Reverend to andypski said:
....
You (and almost all of the ATI personnels that participate here) have conducted yourself in a very complimentary manner so far. This hasn't changed and I certainly don't want to appear like I naturally "distrust" postings by IHV personnels. It (my posts here) was just one of the reactions upon reading your "... and was chosen to sound plausible." comment (and of which you didn't expand upon).
I was surprised because that didn't sound like you! Anyway, thanks for clarifying it above. I appreciate your response, Rev.
About the "....was chosen to sound plausible" remark, I completely agreed with it, and thought it was a rather obvious observation, myself, based on the text of Kirk's interview. Although PR people of this "stripe" (and I dont necessarily mean that prejudicially, as I'll explain a couple of paragraphs down) may instantly respond to questions asked, it does not mean those responses will either appropriately address the questions asked, or that the responses will contain meaningful information or accurate information. What is often the case, in my experience, is that PR people build up, over time, a large intellectual portfolio of "automated responses" which are triggered by certain questions. Heh...
They have an entire psychological superstructure in place consisting of rationalizations, half-truths, evasions, insinuations, and even at times calumny, which, in a rehearsed and practiced fashion, they draw upon almost mechanically when "answering questions" posed by an interviewer in a professional, work-related context. Some PR people operate this way. The really best PR people are the ones who know how to tell the unvarnished truth and make it sound completely beneficial at the same time, and have enough basic knowledge about the subjects they address so that they can properly synthesize the information they receive from inside the company concerning those subjects.
Sometimes, too, when people in PR are not directly involved in the technological design/production work of the company, or else have limited/past experience in the field or a related one, they will often approach the real "movers and shakers" in a company (generally invisible to the public), for answers to questions they consider pertinent to their PR work. I mean to say that sometimes the desire of these people to learn and understand is genuine, and the last thing they expect is that they themselves will be manipulated by the people in the company who have the answers they need. But unfortunately, this happens. Thus, PR people themselves may dispense information publicly which they have every confidence in, because they trust the people in the company from whom they obtained this information, but may wind up dispensing falsehoods without even being aware that they are doing so at the time.
So, when the phrase..."was chosen to sound plausible" is used, I can instantly understand it. Much of PR is "chosen to sound plausible," but the key is in determining what actually
is plausible in whatever PR propaganda one might be exposed to. Only with experience in the related fields can one expect to be able to do this, though.
All PR is not necessarily propaganda. Good PR is recognizable by its lack of inflammatory spin, IMO. "Bad" PR, or "gutter PR," as I define it, seeks to capitalize on the ignorance of its intended targets, and to misrepresent evident truths, to spin them, such that only the technologically inexperienced might be swayed by the "arguments" presented. Bad PR seeks essentially to deceive. Good PR educates its market even as it persuades its market. Good PR *never* condescends, but bad PR almost always does. The best PR is based wholly on the truth. Of course, it goes without saying that this is my personal attitude on the topic, and I wouldn't pretend to speak for anyone else.
To one specific of Kirk's--the, in my opinion, meaningless drivel about "24-bit fp pipelines being based on an incompatible mathematical progression," or whatever he said, isn't a technological statement. Most obviously, of course, R3x0 proves the statement devoid of foundation. It might be valid in numerology or astrology somehow, but certainly isn't a technical statement of any kind as relates to fp pipeline precision. Technically, at best it's gibberish, and at worst it's calumny. To that end this kind of statement is, IMO, "gutter PR." As well, it does nothing except to completely undermine Kirk's designated title as "chief scientist" at nVidia, as far as I am concerned. It merely underscores the proposition that his title as such is a complete affectation designed to elicit attention from a PR standpoint. People tend to listen more carefully to a "chief scientist" than to a "Vice-President of Public Relations."
Kirk said:
I think that’s a very interesting question. If you go back to how DX9 was developed it wasn’t clear as we were doing our development and Microsoft was doing their development and ATI was doing their development what the target precision was going to be for DirectX. If you look at processors FP24 doesn’t exist anywhere else in the world except on ATI processors and I think it’s a temporary thing. Bytes happens in twos and fours and eights -- they happen in powers of two. They don’t happen in threes and it’s just kind of a funny place to be.
FP24 is too much precision for pure color calculations and its not enough precision for geometry, normal vectors or directions or any kind of real arithmetic work like reflections or shadows or anything like that.
I think what ended up happening was during the course of DX9 development and discussions between the various parties the targeted precision changed several times and we took a snapshot when the precision being discussed was 32 and ATI took a snapshot when the precision was 24. In fact DX9 was released without any guidelines as to precision and a clarification was made later and the clarification that was made was very timed to ATI in that it did not make a statement that 24 was not enough.
