AMD Radeon HD 4870 X2 On Linux

[phoronix]

Phoronix: AMD Radeon HD 4870 X2 On Linux

While details on AMD's high-end ATI Radeon HD 4870 X2 graphics card began surfacing last month, today this dual-GPU graphics card has formally launched and will begin shipping to retailers. The good news this time around is that this graphics card will actually be something worth looking at under Linux now that there's the needed CrossFire support on the horizon. We're not yet permitted to publish the Linux CrossFire benchmarks, but today we have a few pieces of information to share about the Radeon HD 4870 X2 on Linux.

http://www.phoronix.com/vr.php?view=12734

[NeoBrain]

Heh, you sound like you could know when the new driver is out :D

Whatsoever, I'm really looking forward to seeing some benchmarks and how well Crossfire setups scale, keep up the good work!

[d2kx]

I can't even imagine what bad things would happen if the users knew the driver's version you've used :) (I know this has been brought up before)

[quintesse]

Oh sure, Crossfire support! I wish they would support the latest XOrg (the one used by Fedora 9) instead! Got this great new card and I haven't been able to use it AT ALL in the last couple of months. Even video stutters. *sigh*

[NeoBrain]

Oh sure, Crossfire support! I wish they would support the latest XOrg (the one used by Fedora 9) instead! Got this great new card and I haven't been able to use it AT ALL in the last couple of months. Even video stutters. *sigh*

First things first, the demand for crossfire under Linux has been there for a much longer time than the demand for Xorg 7.4 support.
Apart from that, that Xorg "release" is not even out or considered stable yet. The official Catalyst driver, as bridgman once stated, is intended for everyday use on stable systems, which don't use the bleeding-edge development trees, whereas radeon and radeonhd are intended to support them.

[RealNC]

First things first, the demand for crossfire under Linux has been there for a much longer time than the demand for Xorg 7.4 support.
Apart from that, that Xorg "release" is not even out or considered stable yet. The official Catalyst driver, as bridgman once stated, is intended for everyday use on stable systems, which don't use the bleeding-edge development trees, whereas radeon and radeonhd are intended to support them.

If fglrx doesn't work, you are forced to go git master to get your card working with radeon/radeonhd. If ATI later says "we fixed it", it would much easier if I could try it out myself on X.Org 7.4 rather than spending the evening downgrading to 7.3 only to discover that it still doesn't work and then having to upgrade again to git.

[NeoBrain]

If fglrx doesn't work, you are forced to go git master to get your card working with radeon/radeonhd. If ATI later says "we fixed it", it would much easier if I could try it out myself on X.Org 7.4 rather than spending the evening downgrading to 7.3 only to discover that it still doesn't work and then having to upgrade again to git.

hm.. I see where your problem is; however, then you could also stick with Xorg 7.3 and use radeonHD 1.0, so you need only one downgrade at all. But if you insist on having 3d acceleration it's not as easy as "if fglrx doesn't work I use radeon(hd)", exactly because of what you described. One must decide wether he sticks with some 3d accel that works with radeon(hd) or wether he gets full 3d accel IF he gets his fglrx running.
I guess in your case it'd be best to spent ~5GB for a second distribution installation in order to test wether a new fglrx release fixed your problem.

I now that's not really the answer you want to hear, but it's the sad reality that binary blobs don't fit too well in the OSS world ;)

[RealNC]

I now that's not really the answer you want to hear, but it's the sad reality that binary blobs don't fit too well in the OSS world ;)
This is true, of course. But I wish AMD had gone the NVidia way; ignore the standards and do your own thing. On NVidia right now, the Linux Desktop/Gaming experience seems much better. I think there's a reason why you see a "don't use ATI, go NVidia" message when you install Cedega. NVidia seems to have invested more in their binary blobs than AMD has. If you provide a binary blob and refuse to open source it so others can fix it, at least fix it yourself dammit :P

[FunkyRider]

I still have to say buying a 4870X2 for Linux is brain dead. By the time the driver, if there is any, comes out, the 4870X2 is already history.

So forget about it!

[leef]

Agreed. Better to work on 3D (OpenGl, memory management, etc.) if resources are limited. No need to have things not work (up to) twice as fast.

Of course, making the code "Crossfire clean" could have benefits elsewhere.

[jeffro-tull]

wait, didn't AMD redo the codebase for fglrx not too long ago? Didn't Crossfire exist when they did that? Why are they molesting code to get it working now?

I know I shouldn't expect much from fglrx, but this is something they should have seen coming.

[Ellipsys]

I'm so disappointed. I'm ready to buy a 4870x2 for use in a dual boot situation - its the fastest, most powerful graphics card available at the moment and a good price/performance ratio to match. However, if I won't be able to utilize it in Linux at all, with proper 3D closed-source driver support, (even if it just runs 1 "core" that's enough for me), I don't know how I can buy it.

