Catalyst 3.2 drivers released

[31 Flavas]

I just read on rage3d.com that the Catalyst 3.2 drivers have been released. This has relavance here because this is supposed to be the release that addresses some RealOne Player playback issues.

So jump on over to rage3d.com and read up.

[Neo Neko]

Yes. It's about time. Perhaps my 8500 will see a bit of perfomance boost. It has been a while since I updated the drivers.

[symonjfox]

Hi, let's hope they will fix the MPEG 2 Hardware acceleration under WinXP. Does it work for you?

In my system, it worked since the arrive of first Catalyst driver (that boosted a lot the 3D graphic, but disabled at all the MPEG2 hardware acceleration).

I don't wanna return to non-catalyst drivers.

Under WinME it works well.

And you?

<Sorry, I'm out of thread>

[Rash]

Just like magic!

Overlay is on, tearing is gone (100%), image quality is better!

Now I can say, love Helix and ATI. Heheh. ;)

[karl_lillevold]

Yep, I did a quick test today, and the bug related to RV playback appears to be fixed. The reason this occurred only for RV is that ATI has hardware support for the postfiltering in RV8. Since RV9 does not use a separable postfilter, you might wonder why this then matters for RV9... You see, we still use the same overlay color format (similar to YV12, but its own 4CC code), but disable filtering, such that now the ATI driver just copies YUV data and does not apply filtering. In any case, it was this special overlay color format that had a bug.

P.S. it is the same hardware accelerated postfilter that is mentioned here in this press release:
http://www.internetnews.com/infra/article.php/1567311
where "ATI announced a deal with DivXNetworks to combine each other's software to improve the quality of digital video playback."

Again, note that this h/w accelerated postfilter is used for RV8, not RV9, but playback of RV9 would show the same bug since the same playback 'pipeline' is used.

[Sirber]

Everyone get a GeForce 4!!! :D

[midiguy]

was this bug present only for the radeon 8xxx series or also the 9xxx?

[karl_lillevold]

i know it was present for 8xxx, and I am pretty sure it was a problem also for 9xxx, but it should be fixed for both with Catalyst 3.2.

[Baalthazaar]

Just for the record these drivers are not perfect, though they do seem to have fixed the problems I was having with RealVideo. They do introduce a few glaring bugs into the catalyst drivers (Half-Life Engine games crash your computer if you press 'escape' with R300 based cards) and a few smaller ones. On the whole they seem to have fixed a lot, though - I can finally play Neverwinter Nights without massive stuttering (but why would I want to?).

Anyway, be sure to do a few minutes of reading up over at the Rage 3D forums (http://www.rage3d.com/board/) before committing.

[karl_lillevold]

while we were looking into improving performance for HD playback of RV9 (1280x720p @ 30 fps) [would you not like that quality all the time?] it looks like we have found another problem with our ATI card. We noticed that memcpy of all that data to video memory took upwards of 30% of the CPU time, leaving only roughly 2/3 to core decoding. So my co-worker wrote a little test app wrapping the calls we use in our renderer to measure the time it takes to copy various color formats to system memory and/or to video memory. What we found was that this copy process takes 4 times as long on an ATI Radeon AIW 8500DV compared to a GeForce4 Ti 4200. On the GeForce4, the copy to video memory is just a little slower than copy to system memory, while on the ATI, it's about 4X slower.

Looking into ATI's advanced options, they have something called SMARTGART. With this you can adjust AGP speed, and enable or disable "Fast Write". Fast Write was OFF, and after turning it ON and a reboot, it unfortunately goes right back OFF again. This setting may or may not affect this transfer speed. We have communicated the problem to ATI and are awaiting an answer if it is possible to improve this performance problem.

It is, however, likely that for standard TV resolutions, this does not really degrade playback, unless you have a somewhat slow system, right on the edge of where playback is smooth (slower than 800 MHz, and then I am sure you have an older video card anyway).

