FP blending in R420?

[Joe DeFuria]

There's a reason why the move for FP filtering/blending, and it's so that games move forward, not backwards.

Why does NV40 support sport FP16 again?

Because there's a lot of operations that doesn't need more precision ?

Doesn't matter...you're "holding the industry back", right?

 

[Joe DeFuria]

Why does NV40 support sport FP16 again?

If you're asking why is doesn't support FP32 but only FP16 for FP blending, you already know the answer Joe: Way too much memory bandwidth overhead to be usable this time around.

No, I'm asking why nvidia supports FP16 at all. All it does it allow devs to "hold games back" by "settling" for FP16 instead of using the available full precision.

Seriously, of course I know the answer....it's called "making a trade-off". there are economic considerations for supporting one feature or not...supportingone feature over another, etc.

It's the idiodic "company X is holding the industry back" comments from folks like Chalnoth that drove this side-track....

 

[radar1200gs]

There's a reason why the move for FP filtering/blending, and it's so that games move forward, not backwards.

Why does NV40 support sport FP16 again?

Because there's a lot of operations that doesn't need more precision ?

Doesn't matter...you're "holding the industry back", right?

Not at all - FP32 is available when required.

 

[KimB]

No, I'm asking why nvidia supports FP16 at all. All it does it allow devs to "hold games back" by "settling" for FP16 instead of using the available full precision.

Seriously, of course I know the answer....it's called "making a trade-off". there are economic considerations for supporting one feature or not...supportingone feature over another, etc.

It's the idiodic "company X is holding the industry back" comments from folks like Chalnoth that drove this side-track....

If you're trying to make an example in the other direction, it's a very poor one. Most operations on the NV4x can now be executed at full speed at FP32, with some improvements in performance to be made at FP16. There's just no reason for developers to skimp on quality with the NV4x.

This is a completely different issue than not even bothering to support basic functionality, such as FP blending/filtering.

 

[Joe DeFuria]

Not at all - FP32 is available when required.

Nope. It's not available at the same performance, and as mentioned above, not available for all operations. Clearly, nVidia is retarding the industry. Until they drop FP16 completely, gamers will languish in gaming mediocrity...

 

[radar1200gs]

Not at all - FP32 is available when required.

Nope. It's not available at the same performance, and as mentioned above, not available for all operations. Clearly, nVidia is retarding the industry. Until they drop FP16 completely, gamers will languish in gaming mediocrity...

FP32 may be slower, but it doesn't need to be used everywhere in the first place.

Just because nVidia uses only FP16 for filtering and framebuffering doesn't mean you can't calculate at FP32 in the pixelshaders (where the precision is required) and writeout at FP16.

FP16 is a very luxurious sort of mediocrity - it's good enough for OpenEXR and ILM, and as I've said before I'm not expecting games to outperform cinema effects anytime soon.

 

[Heathen]

This is a completely different issue than not even bothering to support basic functionality, such as FP blending/filtering.

Since when was FP Blending & Filtering considered 'basic functionality'?

 

[radar1200gs]

This is a completely different issue than not even bothering to support basic functionality, such as FP blending/filtering.

Since when was FP Bledning & Filtering considered 'basic functionality'?

Well, if you are content with worse than S3 virge functionality then I guess it doesn't matter...

 

[Dave Baumann]

Mr Sanford - haven't embarrassed yourself with your "technical knowledge" in this forum enough times over the past few months? If you really must continue to post please restrict it to things you know about.

 

[Joe DeFuria]

FP32 may be slower, but it doesn't need to be used everywhere in the first place.

Of course it does. FP24 and FP32 are simply superior to FP16 in all quality aspects. We need it...no substitutions. Stop holding the industry back nVidia!

 

[KimB]

If you want to take that stance, Joe, FP32 is superior to FP24 in all respects.

 

[radar1200gs]

Dave, are you suggesting that shader output can't be placed in a framebuffer with less precision than the shader?

Don't you think most people consider bilinear to be the bare basics for texture filtering? This isn't 1995/96 you know.

Then again, if you think point sampling is yummy, feel free to indulge yourself...

 

[Briareus]

Did ATI leaving out fp blending in the R420 have something to do with their use of fp24 instead of fp32/fp16?

Blending and texture filtering aren't done in the shader, so no, this isn't a reason. Besides, nobody would create a FP24 framebuffer, due to bit alignment issues.

Thanks, I didn't know if they shared any hardware. I'm still learning.

 

[Dave Baumann]

Greg - that post is not helping your case

 

[GameCat]

Obviously, the best precision is the one mandated by IEEE 754 (i.e. 7+5+4=16 bits) Geez, stop giving radar such a hard time...

 

[radar1200gs]

nVidia has brought FP framebuffers and FP filtering anf HDR rendering to consumers by way of proven, established, standardized technologies that have a track record of working.

ATi is making it up as tey go along (especially with FP24), aided and abetted by the king of non-standards/invent-your-own standards - Microsoft.

 

[WaltC]

If you want to take that stance, Joe, FP32 is superior to FP24 in all respects.

Right, just as is true for fp24 vs. fp16. The problem in the NV4x pipeline is not fp32, but fp16, since fp16 is the reason nV4x cannot be said to always render to a higher precision than R4x0, but often renders to a lower precision than R4x0, instead, despite the fact that it supports fp32. Where nVidia gives fp32 on one hand, they take it away on the other with fp16. If nVidia wants to legitimize its claim to "fp32 precision throughout the pipeline," it's simple enough to do so--just drop fp16 support completely and run everything at fp32 precision.

It's kind of funny to hear comments like, "Sometimes it makes more sense to run fp16 instead of fp32 within nV3/4x," when you consider that fp16 doesn't make any sense at all in R3/4x0...

 

[WaltC]

nVidia has brought FP framebuffers and FP filtering anf HDR rendering to consumers by way of proven, established, standardized technologies that have a track record of working.

ATi is making it up as tey go along (especially with FP24), aided and abetted by the king of non-standards/invent-your-own standards - Microsoft.

Heh... Radar--you'd fit right in at nVidia PR where the operative principle always is:

World = wrong

nVidia = right

You should apply immediately.

 

[WaltC]

Obviously, the best precision is the one mandated by IEEE 754 (i.e. 7+5+4=16 bits) Geez, stop giving radar such a hard time...

Hehehe.... That was truly funny... GF "256," anyone? Thanks for the reminder...

 

[andypski]

Don't you think most people consider bilinear to be the bare basics for texture filtering? This isn't 1995/96 you know.

That depends - don't people think that a full selection of wrap and clamping modes and non power of two support is a bare minimum for textures these days as well?