Two new demos on my site
- Thread starter Humus
- Start date
Bigus Dickus
Regular
Well, fragment AA that could also AA polygon intersections and texture edges as well would be way cool. 
SS effect but still only requiring AA on a fraction of the screen.
SS effect but still only requiring AA on a fraction of the screen.
May I suggest you try these if you haven't already:
http://www.ati.com/support/infobase/4081.html
They fixed the problem for many other 9700 users.
http://www.ati.com/support/infobase/4081.html
They fixed the problem for many other 9700 users.
Bigus Dickus said:Well, fragment AA that could also AA polygon intersections and texture edges as well would be way cool.
In Fact... this kind of AA exists... And I am proud inventor of it... I am not sure where the development is now, because since I was busy, I just dropped the idea one of my friends at September 2001... well.. better tell the whole story...
This mixes up so many ppl anyways...TIME OF BRAINSTORMING
------------------------------
about year ago, August, I was in unbeliable brainstorming mode. Got huge amount of ideas new kind of implementations of different things on Hardware 3D... I think all got on it's way when I got some Info about upcoming Axe chip. So basically at least first, things gotten started with thoughts 'How I would do this...' But because never got anything out of from the boys at Espoo, never got confirmation if my ideas were right... Anyways, when you get started you just can't stop. So I started to ruin my finnish lessons and my notebook got all kind of wierd drawings.
The real first try to get things rolling was my invention eDRAM based Dual chip system. Right now even I don't remember how it would have been worked, but this was one of the first things that gotten a slight eye brown lift at Bitboys. I actually got feedback from it after introducing it to some fellas over there. still they were able to dig up some heavy failures my desing had, but still, it was a giant compliment to me. Here at least Ceiser Söze and Humus can confirm seeing a funny digital photographed drawing representing 2 eDRAM chips with 12 Megabytes of eDRAM on each. To introducing it here, I would definately need to dig up that photo and look for ICQ logs... but maybe more about that later.
Next thing was the NaNodiaAA... while thinking multichip I somehow got an idea of AA.. and again, it needs a lot bandwidth. idea was to have completely independent AA unit on the end of IMR. Idea was get pixels in the 3x3 tiles AFTER the whole frame was finished and check on RGBA level if there was need for AA. if yes, then take 2 samples from middle, and sum on the directions that had need for AA.
At this point things started to fly a way fast and I needed someone to keep brainstorming on the line. I decided to talk about my idea one of my friends. First, he said that he really wasn't up to this kind of thing, but after second cup of coffe on our school's caffeteria, he started to interest. On next day we had very simple software simulation without 2 samples from the middle so it was practically just a blur. Still results were encouraging and we continued collecting ideas.
On the next week, we had approximate gate design for Blur version and quick calculations showed that on eDRAM system we could have 27 ALUs in paraller for working AA (or Blur so far.) with 175Mhz clock rate that would have mean about 800M pixels with Blur enabled. On this point I introduced our idea to Dave Barron (it was propably November already.) and I will propably remember his ICQ comment rest of my life: '27 in paraller?? that's insane. I don't think you are lost at all...'
we were still missing the AA and my school & other things started to take more time, so I pretty much left development for Tommi Paju. He actually made pretty good progress on Algorithm development during winter and even with the 2 subsamples on middle pixel, it should not slow it much down. right now I know that he has practically made all over again, so only basic idea of making AA to ready frame is only left. Name is changed to NordiaAA, but I have to be proud of starting this. we were to close contact at BB all the time of development and I don't know much more the situation, but I know that everything exists still.
The future?? I really don't know. The fact is that developing that AA is endless task. you can continue developing it further and further, but atm there's no manufacturer who could use it on reasonable speed. Just because it's bandwidth needs are endless. for nowadays DRAM busses.
and what comes to my brainstorms I ocasionally get, there's few ideas again, but haven't thought much further those... Major one is making a ShadowCaster in the end of pixelpipelines. (one to each other.) So after rendering the whole pixel, this part would check if any polygon is blocking the line between point being rendered on polygon and each of the lights. (needs HW light setup, but so does everything nowadays.) if yes, then calculation of shadow color with blocking polygon's texture (needs Texel fetcher... point sampling would be good enough for start.) and combining to basic pipeline's result. If no blocking ocurs, continue with next pixel. This needs some ideas and thinking and development so it's not anywhere near far as multichip and AA...
I really don't know if I have been re-inventing things but that happens when you got brainstorm.
I hope that you liked my little story, if someone bothered to read it thru. and oh... if anyone is more interested about NordiaAA, let me know and I'll try get Tommi here to tell more about it, where it is now and so on...
Recheck that link Hyp-X.
It is no longer the D3D-Patch only, but a link to the 7.77 drivers (full 21MB bundle). It includes the D3D patch inside, but also has a new mini-vdd and a few other new changes (INF and the like). It also includes the drivers + CP all in one handy install.
It is no longer the D3D-Patch only, but a link to the 7.77 drivers (full 21MB bundle). It includes the D3D patch inside, but also has a new mini-vdd and a few other new changes (INF and the like). It also includes the drivers + CP all in one handy install.
EvilEngine
Newcomer
Too bad this doesnt work for NV cards
*off-topic*
Humus, your shadows demo is now 3x faster than before on my GF3 with the newest display drivers from Nvidia. Thought I would let you know incase any geforce users ask for tech support. Heres a screenshot:
*off-topic*
Humus, your shadows demo is now 3x faster than before on my GF3 with the newest display drivers from Nvidia. Thought I would let you know incase any geforce users ask for tech support. Heres a screenshot:
EvilEngine
Newcomer
It has 8x aniso and -1.5 lodbias.
If I set everything to the lowest possible levels the fps doesnt change. Same if I pump the AA to 4x or 4x+9tap with 8x aniso and -3 lodbias, no fps change.
If I set everything to the lowest possible levels the fps doesnt change. Same if I pump the AA to 4x or 4x+9tap with 8x aniso and -3 lodbias, no fps change.
EvilEngine said:Too bad this doesnt work for NV cards
*off-topic*
Humus, your shadows demo is now 3x faster than before on my GF3 with the newest display drivers from Nvidia.
I suppose "this" refers to the laplace demo? The dither demo should work for GF3.
Nice to see the improvement in fps. The bottleneck is the rendering to cubemap though (at least in default res windowed mode), so that explains why AA and aniso doesn't change the fps much.
