GF FX: best graphics processor?
- Thread starter BoardBonobo
- Start date
Well, the award is for "...the most promising new microprocessor technology unveiled in 2002...", which I believe strongly that the NV30 is eligible for.
Unfortunately it is also a firm favourite for the "Most Disappointing Microprocesser Produced Belatedly in 2003 Award".
MuFu.
Unfortunately it is also a firm favourite for the "Most Disappointing Microprocesser Produced Belatedly in 2003 Award".
MuFu.
OpenGL guy
Veteran
When was the GeForce FX unvieled in 2002?MuFu said:
BoardBonobo said:
I would suggest that this is one of them bought and paid for awards. Considering that no end users own one and all that it is pretty hard too see just how they came to this conclusion. Methinks they will have to renig the award when the R400 hits the market. BAH nvidia is really pushing their garbage on the PR side of things more then ever it seems. I can hear them now "lets put some lipstick on this bitch".
OpenGL guy
Veteran
Well, normally you unveil a product, not a piece of paper.tEd said:OpenGL guy said:
paper launched in november 2002
Both the NV30 and Athlon 64 were demonstrated in 2002.
The Athlon 64 demonstration exhibited new heights of performance for its clockspeed. There was nothing that compared as favorably with the expected launch performance of the architecture, so I can sort of think that particular vaporware choice had merit.
The NV30 "sets new records for performance, complexity, and cost" according to them. Looking at the Athlon 64 win, I don't get the reasoning at all about performance when the 9700 was also being considered, and given the featureset, I don't see the sense unless "128-bit" versus "96-bit" sounded like a clear win to them, or they are easily impressed by "the longest PR documentation for featureset". As for cost, unless setting a new record for highest/i] cost is a good thing, that consideration seems pretty mutually exclusive with the other concepts. As for complexity, they may have simply been impressed by the transistor count.
I'd hope they were a bit more competent than I believe from the above...but that might just be a silly expectation on my part.
About them being bought: I can't conceive of AMD being able to out-buy Intel, or having the marketing awareness to do so. Could easily be a "political" or vested interest thing, though.
The Athlon 64 demonstration exhibited new heights of performance for its clockspeed. There was nothing that compared as favorably with the expected launch performance of the architecture, so I can sort of think that particular vaporware choice had merit.
The NV30 "sets new records for performance, complexity, and cost" according to them. Looking at the Athlon 64 win, I don't get the reasoning at all about performance when the 9700 was also being considered, and given the featureset, I don't see the sense unless "128-bit" versus "96-bit" sounded like a clear win to them, or they are easily impressed by "the longest PR documentation for featureset". As for cost, unless setting a new record for highest/i] cost is a good thing, that consideration seems pretty mutually exclusive with the other concepts. As for complexity, they may have simply been impressed by the transistor count.
I'd hope they were a bit more competent than I believe from the above...but that might just be a silly expectation on my part.
About them being bought: I can't conceive of AMD being able to out-buy Intel, or having the marketing awareness to do so. Could easily be a "political" or vested interest thing, though.
NV30 was at least officially "launched" in 2002, which A64 was not. Yes, this leaves the definition of "launch" entirely up to the IHV, but nonetheless, paper launch in 2002, limited reviews in January, availability (of sorts) in February-maybe-March is at least justifiable under a suitably generous interpretation of "2002". Launch and reviews in September 2003 with availability presumably September or October is not, IMO.
Feh. Performance "for its clockspeed" isn't really a relevant measure, and isn't going to impress MDR-types in particular. Hammer certainly will set performance/clock records for x86, but by the same token will still be trailing most server-oriented CPUs in this category.
This is more to the point. If the "2002 awards" actually referred to products released in 2002, the winner should have been Northwood (.18um P4), released last January...but that wouldn't have been very exciting a choice, I guess. Sort of like how all the big Oscar-contender movies get released Dec. 28 in 2 theaters in New York and LA, and a good movie that happened to have come out at the start of the year gets forgotten. Except that movies do base their release schedules around the Oscars, while chip makers (hopefully) rarely base theirs around the In-Stat/MDR Analysts' Choice Awards.
