Writing an update article to the Editor's Day coverage

[Althornin]

Your remark didn't contribute at all. You didn't say anything useful, you just criticized and didn't offer an alternative.

For the record:

A). I didn't ever assume NVIDIA was telling the truth.
B). I posted what I did, presented NVIDIA's point of view, and then asked a question to spark discussion.
C). YOU, personally, took it to mean that's what I believe.

If I implicitly believed what they told me, I wouldn't be posting it here and asking for commentary, would I?

for the record:
A question like:

Is ShaderMark flawed then?

following the BB quote sure sounds like you were believing them.
If you had asked - is this info accurate, or something else that didnt seem to assume then it wouldnt have seemed that way (duh).
You getting nothing out of my comment says alot about your attitude. Sure, my comment wasnt very helpful (in the information sense), but i dont think it was "abusive" in anyway. You getting nothing from being told you are approaching things wrong is YOUR problem - and its a major one. I thought that maybe you'd say "ok, so what is the right question", or "ok, <insert some comment here that shows your brain is working>". Sorry that your attitude made that impossible.

 

[bloodbob]

Well I don't like how he has used 3dmark2k3 in the update just before a new version is due out which forces the reviewer to only use futuremark approved divers :/

 

[Reverend]

Jakub,

Here's another absolutely useless post for you to read.

You're too sensitive. And probably too serious for your own good if you intend to participate in public forums. More knowledgeable folks (than me, not you, of course) participate in public forums and are willing to ignore posts that they deem are "useless".

Lighten up. You said that you have extensive Internet experience... which means you should know better (than to respond to posts that apparently do not answer whatever questions you post in a public forum). Don't respond to posts that "annoys" you if you want to participate in public forums and if you're not willing to acknowledge what the Internet is all about. Use emails instead.

Beyond3D, as a site and as a forum, will not be much different than any other public forums regardless of what its staff wants.

PS. Don't ask PR personnels if you want straight answers. Just a hint.

 

[LeGreg]

no filtering (infact, only GL_NEAREST or GL_NEAREST_MIPMAP_NEAREST)
no alpha test
no alpha blend
no logic op
no dither

ATI doesn't support blending modes with FP targets as well
(or emulated through a pixel shader but same for NV)
it doesn't support filtering (other than Nearest) and doesn't support multisampling with FP rendertarget (same as NV).

The major differences so are:
support of all addressing modes (wrap, mirror), support of mipmapping.

That alone make them acceptable to pass WHQL and expose the cap whether nvidia cannot.

So nothing mysterious here

LeGreg

 

[LeGreg]

For those who are interested in details
here is Doug Rogers (nv devrel) and Richard Huddy (ATI devrel) answer on dxdev list:

Date: Mon, 13 Oct 2003 15:46:11 -0700
From: Doug Rogers <DRogers@NVIDIA.COM>
Subject: Re: Is FPRTF or MRTs the requirement of DX9 compliance?

MRTs and floating point textures are not required for DirectX. The GeForceFX
series do not have MRTs, so are not supported.

The floating point textures in the GeForceFX do not support wrap mode, nor
MIP maps. Wrapping and MIP maps are required attributes of floating point
textures for DirectX and so the the caps bit is not exported for GeForceFX.


The new drivers will not expose floating point textures.

We are working with Microsoft on a mechanism to expose our floating point
texture format implementation.


-Doug Rogers
NVIDIA Developer Relations

Date: Tue, 14 Oct 2003 07:22:57 -0400
From: Richard Huddy <rhuddy@ATI.COM>
Subject: Re: Is FPRTF or MRTs the requirement of DX9 compliance?

Guys,

Just to add a little detail to Doug's posting - which I believe is wholly
accurate.

>MRTs and floating point textures are not required for DirectX. The
>GeForceFX series do not have MRTs, so are not supported.

Support for MRTs and FP textures are not a requirement for a driver to offer
the DX9 interfaces. Indeed a driver could expose standard DX8 hardware
through the DX9 interfaces (and many do exactly that since it's easier to
maintain just the DX9 driver code base).

>The floating point textures in the GeForceFX do not support wrap mode, nor
>MIP maps. Wrapping and MIP maps are required attributes of floating point
>textures for DirectX and so the the caps bit is not exported for GeForceFX.

The Microsoft WHQL tests specify minimal levels of functionality - for
example in order to expose the caps stating that hardware supports floating
point textures the tests require that both wrapping and mip-mapping work in
all cases that the test team have considered (and, yes, they try to be
thorough).

i.e. The WHQL tests are stating that you, the programmer, must be able to
rely a minimum standard of support from hardware before it offers this
facility. This avoids the alternative of having many more caps bits...

In particular the WHQL tests try hard to enforce the philosophy that
capabilities should generally be independent (or 'orthogonal') wrt to one
another.

>The new drivers will not expose floating point textures.
>
>We are working with Microsoft on a mechanism to expose our floating point
>texture format implementation.

Although the WHQL tests do not presently allow a GeForce FX to expose the
floating point texture capabilities it makes sense for Microsoft to want to
expose whatever lesser capabilities the hardware may have.

As I understand it this would require a change to the spec, or to the WHQL
tests.

Thanks,


Richard "7 of 5" Huddy
European Developer Relations Manager, ATI

LeGreg

 

[PatrickL]

Already posted in that thread: http://www.beyond3d.com/forum/viewtopic.php?t=8546&highlight=

 

[digitalwanderer]

Beyond3D, as a site and as a forum, will not be much different than any other public forums regardless of what its staff wants.

Sorry to disagree with you here Rev, but I gotta call "BULLSHIT!" on this one.

