[nelg]

Thanks Deano, that answers my question.

[arjan de lumens]

Originally Posted by DiGuru

If we talk about precision, why use MIP-maps and detail textures? Why not store all textures in FP 32? And why 32 bits? Just because it seems large enough, it is an easy multiple of 8 and most CPUs use them? In that case, you could make a very convincing case for 80 bits. And why not 100 bits then, just to make sure? And require four such values to be computed at the same time, to ease the use of calculating vectors and color values...

Everything is gradual and needs to fit the task it is designed for.

The problem solved by mip-mapping (aliasing/popping/sparkling because features in the texture map become smaller than 1 pixel; mip-mapping is a somewhat crude way of simulating the process of taking multiple samples per pixel) is not the same problem that FP32 would solve.

[Chalnoth]

Originally Posted by DiGuru

If we talk about precision, why use MIP-maps and detail textures? Why not store all textures in FP 32? And why 32 bits? Just because it seems large enough, it is an easy multiple of 8 and most CPUs use them? In that case, you could make a very convincing case for 80 bits. And why not 100 bits then, just to make sure? And require four such values to be computed at the same time, to ease the use of calculating vectors and color values...

For storage, you want to store objects in power of 2 sizes to make the memory controllers simpler.

[Reverend]

Originally Posted by DeanoC

And yes having to go through my shader that work full speed on ATI, reducing things to FP16 on GFFX is a pain.

Can you expand on this? What is your particular shader where precision can matter a great deal (or not) to you? What does "full speed" mean in this context?

[DeanoC]

Originally Posted by Reverend

Can you expand on this? What is your particular shader where precision can matter a great deal (or not) to you? What does "full speed" mean in this context?

It means when all the shaders are working at a good FPS on ATI or NV40 cards but sucks on GFFX, I have to go through the shader code looking for places to partial precision it.

This is harder than it sounds as unlike teapot renderers, many games (i.e. Valve, Crytek, Ninja Theory) are using auto-generated shaders. This make adding partial precision much harder as truncating the precision too early in the shader code, can look fine on some shaders and rubbish on the more complex ones. Our current system has the ability to override (by material name) shaders from the auto-generated ones but that takes work to a)find which shader need optimising and b)optimise them.

As the majority of the cards we are targeting (ATI R3x0, R4x0 and NV40) generally don't need this work, having to do it just for NV3x is a pain.

I think the problem that often missed in this discussion is that in games we don't really work on 'a' shader but lots. I don't actually know how many pixel shaders we have but I know that total shaders (vertex and pixel) is over 6000.

[hovz]

deano what company do u work for and what game are u developing

[DeanoC]

Ninja Theory (previously known as Just Add Monsters) and the game is Heavenly Sword.

[Chalnoth]

Originally Posted by DeanoC

Ninja Theory (previously known as Just Add Monsters)

Sounds like a theory about MMORPG's and looting practices....

[Arun]

Originally Posted by Chalnoth

Sounds like a theory about MMORPG's and looting practices....

ROFL!

[DegustatoR]

Originally Posted by DeanoC

As the majority of the cards we are targeting (ATI R3x0, R4x0 and NV40) generally don't need this work, having to do it just for NV3x is a pain

Isn't NV4x benefit from this also?

[hovz]

deano are there any pics of your game? couldnt find any on ign.

[Ostsol]

Originally Posted by hovz

deano are there any pics of your game? couldnt find any on ign.

http://www.eurogamer.net/article.php?article_id=56501

[Frank]

Originally Posted by arjan de lumens

The problem solved by mip-mapping (aliasing/popping/sparkling because features in the texture map become smaller than 1 pixel; mip-mapping is a somewhat crude way of simulating the process of taking multiple samples per pixel) is not the same problem that FP32 would solve.

Yes, I know that. But you store multiple sizes, to be able to use the one that fits the need. That is analogous to the bit depth: you use the one that looks good enough and uses as little space as possible. You might WANT to use the largest textures you can to cover everything, but it is not feasible, unless you think four textures are enough for everything...

[Frank]

Originally Posted by Chalnoth

For storage, you want to store objects in power of 2 sizes to make the memory controllers simpler.

That depends. You might want to use a size that has little spill-over for the cache-filling bursts. It might be optimal to use 21 bit color values, 7 bits per color, to be able to use the bandwith to it's full potential in many scenario's. I don't know. As long as the total burst size transfers the data you need and fits a multiple of the bus width, it can be more optimal than an even amount of bytes/words/longs.

Btw. How would you do a 32/48/64/80 bit computation on a R300? I think it probably isn't that hard, so if you need it for some special cases, you might want to use macros that do multi-word calculations.

[Chalnoth]

Originally Posted by DiGuru

Btw. How would you do a 32/48/64/80 bit computation on a R300? I think it probably isn't that hard, so if you need it for some special cases, you might want to use macros that do multi-word calculations.

I don't think current architectures support the bitwise operations you'd need to perform higher-precision operations than officially-supported. You could do some packing to get somewhat better precision, but you won't get FP64 with two FP32 registers.

[Unknown Soldier]

Originally Posted by hovz

deano are there any pics of your game? couldnt find any on ign.

Hovz .. there was also a trailer (or 2) that was released if I remember correctly.

Yes I'm pretty sure. Wasn't there DeanoC?

US

[Unknown Soldier]

http://www.beyond3d.com/forum/viewto...d&start=20

US

[Geo]

Originally Posted by DeanoC

I think the problem that often missed in this discussion is that in games we don't really work on 'a' shader but lots. I don't actually know how many pixel shaders we have but I know that total shaders (vertex and pixel) is over 6000.

Any sense how many person-hours (or person-days, if that is easier) were/will be required to do the analysis of which shaders need to be switched to partial-precision and the actual coding to do so for your game?

[nelg]

Originally Posted by geo

Any sense how many person-hours (or person-days, if that is easier) were/will be required to do the analysis of which shaders need to be switched to partial-precision and the actual coding to do so for your game?

Good idea. Then they could calculate how large of a bill to send to nV.