I really wish AMD/ATI would get their drivers up to par - I don't want to buy an underpowered, overly expensive Nvidia card, and I don't want to support their very closed mindset, but I need a $500 graphics card to WORK.

[NeoBrain]

wait, didn't AMD redo the codebase for fglrx not too long ago? Didn't Crossfire exist when they did that? Why are they molesting code to get it working now?

I know I shouldn't expect much from fglrx, but this is something they should have seen coming.

Well, actually they didn't rewrite their whole codebase. It was "just" the 3d part and one more low-level thing in order to improve performance and add AIGLX support. The part was (well, at least mostly) completely independant from any future Crossfire support, so it would have been the same mess anyways.
By the way, the driver was buggy enough with 8.42, I don't even want to imagine how buggy it would've been if they even had Crossfire implemented at that time.

[NeoBrain]

I'm so disappointed. I'm ready to buy a 4870x2 for use in a dual boot situation - its the fastest, most powerful graphics card available at the moment and a good price/performance ratio to match. However, if I won't be able to utilize it in Linux at all, with proper 3D closed-source driver support, (even if it just runs 1 "core" that's enough for me), I don't know how I can buy it.

I really wish AMD/ATI would get their drivers up to par - I don't want to buy an underpowered, overly expensive Nvidia card, and I don't want to support their very closed mindset, but I need a $500 graphics card to WORK.

Well, one important keypoint of this article was to explain that this card will probably be supported by the next Catalyst release. I guess, they either activate support with 8.8 or with 8.9, so you'll be able to enjoy Crossfire fun next month ;)

PS: In regards of using at least one core of the card, this is actually supported, at least after you add the ChipID in the Xorg.conf file.

[jeffro-tull]

Well, actually they didn't rewrite their whole codebase. It was "just" the 3d part and one more low-level thing in order to improve performance and add AIGLX support. The part was (well, at least mostly) completely independant from any future Crossfire support, so it would have been the same mess anyways.
By the way, the driver was buggy enough with 8.42, I don't even want to imagine how buggy it would've been if they even had Crossfire implemented at that time.

right, and fglrx is just awesome now....

sarcasm aside, thanks for the info. I didn't know what they redid, just that they redid stuff.

[#cruncher]

Thus far, the card hits a BIOS post:

Kernel Librray version:03.05.00.03
StrongROM version:03.30 Build:_P
ERROR
System Configuration Data Read Error
ERROR
Warning: IRQ not configured- PCI in slot 04
Bus:83, Device:00, Function:00
ERROR
Warning: IRQ not configured- PCI on Motherboard
Bus:84, Device:04, Function:00
ERROR
Warning: IRQ not configured- PCI on Motherboard
Bus:84, Device:08, Function:00
ERROR

but does later configure with radeonhd driver for minimal functionality. Unfortunately when run through aticonfig to get the fglrx driver (9.1 release, 2009/02/09) the screens are dark. No serious complaints in fglrx-install.log:

>>>>>> more /usr/share/ati/fglrx-install.log
[Message] Kernel Module : Trying to install a precompiled kernel module.
[Message] Kernel Module : Precompiled kernel module version mismatched.
[Message] Kernel Module : Found kernel module build environment, generating kernel modu
le now.
AMD kernel module generator version 2.1
doing Makefile based build for kernel 2.6.x and higher
rm -rf *.c *.h *.o *.ko *.GCC* .??* *.symvers
make -C /lib/modules/2.6.27.7-9-default/build SUBDIRS=/lib/modules/fglrx/build_mod/2.6.x modules
make[1]: Entering directory `/usr/src/linux-2.6.27.7-9_thor'
CC [M] /lib/modules/fglrx/build_mod/2.6.x/firegl_public.o
/lib/modules/fglrx/build_mod/2.6.x/firegl_public.c: In function ‘KCL_MEM_VM_GetRegionPhysAddrStr’:
/lib/modules/fglrx/build_mod/2.6.x/firegl_public.c:3260: warning: return makes pointer from integer without a cast
/lib/modules/fglrx/build_mod/2.6.x/firegl_public.c:3261: warning: return makes pointer from integer without a cast
/lib/modules/fglrx/build_mod/2.6.x/firegl_public.c:3262: warning: return makes pointer from integer without a cast
/lib/modules/fglrx/build_mod/2.6.x/firegl_public.c:3264: warning: return makes pointer from integer without a cast
CC [M] /lib/modules/fglrx/build_mod/2.6.x/kcl_acpi.o
CC [M] /lib/modules/fglrx/build_mod/2.6.x/kcl_agp.o
CC [M] /lib/modules/fglrx/build_mod/2.6.x/kcl_debug.o
CC [M] /lib/modules/fglrx/build_mod/2.6.x/kcl_ioctl.o
CC [M] /lib/modules/fglrx/build_mod/2.6.x/kcl_io.o
CC [M] /lib/modules/fglrx/build_mod/2.6.x/kcl_pci.o
CC [M] /lib/modules/fglrx/build_mod/2.6.x/kcl_str.o
CC [M] /lib/modules/fglrx/build_mod/2.6.x/kcl_wait.o
LD [M] /lib/modules/fglrx/build_mod/2.6.x/fglrx.o
Building modules, stage 2.
MODPOST 1 modules
CC /lib/modules/fglrx/build_mod/2.6.x/fglrx.mod.o
LD [M] /lib/modules/fglrx/build_mod/2.6.x/fglrx.ko
make[1]: Leaving directory `/tmp'
build succeeded with return value 0
duplicating results into driver repository...
done.