Complete nonsense, of course, in every respect. First of all, you don't invest $400 million in the development of an architecture based on mere "snapshots"--at least, no company with a brain in its corporate noggin would commit without a clear picture of exactly where it wanted to go and what it wanted to support. What a ridiculous concept. And of course, Kirk never explains the value of fx12 or fp16 in the nV3x architecture, which could be criticized in exactly the same fashion, nor does he bother to point out that using fp32 in nV3x results in extremely uncompetitive 3d performance. Strangely enough, the above comments of his pretend that the only precision in the nV3x pipeline is fp32. Which is kind of interesting, considering the question was about fp32 in the first place--gosh, based on his evaluation of fp24 as stated above I guess Kirk must believe that fp32 has
way too much color precision, then, and that, of course, fp16, which he doesn't predictably mention here to avoid the contradiction, has "far too little" precision for geometry...
Heh...
Sounds like a pretty good reason to me to skip both fp16 and fp32...
The following statement I thought was very telling:
Certainly one of the choices that Microsoft could have made is that it has to be 32 or nothing. They could have also made the choice that it has to be 16 or nothing.
I translate this as nothing less than: "Microsoft should have let us determine the DX9 specs instead of brazenly thinking they could actually manage the API themselves." Apparently, Kirk also believes that M$ itself is "numerologically challenged" with respect to the practicality of a 24-bit fp pipeline. The bottom line for Kirk is that everyone else is wrong, and only nVidia is "right." It's really sad to see the "chief scientist" of a company making superstitious statements on the value of undefined numerological conventions as applied to fp pipeline precision in a 3d chip (not that defining them would make his statements more valid, heh...
).
I personally think 24-bit is the wrong answer. I think that through a combination of 16 and 32, we can get better results and higher performance.
So, nVidia's "chief scientist" thinks that 24-bits of fp precision is the "wrong answer"; yet ironically, the R3x0 with fp24 outperforms nV3x in either fp16 or fp32. Maybe that's why ATi thought it was the "right" answer--not to mention M$, of course? I'm not sure Kirk has reasoned it out that far--at least it's not apparent from his remarks. Apparently, he is unable to digest this fact from a scientific viewpoint, although it is in evidence all over the Internet and is probably apparent in the R3x0 products nVidia has bought to study in its labs.
From a numerological/astrological position, it's interesting to note that when you combine 16 + 32 and average them, the result is 24, which just so happens to correspond to R3x0's level of fp precision. So is Kirk unknowingly contradicting his own numerological premise? Heh...
This was amusing:
The major issues that cause differing performance between our pipeline and theirs is we’re sensitive to different things in the architecture than they are so different aspects of programs that may be fast for us will be slow for them and vice versa. The Shader Day presentation that says they have two or three times the floating point processing that we have is just nonsense. Why would we do that?
I thought the "sensitive" comment was pretty amusing--and notable of course for the amount of specific detail the "chief scientist" supplies to support his proposition. Well, I guess admitting to designing chips based on numerology would be pretty embarrassing--so I guess that's why he doesn't feel the need to supply credible examples to back up his statements. As well, I thought it was pretty funny that he talks about "Shader Day" without realizing that "fp precision" might possibly be construed as a subject other than shaders--but that's only a "scientific" distinction, certainly. I really liked his "Why would we do that?" remark. Why would you do what, DK? Why would you design and ship a much much slower chip than ATi? Well, obviously, don't you think it might have something to do with that fact that
you had nothing do do with what ATi shipped? It's really strange to hear that nVidia apparently believes it has some sort of direct control over the products its competitors ship, because the only possible inference that can be drawn from this question is, "Why would we allow ATi to get that far ahead of us?" Heh...
As though it was up to nVidia in the first place. I have no idea why, DK. Presumably you do, though, so how's about you let us in on it?
Next, Kirk speaks scientifically:
Well one example is if you’re doing geometric calculations with reflections or transparencies and you need to do trigonometric functions. Our sine and cosine takes two cycles theirs takes eight cycles, or seven cycles I guess. Another example is if you’re doing dependant texture reads where you use the result of one texture lookup to lookup another one. There’s a much longer title time on the pipeline than there is in ours. So it just depends on the specific shader and I feel that for the calculations I mentioned are pretty important for effects and advanced material shaders and the types of materials that people use to make realistic movie effects. So they will get used as developers get more used to programmable GPUs and we’ll have less of a performance issue with those kinds of effects.
"So it depends on the specific shader..." Really, who'd a thunk it? Who'd a thunk that nV3x has a very hard time with ps2.0? Gosh, and I guess if your desire is to forego use of nV3x as a DX9 3d chip, and to pretend that R3x0 isn't designed for 3d, then things like trigonometric function cycle time might be of interest, I suppose--if you want to use the vpus as cpus, maybe, and *all* you need to do is run sine and cosine functions, etc. But really he answers himself here thusly:
...and I feel that for the calculations I mentioned are pretty important for effects and advanced material shaders and the types of materials that people use to make realistic movie effects. So they will get used as developers get more used to programmable GPUs and we’ll have less of a performance issue with those kinds of effects.
