the total failure of and lousy IQ caused by post-Voodoo5 AA and filtering methods

[Anarchist420]

Anisotropic filtering has actually always looked phony; it interferes with trilinear filtering because it compounds it.

Larger textures with VQTC ratios not insanely high and 4x sparse grid super sampling with proper LOD bias adjustments were a better idea. I realize that FXT1 was S3TC based and that 8 bit palettized textures looked lousy. However, lower-bit and lower-res lossless textures are too often a better option than any block texture compression formats that have been dominant for the past 15 years. I know (as do many others) that some implementations were worse than others but I also observed that 98% of the time they were all rude-smelling farts.

MSAA is only good for geometry; and using other than 4 samples of it was just patchwork. FXAA and TXAA didn't look all that good to many people and they looked downright lousy to some people including the man with the AT forums user name Anarchist420.

Anyway, I think the most logical solution starting now is for the IHVs to make OpenCL renderers that do all the rendering esp given the horsepower and computational features in the R290 or even GTX titan.. they could def balance the rendering load quite well between either of those and a 4790K.

I mean, even considering programming any more graphics in DX and OpenGL is a terrible idea considering that games will look better and perform better NOT using tex units, depth units, tess units, ROPs or any other hw rendering units.

 

[Red Hawk]

And what gave you these ideas, Anarchist?

 

[BFG10K]

Anyway, I think the most logical solution starting now is for the IHVs to make OpenCL renderers that do all the rendering esp given the horsepower and computational features in the R290 or even GTX titan.. they could def balance the rendering load quite well between either of those and a 4790K.

Again, why keep posting here about it?

Go right ahead - implement a hybrid software/shader renderer that avoids all fixed function GPU units, and demonstrate to the industry the "superiority" of your ideas.

While you're at it, be sure to use a free Linux distribution and code it on a Larrabee, then give it all away for free so anyone can profit from your efforts without restriction.

I mean, even considering programming any more graphics in DX and OpenGL is a terrible idea considering that games will look better and perform better NOT using tex units, depth units, tess units, ROPs or any other hw rendering units.

LOL.

Again, demonstrate this brilliant code of yours running a real game to convince us and the industry.

 

[Pottuvoi]

Certainly would love to hear what kind of renderer he is proposing.
Surfel splatter, Reyes/Micropolygon renderer, raytracer (voxel, solid, polygon) or something else..

We know there is at least one polygon rasterizer using cuda.
https://research.nvidia.com/publication/high-performance-software-rasterization-gpus

 

[NTMBK]

Feel free to write a compute based raytracer. I recommend taking a look at Embree for a good starting point.

Good luck getting it to run at 1080p 60fps though.

 

[ThatBuzzkiller]

Anyway, I think the most logical solution starting now is for the IHVs to make OpenCL renderers that do all the rendering esp given the horsepower and computational features in the R290 or even GTX titan.. they could def balance the rendering load quite well between either of those and a 4790K.

I mean, even considering programming any more graphics in DX and OpenGL is a terrible idea considering that games will look better and perform better NOT using tex units, depth units, tess units, ROPs or any other hw rendering units.

This is a horrible idea ... There's no need to make the graphics pipeline fully programmable. Even with Larrabee it also didn't go to the extreme of been fully programmable either since it too had texture units to accelerate the pipeline.

 

[Anarchist420]

Certainly would love to hear what kind of renderer he is proposing. Surfel splatter, Reyes/Micropolygon renderer, raytracer (voxel, solid, polygon) or something else.. We know there is at least one polygon rasterizer using cuda. https://research.nvidia.com/publicat...erization-gpus

thanks however, the paper linked to is 3 yrs old so i plan to go ahead with a new and original OpenCL software rasterizer (raytracing faces an uphill battle if it wants to reign supreme for even 1/4 as long as rasterization has) which is potentially better than CUDA apps (and anything else closed including D3D which screwed pc gamers since HW T&L first came out in the first GeForce a full 15 years ago) given that i could never really contribute to the future of CUDA unless I started working for nvidia. but i could never consider working for nvidia for one minute so...

i'll use OpenCL for sure given my democratic, egalitarian (except i will always prefer matriarchal rule by the most polyandrous and non-moralistic women), anti-Federalist soul all going into a Good Future.

 

[hawtdawg]

At first glance, this thread appears to be a product of Autism.

 

[waffleironhead]

At first glance, this thread appears to be a product of Autism.

Op has admitted elsewhere to being off of his meds.

 

[Anarchist420]

Op has admitted elsewhere to being off of his meds.

actually i was just discharged with a diagnosis of Autism Spectrum Disorder and the doctor recommended no more anti-psychotics. i hope that doesn't get me hurt but it really shows how ignorant if not mindless followers that some people are.

thank you

 

[AdamK47]

You don't have extra-ordinary powers of perception. I know you're struggling with it. There isn't anything special or hyper-perceptual about you that makes you stand out from the rest of humanity.

