470 performance

[Laco Gaal]

could someone reupload the Candle test?

thanks

[Margus Voll]

cuda manual update did not change stuff. will see if update did put back this kext.

I still have not found: “KEXT power management modification steps as suggested”

Any help would be perfect.

—

Margus

https://iconstudios.eu

[Margus Voll]

Seems thats system update did put kext back.

Had to remove it again.

—

Margus

https://iconstudios.eu

[Vladimir Kucherov]

Yep, same thing happened to me.

I gotta say, my 470 on the mac is slower still than the same 470 running in Windows on the same mac.

[Margus Voll]

what windows do you have ?

Interesting idea to play with

—

Margus

https://iconstudios.eu

[Juan Salvo]

Here you go Margus:

https://macvidcards.com/2011/29/fermi-cards-come-to-the-mac-pro/

It’s the section called EDITING AGPM.

[Vladimir Kucherov]

Running Windows 7, with Resolve 8.2 beta 3.

I don’t have a benchmark to back that up though. I’ll try the blur node and nr node comparison test as soon as I have a chance.

[Vladimir Kucherov]

Thanks for this link! This article implies that actually properly editing the kext should yield better performance than deleting it. I’ll try it out

[David Pirinelli]

From those CUDA-Z numbers it also app[ears that you have the 470 in a 4x slot.

Host to Device will be around 5000 when in a 16x slot

The Single Precision Float number is the one that seems to predict Resolve performance.

When the kext is properly deleted or edited, that number should be 1080 or so with a 470.

A GTX285 1 Gig card will achieve around 700.

A GTX480 at full speed will hit 1300-1400.

And a Quadro 4000 usually gets around 450 or so.

BTW, we now have a GTX285 that only requires a single power cable. It is NOT as fast as a regular GTX285 as it is down clocked, but it IS faster than a Quadro 4000. It is a double width card, but only having a single power connection means it can work with either a Quadro 4000 or a AMD 5770 with no splitters, adapters or other buggery trickery.

It scores around 500 in the single precision float, so a hair faster than a Quadro 4000.

But everyone with a GTX470 or GTX480 will need to realize that each OS update will probably include a new AGPM kext. FWIW, I believe that the GTX470 still uses it’s own planned power management when this kext is deleted. You can verify this with the OpenGL View test. When you first run the test the numbers will be in the hundreds, after a second or two the drivers register the additional load and kick the card into high gear, you will see the number LEAP into the thousands. If you re-run the test right away they will start in thousands again. But if you leave the machine idle for a few minutes, the score will start in hundreds again. So I am 99% certain that the 470 is just fine without that kext.

The odd thing about the AGPM kext is that it only has 2 @ Nvidia cards defined, the GT120 and a GTX260. (device ids of 0640 and 05e2) This is odd because Apple never shipped or offered a GTX260 and yet the GTX285 which uses device id of 05e3 runs fine.

The reason I include instructions for changing the “05e2′ entry into whatever the 2nd card is is so that the 0640 entry stays and the GT120 remains. I am not certain that it is important but since adding it properly does make it run the card, it doesn’t seem to hurt.

[Margus Voll]

I think it was the AGPM kext after update.

After removing it i got the same number Sasha got in he’s screenshot.

I will also look in to editing AGPM kext.

I’ll let you know.

—

Margus

https://iconstudios.eu