FP blending in R420?

Hyp-X said:
pat777 said:
Then again, Half-Life 2 got HDR to work well on X800 XT.

They used FX16, not FP16.
But I don't know how they handled blending (if at all).

Isn't FX16(fixed point)the same as I16(integar)? Didn't OpenGLguy say R420 doesn't support I16?
 
pat777 said:
Hyp-X said:
pat777 said:
Then again, Half-Life 2 got HDR to work well on X800 XT.

They used FX16, not FP16.
But I don't know how they handled blending (if at all).

Isn't FX16(fixed point)the same as I16(integar)? Didn't OpenGLguy say R420 doesn't support I16?

I doubt they are doing anything that REQUIRES integer maths so they can just emulate it with FP.
 
pat777 said:
Isn't FX16(fixed point)the same as I16(integar)? Didn't OpenGLguy say R420 doesn't support I16?

I'm talking about D3DFMT_A16B16G16R16 - what do you mean by integer?
It stores values in the 0..1 range I wouldn't call that integer.

And R420 supports this format.
What OpenGLguy said is that it doesn't support blending with that format.
 
I think that's what everyone calls integer nowadays. The range that the values actually represent is pretty meaningless aside from a scale factor that may need to be done.

I guess FX is somewhat more correct because it means "fixed point", but I always thought of that as an NVidia term. Not only is NVidia's "internal FX12 units" the first time I saw it, but I don't see why I have to be reminded of the GeForceFX, IMO the most poorly executed graphics architecture from either of the big two for many years. :D

I think the best (though incomplete) solution is just do blending in the final R8G8B8A8 buffer. You won't get HDR effects from things like reflections off transparent windows, and you'll have to use texkill instead of alpha testing, but you'll at least get most of the effect with minimal effort. Tone mapping could be a serious problem, though.

The biggest shortcoming of this solution is that things like propulsion flames or futuristic weapon fire won't be able to get HDR effects applied. One possibility is to just use alpha textures with halos/glows/streaks already applied, as they often do in today's games. The other possibility is to just go Tron 2.0 style for all alpha blended effects, and do a fake HDR using ordinary R8G8B8A8 buffers.

Yes, it is disappointing that ATI skipped FP blending support, but there's still plenty of HDR effects available. It's possible that some of these hacks could increase performance on NV40 as well.
 
Mintmaster said:
Yes, it is disappointing that ATI skipped FP blending support, but there's still plenty of HDR effects available. It's possible that some of these hacks could increase performance on NV40 as well.
It sounds to me like you're saying, "If game developers do things the way they've been done for the last couple of years, performance could increase on the NV40." Huh? There's a reason why the move for FP filtering/blending, and it's so that games move forward, not backwards.
 
Joe DeFuria said:
Chalnoth said:
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 ?
 
Chalnoth said:
Mintmaster said:
Yes, it is disappointing that ATI skipped FP blending support, but there's still plenty of HDR effects available. It's possible that some of these hacks could increase performance on NV40 as well.
It sounds to me like you're saying, "If game developers do things the way they've been done for the last couple of years, performance could increase on the NV40." Huh? There's a reason why the move for FP filtering/blending, and it's so that games move forward, not backwards.
Backwards compared to what? Stop being so melodramatic.

-FUDie
 
Joe DeFuria said:
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.

Full FP32 support on the math side (silicon) of things was probably do-able, but we don't have mem bandwidth in surplus so it was better to come up with a good alternative (FP16) than have nothing at all (FP32). (Forget about the FP16 fiasco from the NV3x with the NV4x me says...)
 
DaveBaumann said:
LeStoffer said:
Performance of what? And what hardware are we talking about?

FP frame buffers. With NV40 fill-rate and bandwidth will be halved. There is no AA either. I suspect that these will be the long term limiting factors.


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

I'm assuming since the bandwidth hit for doing fp16 is as high as it is that fp24 would be even larger and possible unacceptable with today's memory.
 
Briareus said:
DaveBaumann said:
LeStoffer said:
Performance of what? And what hardware are we talking about?

FP frame buffers. With NV40 fill-rate and bandwidth will be halved. There is no AA either. I suspect that these will be the long term limiting factors.


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

I'm assuming since the bandwidth hit for doing fp16 is as high as it is that fp24 would be even larger and possible unacceptable with today's memory.
Because their chip was to big as it already was they couldn't put in FP blending of SM3.0 support without making a massive loss. Why could NV do it? maybe they had better engineers or maybe they are fine with a small profit margin I dunno.
 
Briareus said:
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.
 
Chalnoth said:
Mintmaster said:
Yes, it is disappointing that ATI skipped FP blending support, but there's still plenty of HDR effects available. It's possible that some of these hacks could increase performance on NV40 as well.
It sounds to me like you're saying, "If game developers do things the way they've been done for the last couple of years, performance could increase on the NV40." Huh? There's a reason why the move for FP filtering/blending, and it's so that games move forward, not backwards.
:rolleyes:
I have no idea why you just said what you did.

How many games support even close to R300's HDR capabilities in the "last couple of years?". Zero. That's what I'm saying. Plenty can still be done.

You think I'm arguing against the support of FP blending? Have you been paying attention to ANY of my posts in this thread? Don't try to make me look like a blind supporter of ATI, because I'm not.

I'm just saying that if developers choose to put a R3xx HDR path in their game (which would be smart if not downright necessary given the current market share), then NV40 could run it too if FP16-blending was too bandwidth intensive and caused a performance drop that some would deem intolerable. Geez, just look at my intentionally inconclusive statement. There are three qualifiers - "possible", "some", and "could". Obviously taking advantage of NV40's FP blending/filtering in an HDR implementation is the most straightforward, easiest, and most complete solution out there.
 
bloodbob said:
Briareus said:
DaveBaumann said:
LeStoffer said:
Performance of what? And what hardware are we talking about?

FP frame buffers. With NV40 fill-rate and bandwidth will be halved. There is no AA either. I suspect that these will be the long term limiting factors.


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

I'm assuming since the bandwidth hit for doing fp16 is as high as it is that fp24 would be even larger and possible unacceptable with today's memory.
Because their chip was to big as it already was they couldn't put in FP blending of SM3.0 support without making a massive loss. Why could NV do it? maybe they had better engineers or maybe they are fine with a small profit margin I dunno.

GeForce 6800U is bigger than the X800 XT. ATI didn't want to spend as much money on the X800 XT. Plain and simple.
 
pat777 said:
Chalnoth said:
Smaller dies also use less power, and thus can be clocked higher.

nVIDIA's double molex counters that unless your PSU is too small.

Just how many NV40's with the GPU clocked at 520Mhz have you seen (with stock cooling)?

I'd say none ;)
 
Back
Top