agpgart headache with radeon

English Hardware
1

I’m running an aging Thinkpad T42 with opensuse 12.3 (and very nice, too!). However - fglrx isn’t supported on the old ATI Mobility 9600 and so I’m using Radeon.

Fine - but even although I have an option to force/request AGPMode to the higher speed, it’s not being honored;

myserver:~ # cat /etc/X11/xorg.conf.d/50-device.conf
Section “Device”
Identifier “Default Device”
Driver “radeon”
Option “AGPMode” “2”
Option “AccelDFS” “on”
EndSection

…the Xorg.0.log is saying it’s ignoring the AGPMode option these days;
(WW) RADEON(0): Option “AGPMode” is not used

…and dmesg still says the following;
myserver:~ # dmesg | grep agp
1.343270] Linux agpgart interface v0.103
1.343395] agpgart-intel 0000:00:00.0: Intel 855PM Chipset
1.357327] agpgart-intel 0000:00:00.0: AGP aperture is 256M @ 0xd0000000
8.079524] agpgart-intel 0000:00:00.0: AGP 2.0 bridge
8.079535] agpgart-intel 0000:00:00.0: putting AGP V2 device into 1x mode

…so I should be seeing AGP running in a 2x, 4x or 8x mode… But I’m not. I’ve googled for some answers but there’s nothing out there - at least, nothing that’s less than five years old!

I appreciate this isn’t cutting-edge hardware, but if you have seen this before - rack your brains and share your wisdom!

Thanks,

Russ.

2

Note:

  • AGP 2.0 supports 1x, 2x, and 4x … so, no 8x support should be expected
  • The option you set would be for placing it 2x mode … you’d use “4” for 4x
  • googling for “(WW) RADEON(0): Option “AGPMode” is not used” provides some hits on the first page that are informative:
3

To add a little more (extrapolative) colour to Eric’s comment:

in the case of some external graphics configurations (eGPU) which are effectively limited to PCIe x1 bandwidth, you can still achieve ~60% or better performance (under intense scenarions like gaming) of what can be recorded by the very same card in a x16 internal slot. Several tech sites have run similar demonstrations of the performance impact.

4

Okay, I understand the first link - this may or may not have a positiive benefit, ymmv… I’d like to try however.

The second link is an Ubuntu one (which I’d seen and tried already) - I’d added the quoted option into /etc/modprobe.d/99-local.*conf *for opensuse rather than the radeon-kms.conf they refer to in the link. I don’t know if I need to then run something so that these changes are picked up - as setting agpmode to 4 in 99-local.conf just seems to be ignored. The linked Ubuntu thread shows them running an initramfs script which doesn’t exist on opensuse.

Apologies for any inaccuracies in this post - at work and not at my Linux laptop. Any inaccuracies in the original post are entirely my fault!

5

Fair enough lol! . My back of envelope thoughts (using some very rudimentary assumptions) figure you’d get, at best, a ~13% performance improvement under those scenarios/apps which could make use of the wider pipeline. Guessing that under ordinary situations, the difference in settings will be quite indistinguishable.

The second link is an Ubuntu one (which I’d seen and tried already) - I’d added the quoted option into /etc/modprobe.d/99-local.*conf *for opensuse rather than the radeon-kms.conf they refer to in the link. I don’t know if I need to then run something so that these changes are picked up - as setting agpmode to 4 in 99-local.conf just seems to be ignored. The linked Ubuntu thread shows them running an initramfs script which doesn’t exist on opensuse.
Try “mkinitrd”

6

Thanks, that did it - I don’t play with this part of Linux too often. It’s not a benchmark, but glxgears has gone from 25 to 50 fps. and more importantly to me it’s removed the laggy behaviour I’d see when trying to scroll or drag a window.

…and just to prove it;

1.341996] Linux agpgart interface v0.103
1.342123] agpgart-intel 0000:00:00.0: Intel 855PM Chipset
1.356048] agpgart-intel 0000:00:00.0: AGP aperture is 256M @ 0xd0000000
8.053614] agpgart-intel 0000:00:00.0: AGP 2.0 bridge
8.053626] agpgart-intel 0000:00:00.0: putting AGP V2 device into 4x mode

Thanks again

Russ.

7

See, just like I said it would … and you said there would be no noticeable impact, pfffft rotfl!

I was in a rush yesterday and didn’t think of it at the time, but another way that you could achieve this is via the boot parm: radeon.agpmode=whatever