As for design innovation, Hammer is just a scaled-down version of what the Alpha team did with the EV7, which is available now (finally), and which got snubbed in the server category. (Although Itanium 2 is a good choice also. (Although while we're on the subject the next Itanium core, Madison, will be released before Athlon64 is! (Although it's just a die-shrink and cache-tack-on, so again not too exciting.)))
Anyways, the discussion of the nominees is revealing. First, because it appears the cutoff for eligability is that the microprocessor in question be sampling in 2002 (which A64 and NV30 both certainly qualify for, although the ARM core which won the IP category was only just available as VHDL in 2002, not in any sampling products). Second, the description of the 9700 as "still the fastest in some crucial respects" is odd given the publication date of Dec. 9.
OTOH, who can argue with this description of GFfx:
MDR said:
Oh...you did!
"Complexity and cost" I think mean exactly what you think they do; yes, both are negatives although also somewhat "impressive" in a certain sort of way. As for performance, well, the Ultra does clearly outperform the 9700 Pro in most games in standard (i.e. no AA/AF) settings, as well as on 3DMark01 and 03, and usually holds its advantage with AA/AF assuming that you compare Nvidia's "balanced" to ATI's "quality" AF. And while the GFfx will certainly never use it to its advantage in game playing, PS/VS 2.0+ is indeed a significant step up from PS/VS 2.0 (as may be FP32 in certain situations). So from the point of view of the "biggest, baddest" design, I think GFfx has a claim over 9700.
From the point of view of well-targeted, balanced design--which is what all good engineering is about, really--there's no question the 9700 is the more successful part. But Titanic probably wasn't the best movie of the year either. There's something about the way awards and spectacle go hand in hand.
Of course, the votes were likely taken before the disappointing benchmarks, the apparent Ultra cancellation, the embarrassing 3DMark whining, the still-ongoing mysteries over its pixel shader performance and its seeming inability to output >4 pixels/clock, etc. Not that many analysts necessarily pay any attention to that sort of thing...
Dave H said:
No, they were both demoed physically in 2002, and both had specifications released in 2002. nVidia called this a launch, AMD did not (well, not to my recollection atleast). In any case, I personally consider them equivalent. You don't? Perhaps it is because I tend towards completely discounting PR (AMD can't be compared with nvidia in this regard, as, IMO, comparatively they do not have any).
I suspect AMD could ship 4000 Athlon 64 chips (the motherboard chipsets are ready). In fact, their emulation of part of the nvidia strategy (releasing lower clocked parts targetted at markets where they can be competitive) appears to be intended to happen soon after it does for nvidia.
If you think the Quadro FX shouldn't be considered part of the GF FX launch (and therefore Opteron is not related to Athlon 64, assuming it is available in a timely manner), I can certainly see your point (though the "3" at the end of the year in the current date with GF FX unavailable still makes me think both "launches" in 2002 are absolutely equivalent
).Well, at the time of the GF FX PR noise, was it as clear that the Athlon 64 would be delayed as much it will as it is now?
Feh?
I'll thank you to keep your phlegm to yourself! Unless you say you are targetting higher clock speeds and there is reason to believe you can achieve that.
Well, the award was most important, not just highest performance. Speculation fits in, but I'd hope speculation based on other than PR documents. That is why GF FX over 9700 still puzzles me, but maybe I expect too much from "analysts".
How is that "most important"? The "promise" it offers really does seem to pale in comparison to the x86-64 architecture. Again, this is not the case between GF FX and 9700 unless you are basing your evaluation on PR.
I really hope not as well.
Though I do think "collecting awards" is just as much a priority for them for OEM recognition, I also think there are plenty of them to go around.Woah, that was deep.
I guess it all boils down to politics and vested interest, after all. Would we be having this discussion on this board if processors not in the public eye got awarded? I wouldn't...the EV7 is only a barely recollected topic of discussion on Ace's Hardware for me. I guess we've answered the question indirectly, haven't we?