Rotated Grid SSAA was good. It only took you so far though. Resolutions were steadily increasing. The use of programmable shaders was increasing. The performance penalty of SSAA was tremendous. Tradeoffs where made to provide a manageable FSAA with the advancements in shader technology and resolution increases. Internal complexity won out. SSAA is too expensive to achieve acceptable framerates.

 

[Zodiark1593]

MSAA is only good for geometry; and using other than 4 samples of it was just patchwork. FXAA and TXAA didn't look all that good to many people and they looked downright lousy to some people including the man with the AT forums user name Anarchist420.

Anyway, I think the most logical solution starting now is for the IHVs to make OpenCL renderers that do all the rendering esp given the horsepower and computational features in the R290 or even GTX titan.. they could def balance the rendering load quite well between either of those and a 4790K.

SSAA (FSAA) fell out of favor for a reason (hint: it's obscenely expensive, no matter what hardware you're running.). I don't suppose you can propose an AA solution that is fast enough for modern games, yet provides similar quality to SSAA, can you?

 

[gdansk]

Trash. There is and has been, already, a continual march toward less fixed functionality on GPU and more general purpose units.

There are already OpenCL and CUDA raytracers: Nvidia has iray and AMD has showed off theirs at the last SIGGRAPH. Who knows how many 3rd party OpenCL/CUDA raytracers are out in the wild. But you're free to write another, just don't expect it to be any bit revolutionary or advancing the frontiers of rendering.

 

[Anarchist420]

You don't have extra-ordinary powers of perception. I know you're struggling with it. There isn't anything special or hyper-perceptual about you that makes you stand out from the rest of humanity.

thank you; i don't think anyone disagrees

Rotated Grid SSAA was good. It only took you so far though. Resolutions were steadily increasing. The use of programmable shaders was increasing. The performance penalty of SSAA was tremendous. Tradeoffs where made to provide a manageable FSAA with the advancements in shader technology and resolution increases. Internal complexity won out. SSAA is too expensive to achieve acceptable framerates.

It may have been too expensive some, but not to everyone.

SSAA (FSAA) fell out of favor for a reason (hint: it's obscenely expensive, no matter what hardware you're running.). I don't suppose you can propose an AA solution that is fast enough for modern games, yet provides similar quality to SSAA, can you?

how about each end user determines what is best for herself or himself? not trying to be a wise ass but expecting the HW devs to come up with one happy balance in AA for any game sounds like a pipedream that could never end, at least to me.

so, one could've logically predicted that i propose that future GPUs not do AA in HW (in fact, I'll never understand why AMD added MSAA back to the ROPs in the 4850).

so i propose 1) game developers program their own AA methods for each game or engine they make as well as 2) independent software solutions being developed and used by the masses-- SMAA injector is pretty cool and SMAA is actually pretty good looking.

 

[JeffMD]

Do you think we will need AA when we hit 4k? We aren't all going to suddenly game on 42 inch monitors, the DPI of 4k on traditional 24-27inch pc screens could reach imperceptible levels of the human eye.

 

[Pottuvoi]

so i propose 1) game developers program their own AA methods for each game or engine they make as well as 2) independent software solutions being developed and used by the masses-- SMAA injector is pretty cool and SMAA is actually pretty good looking.

On console side this is already happening due to more programmability available in GPUs.
IE. access to vertex interpolators in pixel shaders. (allowing cheap GBAA and so on.)

 

[Anarchist420]

yea, i forgot to mention more evidence that HW-based rendering is inferior overall: i read that the guy(s) who programmed RE4 for Gamecube utilized Gekko (i.e., the CPU) for lighting more than the fixed function T&L unit in the flipper.

On console side this is already happening due to more programmability available in GPUs. IE. access to vertex interpolators in pixel shaders. (allowing cheap GBAA and so on.)

Yep, actually now that i can think back about it and clearly so, i can logically conclude that King Kong Gamers Edition (released all the way back in 2005) for the PC did its AA (that was said to be perfect as it neither smeared sperm anywhere nor was did it leave a single jaggie remaining with only DX 9.0c shaders given that i am pretty sure that it used a Floating point back buffer (or float render targets or both float back buffer and float RTs) and that i don't remember reading about quality or performance issues or differences between ATi's R500 series (as of which the 1950XTX was the top) and NV40/G70 hardware (like the Geforce 6800 to the Geforce 7950 series)

So... the fact that it looked so good yet was abandoned for so long is a massive mystery (at least in my mind). Fortunately and most cheerfully, it's coming back and to stay

Do you think we will need AA when we hit 4k? We aren't all going to suddenly game on 42 inch monitors, the DPI of 4k on traditional 24-27inch pc screens could reach imperceptible levels of the human eye.