Here are the timing numbers from my home system copying 1000 frames of 1280x720:

P4 2.2 GHz / 400 FSB, running at 2.4 GHz
nVidia GeForce4 Ti 4400.
format system video
YV12 2414 2431
YUY2 3112 3145
'system' = copy to system memory
'video = copy to video memory

For ATI Radeon AIW 8500 DV, the numbers were
format system video
YV12 1917 9557 *not good*
YUY2 2323 12292 *not good*
The system memory numbers are better due to this being run on a 2.6 GHz i850 system with RDRAM PC800 memory, while my home system uses DDR PC2100 and an i845 chipset.

You can get the zip file with the test utility here:
http://www.lillevold.com/files/cscdiag.zip
We would be curious to know how these numbers vary for older and newer video cards and memory types. If you have time to run the little test, please send me e-mail or PM with the numbers you get. Remember to read the readme.txt file, and change the settings to 1280x720 and 1000 frames so that the numbers are comparable, and include your system information.

[Rash]

Hi there Karl. It really seems that it is a problem with your Fast Writes.

I've been reading about the Catalyst 3.2 disabling this option on machines that used to have it "on" on 3.1. I heard it is not really a bug, just that ATI found out that the drivers shouldn't support that option to that chipset anymore. Fortunately, it is On by default here. ;)

P4 1.6Ghz / 125Mhz FSB, running at 2.0Ghz
ATI Radeon 8500 64Mbs (with Catalyst 3.2)
Format / System / Video
YV12 / 2897 / 3232
YUY2 / 3863 / 4221

[karl_lillevold]

thanks for this useful information! It seems ATI (8500 only?) has a problem with Fast Write and the Intel 850 chipset then, which is too bad considering how the 850 with its RDRAM support is a (somewhat) higher performance chip than the 845. For now it sounds like we need to use nVidia for this demo machine.

[karl_lillevold]

I received feedback from ATI, and just wanted to correct the previous post. It is not the i850 chipset which has a problem with the Radeon, it is simply the Radeon 8500DV which does not work with fast write, and has pretty bad video playback performance, since it has this problem with very slow copy to video memory.

So, the plain 8500 should be fine along with other Radeon flavors as long as it's not the 8500DV (AIW). I think I will get a 9500 for this demo system, or perhaps a GF4 Ti 4200, both at about $140 or so.

[deXtoRious]

Definately use 9500!!! It's way better than GF4! Besides, 9500Pro can be easily changed into 9700Pro at no cost!

[Morbo]

While I was here surfing,I thought Id let you Real guys know my problem.....

My 9700 wont play RV8(cbsnews.com) unless I slide the overlays around and/or back the cpu off too...

I dont recall this happening w/ 3.1,but when I try to play it I get BSOD and/or a reboot.

RV9 gives me no problems at any setting...

I have the cats,control panel,MDAC(version 8),all ati multimedia suite updated......

Also,of course,it does it in a "windowed IE player",and the player itself...

Good Luck!!

System:Asus A7n8x(non-delux),Santa Cruz,Ati 9700PRO,512 2700 @ Agressive....

[karl_lillevold]

Interesting problem. Maybe now we can justify expensing a 9700 or at least a 9500Pro :). 3D capabilities have not been taken into account when we get to buy a new video card.

Seriously, I don't know what this problem could be other than something to do with ATI drivers and the special format we use for the hardware postfiltering on ATI cards. Currently you can not change this format unless you back off on the CPU slider, which you already tried, but a reg key has been added, which I will tell you about when it has been released.

[karl_lillevold]

as it turns out, version 3.2 also had an issue with the special acceleration used for RealVideo. See this thread on rage3d:
http://www.rage3d.com/board/showthread.php?s=&threadid=33675184
As you can see, 9xxx cards had a hard lock-up problem (for certain content). This thread also lists a regkey which can be used as a work-around until new Catalyst drivers are out later this month, in case you have this problem.