Emphasis mine. Ok, so, yes, "where nVidia is better than ATi" it's not of value to 3d-gaming support, but according to what the "chief scientist" says here, will impact "people who want to make realistic movie effects," but with the gotcha' that that isn't something which is immediately apparent from the superiority of NV3x's trigonometric function processing capability
right now, but is something that
will come in time and is dependent on the ability of "developers" to "get more used to" programmable gpus. I really wish he'd make some sort of scientific distinction between what kind of "developers" he's talking about--3d game developers, or developers who write software to "make realistic movie effects." He doesn't seem to appreciate a difference, it would seem.
Yeah, I would say that one of the issues is that since our hardware came out a little bit later some of the developers started to develop with ATI hardware, and that’s the first time that’s happened for a number of years. So if the game is written to run on the other hardware until they go into beta and start doing testing they may have never tried it on our hardware and it used to be the case that the reverse was true and in this case now it’s the other way around. I think that people are finding that although there are some differences there really isn’t a black and white, you know this is faster that is slower between the two pieces of hardware, for an equal amount of time invested in the tuning, I think you’ll see higher performance on our hardware.
For some reason, the "chief scientist" at nVidia doesn't understand API support in a 3d game, and API support in a 3d-card's drivers, and how that card's drivers are supposed to bridge the gap between the API support in the game code and the API feature support found in the 3d hardware. Reading this, one might think that APIs simply do not exist, and that what developers really do is to custom-program support paths which are different for everybody's hardware, and that developers can't really do that "until they go into beta." Presumably, everybody's still "in beta" at present--even the shipping DX9 titles, I guess. Yep, nothing like setting us back 8 years, is there, DK?
I really don't understand his comments. Is he saying that he thinks the development state of HL2 is in the pre-beta stage, and that when Valve "goes into beta" they'll discover....what, exactly? I mean, for him to sit there and say that "I think people are finding out..." something or other about there being no difference between nV3x and R3x0, is remarkable in its appalling ignorance of current events.
After Valve's recent presentation--where it wasn't a matter of "equal time," it was a matter of Valve spending 500% more time creating an nV3x-specific code path in the software than it spent in creating the DX9 code path for the game (suitable for all DX9-compliant hardware, including nV3x/R3x0), it's strangely negligent of Kirk to pretend that "people are finding out" anything more than doing a vendor-specific, mixed-mode code path for nV3x is a waste of time and money and that the nV3x architecture performs best under a generic DX8.x code path. Valve has a certain amount of stature here that I think it would be foolish of Kirk to ignore. But if Kirk and nVidia won't listen to M$, and developers like Valve, I suppose it is foregone they won't listen to anybody.
This post is far too long so I'll end it with some positive, heartfelt advice for DK:
Rid yourself and the company culture there with this obessesion you've got with frame-rate numbers in benchmarks. It has blinded you to everything your potential market has been telling you all year long: We Want Image Quality. Burn that into your psyches. Stop "optimizing shaders" at the expense of IQ. Rip out all your optimizations in your drivers relative to preventing the consumers of your products from getting things like full trilinear support, and full AF support. Give them the IQ features they want, and worry about frame rates *later.* You are now engaging in doing exactly the opposite of what your market wants. You will reap what you sow, in other words.
Face the truth that everybody knows that R3x0 is superior to nV3x. No amount of posturing, pomposity, bravado, bluffing, equivocating, and prevaricating will change that. Face that and move on to something you can control and you can improve in your own products right now: Image Quality. You can do a lot to improve that.
The way nVidia has conducted its affairs all year long has provided people with the firm conviction that buying nVidia 3d products means you get less IQ AND less performance. Since you are limited by the current architecture to taking second-place in the performance department--regardless of how you strip out IQ in order to attempt performance parity with R3x0--I advise you to reverse your present course with regard to frame-rate performance and to instead attempt to beat R3x0 in terms of image quality--if it's possible. Granted, you know nV3x much better than I do, but your comments indicate you appreciate little to nothing about R3x0 and that you are completely out of touch with your market. It could well be that you already know you cannot beat R3x0 in either image quality or performance, and that knowledge would explain much of your company's present conduct.
Nevertheless, your customers are demanding IQ--not
your definition of it--their definition. I would strongly suggest you start listening. No one wants nVidia out of the race--I know that I don't. But by the same token if you continue in the fashion you've become accustomed to all year long, few will eventually care if you withdraw from the 3d market. Consumers have an ingrained habit of polarizing around the companies which make a concerted effort to meet their demands.
ATi reshuffled itself from top to bottom and has clobbered you. If you guys continue to be slow on the uptake that it's no longer Business As Usual, and do nothing to address the needs of your markets other than to continue in your present path, I'm sure the clobbering will get a lot worse for you before it gets better--if indeed it ever does.
