DX 9.1 True?
- Thread starter Unit01
- Start date
I will!geo said:
Dave's high and just falling for the hype, it ain't gonna happen. 8)
Hyp-X said:
Agreed. I first recall hearing about "9.1" on some website months ago (I can't recall where), and it wasn't presented as a straight-up rumor, as I remember, but it was presented as a kind of apology for nV3x and DX9 and its context was, paraphrased, "M$ is going to help nV3x with DX9 by way of DX9.1".
Flash to the present, and not only has the original projected ship date for "9.1" come and gone, but people like Anand Shrimpi are expanding the 9.1 topic to include M$ switching to fp32 as the API standard instead of the DX9 fp24 standard M$ has already set for DX9. Lacking a credible source for this information, Anand candidly admits his 9.1 info is based only on what he has "heard," but he is either unwilling or unable to reveal from whom he has "heard" it. (An anonymous email from www.nvidia.com , perhaps?)
The fp32 thing makes zero sense to me, seeing that R3x0 doesn't do fp32 as nV3x does it, and the way nV3x does it so unsuitable for 3d gaming that nVidia uses fp16, instead. So if 9.1 moves the API standard to fp32, does this mean that fp24 becomes the partial, and fp16 support simply drops off altogether? Of course, that makes no sense, either, seeing how nVidia doesn't do fp24, which is why I have a hard time believing it...
Having said that, if nV4x or R4xx both support fp32 in a manner suitable for 3d-gaming such that partial precision is no longer required, it might make sense--except for the millions of R3x0 cards out there in current use (don't forget the two hundred or so nV3x cards in current circulation)...
Seriously, though, when I start hearing about "9.1" from credible, identifiable sources, in a context other than an nV3x/DX9 apology, and in a context which lays out what 9.1 entails specifically, then, and only then, will I emulate Anand and start listening.
You should always listen, how else are you going to laugh at it? :|WaltC said:
There's no way to move to FP32 as the standard for PS2.0/PS3.0 as it would break backward compatibility. Microsoft didn't increase the precision requirement of PS1.1 or PS1.4 in DX9 did they?
In theory if all the PS3.0 will have FP32 as the default (including XGI), the IHV's and Microsoft could agree to increase the minimum requirement - but even in this case it would be pointless.
I'm pretty sure PS4.0 will have at least FP32 but that won't happen until DX10.
In other words it might be true that R420 will be FP32 only, but that wouldn't require DX9.1
In theory if all the PS3.0 will have FP32 as the default (including XGI), the IHV's and Microsoft could agree to increase the minimum requirement - but even in this case it would be pointless.
I'm pretty sure PS4.0 will have at least FP32 but that won't happen until DX10.
In other words it might be true that R420 will be FP32 only, but that wouldn't require DX9.1
Put it like this. There is some pushing and pulling going on, and no-doubt some serious lobbying to Microsoft. It'll become clearer why in the fullness of time.
Lemme guess: NV40 = 16FP/32FP so nVidia is pushing like a bad dog to get it modified and M$ is damn near about to give in just 'cause nVidia fucked up so badly.DaveBaumann said:
How about this for an idea:
Nvidia did not fix the register usage problems in NV40, and to unlock PS/VS 3.0, microsoft will release a new hlsl compiler (aka 9.1 sort of) which will give nvidia a more optimal shader on compile. Or perhaps, By default _pp will be enabled and only statements with a different flag will be run at full precision.
Nvidia did not fix the register usage problems in NV40, and to unlock PS/VS 3.0, microsoft will release a new hlsl compiler (aka 9.1 sort of) which will give nvidia a more optimal shader on compile. Or perhaps, By default _pp will be enabled and only statements with a different flag will be run at full precision.
Deathlike2
Regular
That's a terrible idea. Besides "breaking" something that was working (hard to explain), it's "moving the industry backwards". I doubt that MS would support FP16 in that context especially when they made FP24 STANDARD for DX9 full precision.
I didn't hear about it... but even so, it would be closer to the truth (or lack thereof) if this performance was gained by hand optimized shaders. (I very much doubt a 60% performance gain in their drivers/an updated patch/an updated DX9).
Performance comes by incremental gain... not a "super boost". Only newer hardware could make that happen anyways...
DaveBaumann said:
Maybe we'll start www.waveywatch.com to chart microscopic changes in your positions and speculate what information we aren't seeing (yet) is behind it.
Deathlike, the winky indicated I was mostly kidding about mixing up the two rumors. I also don't expect a mere DX9 update to offer anywhere near a 60% speed increase, even with the current disparity in architectures. A recent post on the Gearbox forums said we should expect big speed gains from a Halo patch, not a DX9 update. I don't follow Halo much, so I don't know how credible the source is. Sorry, I can't even find the news item that alerted me to the forum, but this B3D thread mentions the Gearbox forum post I'm referring to.
In other words--DX9.1 is real. After all, it's the Inq.jimbob0i0 said:
Similar threads