Well, given that more pixels mean more detail, the AA will be needed MORE rather than less. In fact, the jaggies were there when i played through most of Indiana Jones and the infernal machine on my older brother's college PC with a Rage 128 and PIII 450 (he brought it home for the first summer he was a college student and he was happy to let me use it without restriction to not only play get to the nub's tomb chapter with some help, but also to watch along with cousins the world is not enough, Deuce Bigalow, Austin Powers the Spy who shagged me and some other movies that were new to DVD back in day; really looked a lot better on that setup than the Panasonic 2nd generation cheap ass model DVD player my parents bought that summer to go with an RCA colortrek bought at the beginning of 93) BUT...

at 800x600 (the MAG monitor was no larger than 17" and i actually think the Rage 128 processor was on a board with only 16 MB so 800x600 was perf) or at least the how rendering or the driver settings i used were back then, the jaggies really weren't annoying... they were nothing like the sensory issue both aliasing and the cheap ass newer AA techniques like TXAA and FXAA are on modern hardware.

Then from almost 2 years later, I don't even remember jaggies (or texture shimmering) being a problem with Return to Castlewolfenstein or PoP3D on a short term system with Pentium III 500 to which i added a visiontek (really reference) Geforce 4 4400 ti.

I really regret that instead of adding (well actually the good folks at DC Software put it in and offered as much as support anyone could want) a Voodoo5 5500 PCI (PCI version b/c i liked the colors on the box better than the almost all orange AGP version) to it instead as i initially wanted to, but I couldn't find any new at any Brick and mortar stores. so the GF 4 Ti 4400 was a sweet ass deal right at compusa; plus it had just come out and my parents wanted to pay for the middle SKU and i was very happy to buy it.

thanks

 

[Phynaz]

Wow

 

[Deders]

actually i was just discharged with a diagnosis of Autism Spectrum Disorder and the doctor recommended no more anti-psychotics. i hope that doesn't get me hurt but it really shows how ignorant if not mindless followers that some people are.

thank you

You're not alone, I bet half the people here are somewhere on the autistic spectrum, myself included. No one else seems to take such an intense interest in computers.

On the subject, have you seen how well temporal SMAA works? and what do you think of the ubersampling option in Witcher 2? It goes beyond standard AA and AF.

 

[Rakehellion]

As soon as I saw "Voodoo5" I knew this was an Anarchist420 thread. Most people don't stare at screenshots and wonder why three pixels aren't where they should be.

You understand that everything a GPU does is downscaled to reduce computational cost? All calculations are lower precision than you'd get with software rendering.

 

[thilanliyan]

So is a Voodoo5 worth anything these days?

 

[Magic Carpet]

So is a Voodoo5 worth anything these days?

A card like this, at least hundred bucks, I would imagine. I have a few. To give you an idea. You can get a brand new 5500 PCI here, for a cool nine hundred bucks. The PCI version is my favorite, since you can use it in a modern PC. The boxed one is at least a hundred more. Some samples I've seen go for ~$300. But it does gain value, so... only time will tell. There is also a MAC version (Lava should get it) which has an extra DVI output. And unlike later cards (such as V5 6000/Rampage), this one is fully useable. V5 6000 all have some bugs, only the latest revision (which I can't tell off my memory) was retail-ready, but those cards are at least a few G's today. So yeah, if you find something Voodoo in your basement, don't just throw it away!

This one has received a fan upgrade:

2) Can't add much to the OP, but NFS 2 SE w/ 4xFSAA on a V5 looked great.

 

[thilanliyan]

^Cool, I have a V5 5500 lying around somewhere. Maybe I'll hold onto it for a while longer. Bought it for a retro gaming PC but haven't had the time to put it together.

 

[Anarchist420]

You understand that everything a GPU does is downscaled to reduce computational cost? All calculations are lower precision than you'd get with software rendering.

Yes, but given how fast general purpose processors can be there isn't much need to make any trade-off given how much more versatile software rendering can be.

On the subject, have you seen how well temporal SMAA works? and what do you think of the ubersampling option in Witcher 2? It goes beyond standard AA and AF.

never played the witcher 1 or 2. Used SMAA with Matrix Path of Neo; it's not perfect, but I was surprised at how good it looked. Programmable AA is definitely the future; continuing to use ROPs for AA is not a good idea as we have seen how MSAA doesn't even look that good in many if not most recent games.

You're not alone, I bet half the people here are somewhere on the autistic spectrum, myself included. No one else seems to take such an intense interest in computers.

Glad to hear that i am not alone. Thanks

 

[bystander36]

Yes, but given how fast general purpose processors can be there isn't much need to make any trade-off given how much more versatile software rendering can be.

Look, the heart of everything is here. It is all about compromises. If you push for perfect precision and high cost AA methods, then other areas of IQ suffer. We can't have all the other details we get with lighting, shadows, tessellation and everything that goes into a game without compromises. If you go bananas on AA, something else must be dialed back.

Everything is a compromise to get the most realistic look possible, with the least computational power. It most certainly is a needed trade off if you want everything else to look good.