Recent Radeon X1K Memory Controller Improvements in OpenGL with AA

We should ask mjtdevries... and perhaps I should edit my previous post... I just assumed that it was mentioned in the firingsquad article, but I did not manage to find anything on AI there...
 
Last edited by a moderator:
The fixes are generic to OpenGL, its nothing to do with Cat AI - you gain a performance improvement under D3 with or without AI enabled.
 
Dave Baumann said:
The fixes are generic to OpenGL, its nothing to do with Cat AI - you gain a performance improvement under D3 with or without AI enabled.
Click to expand...

This has to do with the new ringbus(memcontroller) right so older cards dont benfit from it correct?
 
I thought I read somewhere that someone from ATI had said that. But reading the threads about this again, I can't find it anywhere, so I was probably mistaken.



Btw this patch has started me thinking about ATI's OpenGL performance in general. Many people think their drivers need a rewrite to get better OpenGL performance. And even ATI seems to acknowledge that their opengl drivers should be improved.

But now we have good performance because of tweaking of the hardware, not because of improvements in the OpenGL driver code.

So does that mean that we can expect another performance leap when they improve the driver? Or does it mean that there was never anything wrong with the driver in the first place?

And if the former is true, can we then also expect similar performance gains in D3D games from this tweaking of the memory controller?
 
mjtdevries said:
Btw this patch has started me thinking about ATI's OpenGL performance in general. Many people think their drivers need a rewrite to get better OpenGL performance. And even ATI seems to acknowledge that their opengl drivers should be improved.

But now we have good performance because of tweaking of the hardware, not because of improvements in the OpenGL driver code.

So does that mean that we can expect another performance leap when they improve the driver? Or does it mean that there was never anything wrong with the driver in the first place?

And if the former is true, can we then also expect similar performance gains in D3D games from this tweaking of the memory controller?
Click to expand...

My understanding was that D3D game performance could also be improved, and that any changes to the OGL driver code itself would/should bring incremental gains. But unless I've missed something, this doesn't seem to have been clearly spelled out yet (Eric Demers was insinuating that there's still a lot of testing to be done with different titles and board configurations).
 
mjtdevries said:
I thought I read somewhere that someone from ATI had said that. But reading the threads about this again, I can't find it anywhere, so I was probably mistaken.
Click to expand...

Well, Wavey's review says that they have the ability to do "individual application profiles" with the MC settings, so perhaps you jumped to conclusions on this particular optimization from there and where "individual application profiles" usually live in ATI's world --except, of course, turns out this one isn't for an individual application. Maybe later tho.
 
Last edited by a moderator:
overclocked said:
This has to do with the new ringbus(memcontroller) right so older cards dont benfit from it correct?
Click to expand...

Correct , sadly , from that review the X800 XL showed no improvement and in fact lost about 1 to 3 fps on most of the tests .
 
mjtdevries said:
normally I hate it when sites test without AF since nobody will ever play a game that way, .......
Click to expand...

******s do .............. lol

Edit : Phanboys do . LOL , I didn't realise that when spelt properly phanboys was a bad word :) .
 
Last edited by a moderator:
I'm really surprised that ATI is doing so well in Doom3/Quake4 now. Between NVidia's free FP16 norm and flexible early-z system, I figured there was no way ATI would catch up. Does anyone know if ATI implemented a min-max Hi-Z scheme this time?

I also noted this from sireric:
sireric said:
X800's also have partially programmable MC's, so we might be able to do better there too (basically, discovering such a large jump, we want to revisit our previous decisions).
Click to expand...
Imagine if R4xx could also have had large improvements through similar tuning. I mean really, this oversight could have cost ATI untold millions in sales.
 
Mintmaster said:
I'm really surprised that ATI is doing so well in Doom3/Quake4 now. Between NVidia's free FP16 norm and flexible early-z system, I figured there was no way ATI would catch up. Does anyone know if ATI implemented a min-max Hi-Z scheme this time?

I also noted this from sireric:

Imagine if R4xx could also have had large improvements through similar tuning. I mean really, this oversight could have cost ATI untold millions in sales.
Click to expand...

http://appft1.uspto.gov/netacgi/nph...87".PGNR.&OS=DN/20050195187&RS=DN/20050195187

16. The method of claim 11 further comprising: (e) updating a hierarchical Z value range and a stencil code in response thereto.

17. An apparatus for hierarchical Z buffering and stenciling comprising: a comparator; a hierarchical Z buffer and stencil cache operably coupled to the comparator; and a hierarchical Z buffer and stencil cache updater operably coupled to the comparator wherein the hierarchical Z buffer and stencil cache provides a cache MinZ, cache MaxZ, and stencil code to the comparator.

18. The apparatus of claim 17 further comprising: a tile comprising a plurality of pixels wherein the tile has a tile MinZ and a tile MaxZ.

19. The apparatus of claim 18 wherein the comparator receives the tile MinZ and the tile MaxZ and compares the tile MinZ and the tile MaxZ to the cache MinZ, the cache MaxZ, and the stencil code to determine if a per-pixel depth operation needs to be performed and to determining if stencil operations need to be performed.
Click to expand...
 
rwolf said:
http://appft1.uspto.gov/netacgi/nph...87".PGNR.&OS=DN/20050195187&RS=DN/20050195187
Click to expand...
Well, I know ATI was planning a min-max Hi-Z/Hi-stencil scheme for the original R400, so I don't think that tells us a whole lot about whether R520 has this or not. It would be nice if ATI releases an optimization guide like they did for the R3xx series.

Unknown Soldier said:
Ye . .it would be great if Pre-X1800 cards also got a huge boost.
Click to expand...
Yeah, great, but also sad in a missed-opportunity sort of way.
 
Mintmaster said:
It would be nice if ATI releases an optimization guide like they did for the R3xx series.
Click to expand...
If you download the October SDK from ATI you get these documents, amongst others:
  • HDR Texturing
  • Programming for CrossFire
  • Radeon X1x00 Programming Guide
Sadly it's a 340MB download. But at some point I expect these'll make it into the wild as individual documents. Maybe they're already out there.

Jawed
 
Rys said:
You certainly are. I'll try and duplicate his results later on. Wavey (and anyone else with the hardware), are you seeing the same?
Click to expand...

Here's some further analysis but, from the results its not the mem map fix thats increasing the non-AA scores, but the newer 8.183 drivers in general. The 8.183's include the mem-map fix but they also appear to improve things further; I've heard that there was something in the drivers for R300/R420 that was now detrimental for R520 and this was removed - it could be changes to how the HeirZ is handles, as Mintmaster is making reference to.
 
You must log in or register to reply here.

Similar threads

T
  • Locked
Tool from ATI boosts Doom 3 (and allegedly other OpenGL) AA performance by up to 35%
Replies
2
Views
2K
Nite_Hawk
N
T
Purevideo and AVIVO
2 3
Replies
48
Views
9K
dman777
D
D
Design lessons from PS2 applied to PS3
2
Replies
33
Views
6K
jpr27
J
M
Xbox 360: First Look at Final Hardware
6 7 8
Replies
157
Views
30K
expletive
E
H
IBM backstabbing sony and neglected Apple ? (long read)
5 6 7
Replies
126
Views
44K
Fox5
F
Back
Top