for a fair price.

Asteroids@Home uses single precision floating point calculations or double precision floating point calculations?

Asteroids@Home uses single precision floating point calculations or double precision floating point calculations?

double.

Also, how many A@H tasks are completed in parallel.

According to gpu-z the gpu load is 99%. This means, it makes no sense to run more than one wu at a time.

I have added another pc to the project. I have now 3 generations of nVidia GTX x50ti running. Everyone who's interested can watch the performance.
A nonvisible difference is the power consumption. The newest card, the GTX750ti, requires ~42Watt electrical power for crunching. The oldest card, the GTX550ti uses >85 Watts for that.

What's the performance difference?

---

Team AnandTech - SETI@H, Muon1 DPAD, Folding@H, MilkyWay@H, Asteroids@H, LHC@H, POGS, Rosetta@H, Einstein@H,DHPE & CPDN

Main rig - Ryzen 3600, 32GB DDR4 3200, RX 580 8GB, Win10
2nd rig - i7 4930k @4.1 GHz, 16GB DDR3 1866, HD 7870 XT 3GB(DS), Win7

double post

---

Team AnandTech - SETI@H, Muon1 DPAD, Folding@H, MilkyWay@H, Asteroids@H, LHC@H, POGS, Rosetta@H, Einstein@H,DHPE & CPDN

Main rig - Ryzen 3600, 32GB DDR4 3200, RX 580 8GB, Win10
2nd rig - i7 4930k @4.1 GHz, 16GB DDR3 1866, HD 7870 XT 3GB(DS), Win7

GTX 550ti ~10,465
GTX 650ti 6,145 - 7,438
GTX 750ti 4,736 - 5,660

So newer cards are faster even with dp wu's, use much less power.
But all in all there seems to be something wrong with the cuda code; the gpu app should be much faster than a cpu wu. They have to calculate in dp as well!

For the moment it makes little on no sense to run gpu-apps (unless one has a high-end gpu).

Looking forward to a new (AMD?) gpu-app!

GTX 750ti 4,736 - 5,660

So newer cards are faster even with dp wu's, use much less power.

Better architecture.

the gpu app should be much faster than a cpu wu.

Definitely not.

Definitely not.

OK, if the Dev. says that I have to accept that.
But just to understand what I mean: I run several gpu projects with cpu and gpu wu's; this is the only project that has no significant increase in performance. I just wanted to express my experiance I made with other projects, I did not want to offend somebody.
Could you please share your knowledge about this with us?

Alexander

I agree with Alex, in MW my GPU is about 51x faster than my CPU!
Also seen massive speed ups in the GPU F@H app vs CPU.

What makes A@H different?
If you did mention it in another thread (I think you may of done??) then by all means link us :).

Btw Alex, big efficiency improvement in the 750 vs 550! :)

---

Team AnandTech - SETI@H, Muon1 DPAD, Folding@H, MilkyWay@H, Asteroids@H, LHC@H, POGS, Rosetta@H, Einstein@H,DHPE & CPDN

Main rig - Ryzen 3600, 32GB DDR4 3200, RX 580 8GB, Win10
2nd rig - i7 4930k @4.1 GHz, 16GB DDR3 1866, HD 7870 XT 3GB(DS), Win7

GTX 550ti ~10,465
GTX 650ti 6,145 - 7,438
GTX 750ti 4,736 - 5,660

So newer cards are faster even with dp wu's, use much less power.
But all in all there seems to be something wrong with the cuda code; the gpu app should be much faster than a cpu wu. They have to calculate in dp as well!

For the moment it makes little on no sense to run gpu-apps (unless one has a high-end gpu).

Looking forward to a new (AMD?) gpu-app!

Have you tried leaving a cpu core free just to feed the gpu yet? If not see if it speeds up the gpu processing.

I have a 760 that does this:
3,899.84 24.41 480.00 Period Search Application v101.12 (cuda55)
I have an older 560Ti that does this:
4,934.89 26.14 480.00 Period Search Application v101.12 (cuda55)
I have another 760 that is doing this:
3,775.34 10.44 480.00 Period Search Application v101.11 (cuda55)

All machines have a cpu core free just to feed the gpu. There are several units I could have chosen from to show you the numbers from, my pc's are not hidden, but most units I am crunching right now are the cpu ones so the gpu units are at the end of the valid lists.

GTX 550ti ~10,465
GTX 650ti 6,145 - 7,438
GTX 750ti 4,736 - 5,660

So newer cards are faster even with dp wu's, use much less power.
But all in all there seems to be something wrong with the cuda code; the gpu app should be much faster than a cpu wu. They have to calculate in dp as well!

For the moment it makes little on no sense to run gpu-apps (unless one has a high-end gpu).

Looking forward to a new (AMD?) gpu-app!

Have you tried leaving a cpu core free just to feed the gpu yet? If not see if it speeds up the gpu processing.

I have a 760 that does this:
3,899.84 24.41 480.00 Period Search Application v101.12 (cuda55)
I have an older 560Ti that does this:
4,934.89 26.14 480.00 Period Search Application v101.12 (cuda55)
I have another 760 that is doing this:
3,775.34 10.44 480.00 Period Search Application v101.11 (cuda55)

All machines have a cpu core free just to feed the gpu. There are several units I could have chosen from to show you the numbers from, my pc's are not hidden, but most units I am crunching right now are the cpu ones so the gpu units are at the end of the valid lists.

Hello Mikey,
oh yes, I did free a core for the gpu job! I have experiance with that; otherwise the crunching times would be even higher!
Anyway, thx for reminding me!

Alexander