Hello,
I've been wondering for a while why the ATI 5870 on my MacPro only crunches 6000 credits a day while the NVIDIA GT650M on my MacBook Pro achieves about 20k.
So I tried to put the display on an older GPU to see how it performs when it can use 100% for computing.
With one surprising Result:
BRP4 Tasks take the same time on both machines now, about 5800 seconds. Before, when I had a display on the ATI it took 15.000 seconds to finish a task.
http://einsteinathome.org/account/tasks&offset=0&show_names=1&state=3&appid=25
BRP5 however vary strongly with only ~16.000 seconds on the notebook but ~40.000 seconds on the ATI Desktop GPU which was about 50k when I had a display connected to it:
http://einsteinathome.org/account/tasks&offset=0&show_names=1&state=3&appid=23
So I'm curios for the explanation, is the cuda app just better optimized. Are the ATI drivers bad for that task? Is it a different GPU architecture? Both cards have 1GB of RAM.
Copyright © 2024 Einstein@Home. All rights reserved.
Varying GPU Performance
)
Howdy,
The AMD\ATI 5870 is an older technology. It came out in Sept 2009 so, 5 years old.
Still, that seems low - I was getting approx 18K on my 5870 when running it full time. That was with running one work unit at a time.
AMD\ATI drivers are fine. (except someone in another thread was experiencing problems with Beta drivers.) Stick with released drivers and you should be good to go.
Bill
yes as mentioned with BRP4
)
yes as mentioned with BRP4 tasks the ATI manages about 20k a day, so the card itself seems fine as well. So the question is what's wrong with the BRP5 Tasks in combination with that Card or why does the NVIDIA deal so much better with them while having the same computing power?
As I'm on a Mac there just is the original driver for this card.