Any ideas?

[#cruncher]

As of today, Asus EAH4870x2 came alive for the first go around:

host >>>>> uname -a
Linux host 2.6.27.7-9-default #1 SMP Mon Feb 9 09:47:41 EST 2009 x86_64 x86_64 x86_64 GNU/Linux
kernel rebuilt with Opteron specific CPU support, not 386 generic

Asus EAH4870x2 2G DDR5 >>>>>> glxgears
40636 frames in 5.0 seconds = 8127.194 FPS
41160 frames in 5.0 seconds = 8231.852 FPS
38216 frames in 5.0 seconds = 7643.188 FPS
42733 frames in 5.0 seconds = 8546.575 FPS
42716 frames in 5.0 seconds = 8542.996 FPS
42739 frames in 5.0 seconds = 8547.710 FPS
42710 frames in 5.0 seconds = 8541.937 FPS
42727 frames in 5.0 seconds = 8545.385 FPS
42725 frames in 5.0 seconds = 8544.861 FPS
42729 frames in 5.0 seconds = 8545.745 FPS
42735 frames in 5.0 seconds = 8546.832 FPS
42712 frames in 5.0 seconds = 8542.352 FPS
42733 frames in 5.0 seconds = 8546.586 FPS
42732 frames in 5.0 seconds = 8546.369 FPS
42735 frames in 5.0 seconds = 8546.971 FPS
42763 frames in 5.0 seconds = 8552.535 FPS


Comparing to an older system:
oldHost >>>> uname -a
Linux oldHost 2.6.22.19-0.2-default #1 SMP 2008-12-18 10:17:03 +0100 i686 i686 i386 GNU/Linux

ATI x1650 512M DDR2 >>>>>>> glxgears
12641 frames in 5.0 seconds = 2528.175 FPS
12865 frames in 5.0 seconds = 2572.728 FPS
12598 frames in 5.0 seconds = 2518.102 FPS
12605 frames in 5.0 seconds = 2520.413 FPS
12324 frames in 5.0 seconds = 2464.615 FPS
12398 frames in 5.0 seconds = 2479.584 FPS
12750 frames in 5.0 seconds = 2549.988 FPS
12329 frames in 5.0 seconds = 2465.761 FPS
12157 frames in 5.0 seconds = 2431.369 FPS
11895 frames in 5.0 seconds = 2378.310 FPS
11740 frames in 5.0 seconds = 2347.959 FPS
12091 frames in 5.0 seconds = 2417.982 FPS
13293 frames in 5.0 seconds = 2658.571 FPS
12682 frames in 5.0 seconds = 2536.338 FPS
12643 frames in 5.0 seconds = 2528.545 FPS
12897 frames in 5.0 seconds = 2579.356 FPS

Installed with ATI 9.1 drivers, circa 2009/02/04 and configures to use the crossfire:

aticonfig --list-adapters
* 0. 84:00.0 ATI Radeon HD 4870 X2
1. 85:00.0 ATI Radeon HD 4870 X2

* - Default adapter
--------------------

aticonfig --adapter=0,1 --add-crossfire-chain
CrossFire chain added
Warning: X needs to be restarted before CrossFire changes take effect.
--------------------
aticonfig --list-crossfire-chains

CrossFire chain for adapter 0, status: disabled
0. 84:00.0 ATI Radeon HD 4870 X2
1. 85:00.0 ATI Radeon HD 4870 X2
---------------------

aticonfig --adapter=all --crossfire=on
CrossFire chain(s) enabled
Warning: X needs to be restarted before CrossFire changes take effect.
----------------------
aticonfig --list-crossfire-chains

CrossFire chain for adapter 0, status: enabled
0. 84:00.0 ATI Radeon HD 4870 X2
1. 85:00.0 ATI Radeon HD 4870 X2
---------------------

I had to manually tweak the xorg.conf file (which is typical, I have always had to fix the aticonfig generated one), but this time it fired up.

The BIOS post messages are still there, including the one about "Option ROM out of space".....


I have not fully checked out this config, but wanted to let folks know it is possible. Hopefully more of us can get this vid working and check against more systems/software for performance.

Regards