B3D is quite a unique site, and I ain't ever been to one quite like it. (And I do mean that in a good way. )

This board has the lowest signal-to-noise ratio of any board I've ever seen. 8)

 

[Aivansama]

This board has the lowest signal-to-noise ratio of any board I've ever seen. 8)

/OT
I would have thought that this board actually has a very high SNR... better than -20dB, which would be remarkably good for a web forum....

OT/

 

[Xmas]

But it's mostly because of the signal being high, not the noise being so low

 

[digitalwanderer]

This board has the lowest signal-to-noise ratio of any board I've ever seen. 8)

/OT
I would have thought that this board actually has a very high SNR... better than -20dB, which would be remarkably good for a web forum....

OT/

Doh!

"Strike that, reverse it."

 

[Reverend]

Beyond3D, as a site and as a forum, will not be much different than any other public forums regardless of what its staff wants.

Sorry to disagree with you here Rev, but I gotta call "BULLSHIT!" on this one.

B3D is quite a unique site, and I ain't ever been to one quite like it. (And I do mean that in a good way. )

This board has the lowest signal-to-noise ratio of any board I've ever seen. 8)

What I meant was that B3D will not be able to stop anyone from posting stuff that aren't perceived by certain folks to be "on topic" or "constructive" but/and only consists of "criticisms". Hell, I do this from time to time myself (i.e. nothing but criticisms in my posts), but only when I am unusually annoyed about certain posts or whatever.

What I was simply suggesting is that, since Jakub had said that (paraphrased) he's been too long on the Internet to fall into the trap of listening/reading posts that doesn't help him (according to him, of course) and presumably falling into the trap of tit-for-tats, that I am surprised by some of the things he's said in this thread... since that is exactly what he has done on the contrary i.e. he has fallen into the "trap" he'd said he only knows too well.

For someone that has admitted he ain't no "tech guru", he's been rather bold with his words, like he's lecturing some of the guys here on the etiquette of Internet forum participation when he'd admitted he knows better.

<shrug> Don't mean to step on Jakub's toes... but perhaps I had presumed a little too much about him based on his first post in the other related thread, which was :

I'd also love to stick around (really, I mean that) and reply in a back-and-forth discussion, but I've been on the internet far too long to make that mistake. While this forum is cleaner than almost any I've seen, I can't trust myself to not give into my resentment at some of the allegations, or to avoid the trolls that surely lurk.

Perhaps what has happened is simply what Jakub said exactly, which is that he can't trust himself to do the right thing, hehe

 

[Aivansama]

But it's mostly because of the signal being high, not the noise being so low

/OT'ish
With high signal power levels distortion becomes an issue... And thus we leave the OT region and return back on topic. Sort of.

OT'ish/

 

[Arun]

I'd make a comment about GFFX 5900s automatically creating noise just by scrolling the page with default IE settings ( seriously ) but that'd be kinda lame

Rev: I'd personally suggest instead to ask PR personnel.
But consider whatever they say as pure damage control and subjectivity. IMO, if you know what you're doing, seeing what they seem "afraid" about can help a lot to understand what's wrong.

Example:
In January 2003, I asked Brian Burke about which format the GFFX benchmarks used, and what the associated performance hits would be.
His response was that they were done in 32-bit mode, since "that was the standard for quite some time". He also said "FP16 is twice as fast as FP32."

Let me insist first that he seriously beleived FP16 = 2xFP32 performance. Got the proof of that much later, in September 2003. But notice how he didn't give a performance hit between 32-bit integer ( or rather, FX12, thus 48-bit, but I doubt he knew what that meant ) and 64-bit floating point.
As we would realize later, 32-bit/FX12 performance = +- 3x FP16 performance.

PR personnel is easy to trick to get the info you want to get. You just have to know what you're doing


Uttar

 

[digitalwanderer]

In January 2003, I asked Brian Burke about which format the GFFX benchmarks used, and what the associated performance hits would be.
His response was that they were done in 32-bit mode, since "that was the standard for quite some time". He also said "FP16 is twice as fast as FP32."

Let me insist first that he seriously beleived FP16 = 2xFP32 performance. Got the proof of that much later, in September 2003. But notice how he didn't give a performance hit between 32-bit integer ( or rather, FX12, thus 48-bit, but I doubt he knew what that meant ) and 64-bit floating point.
As we would realize later, 32-bit/FX12 performance = +- 3x FP16 performance.

PR personnel is easy to trick to get the info you want to get. You just have to know what you're doing

I can't hear about that BB quote/story enough, it just keeps getting funnier to me each time. It's just amusing/sad on sooo many levels...

BTW-Rev-Yeah, I guess I missed your point. I just know Jakub from a few other forums and he's not as bad as his posts here indicate, I think he just got off on the wrong foot. (You'd like him better with his nVnews avatar. )

 

[Jakub]

Reverend,

A). You're right, I shouldn't have commented on the useless post. My opinion of it hasn't changed but I simply contributed to what I had tried to prevent.

B). I imagine you were joking with the PR straight answer comment. Yes, I'm fully aware of it, however, it helps to know the official line before you can dispute it.

 

[Jakub]

BTW-Rev-Yeah, I guess I missed your point. I just know Jakub from a few other forums and he's not as bad as his posts here indicate, I think he just got off on the wrong foot. (You'd like him better with his nVnews avatar. )

I'm not on nVnews...

 

[Oblivious]

BTW-Rev-Yeah, I guess I missed your point. I just know Jakub from a few other forums and he's not as bad as his posts here indicate, I think he just got off on the wrong foot. (You'd like him better with his nVnews avatar. )

I'm not on nVnews...

I think you mean jAkUp not Jakub, DW

 

[dominikbehr]

RGBE format is all nice until you try to express a color like RGB(0.0001, 10000.0, 0.0)
this is why floating point render target support makes it easier to implement hdr rendering and gives more accurate results