I suspect such analysts are targetted for first hand demonstrations. I'd expect the GF FX comparative performance was established as real in their minds, and I've come to think this is the successful side of their PR campaign. It might have worked more on OEMs without the publicity about the fan, which I think is the real reason the Ultra has been de-emphasized.
OTOH, who can argue with this description of GFfx:
MDR said:
Oh...you did!
aka "easily impressed by "the longest PR documentation for featureset".
As far as performance, that is still contradictory IMO, as absolute performance is not the only metric. But if they are giving awards for highest cost (which still strikes me as insane, and I'm sure nvidia would prefer was not in the list without a "low" in front of it), I guess consistency doesn't matter.
Yes, it all boils down to that last statement, I think...
...or to this alternate way of describing the same situation.
THe_KELRaTH
Regular
The effect of any fool being able to create a website!
No, I do. However, Microprocessor Report is a very industry-standard sort of journal, and thus they tend to take official statements seriously as a matter of professional respect. (After all, only a very small fraction of microprocessors are marketed to general consumers, so this is not always a bad assumption. Although still probably too often.) Although anyways it seems the actual critereon was if the chip sampled in 2002, which is an objective measure, if a slightly odd one.
Yeah, it gets down to splitting hairs eventually. Do note, though, that Opteron was up for an award in the server category. To me January/February and September just aren't "absolutely equivalent" when it comes to a so-called 2002 award, but then again if it were up to me I'd stick to "shipping in 2002" like any normal person.
Now that you mention it, I don't think it was known, particularly when the nominations were made (and probably when the votes were cast).
On a side note, this might be a good time to bring up Freelancer, 1999's E3 Best of Show. (Release date: spring 2003).
Why not just talk about performance? It is very easy to trade off performance/clock for clock rate and vise versa; it's increasing overall performance that's the hard part. Knowing this, real EEs don't put stock in the notion of performance/clock as an end unto itself.
Having said this, Hammer (Opteron in particular) looks to have truly stellar performance, and would certainly be a strong candidate for such an award. (If it were the right year, of course!
)Northwood is IMO a phenomenally successful chip (design wise I mean). It got very impressive gains at equivalent clocks over Willamette, due to extra cache and some speed tweaks. The clock rate has ramped beautifully, plus it's been able to accomodate a big FSB bump (with a bigger one coming). With HT enabled in the new versions it marks the first implementation of simultaneous multithreading, which is a very elegant technique that will be extremely important in the future and which marks a very quick move from academia to industry.
Generally, it's the first halfway mature implementation of the "Netburst" architecture, which itself is extremely elegant and forward looking (including other techniques plucked off of research papers, like the trace cache) but was rather ungainly in Willamette. And, basically, because over the course of 2002 Northwood moved the Netburst architecture from maybe 30% to 75% marketshare of all general purpose CPUs. And most of all because the only other new PC processor core introduced in 2002 was Thoroughbred (except for CPUs from VIA and Transmeta, I mean.
), and Northwood was a much more successful design. Of course Northwood's performance is less than that of Athlon64's--it will have been released 20 months earlier! As for importance and performance for its time, I think it will have won out.
Shades of LISP...
Well, except that Microprocessor Report's readership is a bit different from Ace's. (Not that I don't enjoy Ace's. And, well, not that I receive Microprocessor Report.
) Anyways, neither did the ARM ARM1136JF-S or the Cavium Nitrox Plus CN1340p get much press, AFAICT.But yes, I think the fact that EV7's design has been known for so long (now that's a chip that's really late, although for the simple reason that its corporate owners have been trying to smother the Alpha for years now) means it got short shrift from a publication like uPR which often attempts to be buzz-compliant.
Well mainstream OEMs were never going to give the Ultra a design win to begin with. (After they heard the fan, I mean.) And the enthusiast-oriented OEMs like Alienware and Falcon would probably still rather use Ultras.
aka "easily impressed by "the longest PR documentation for featureset"
Well I don't think the "+" on the end of PS/VS 2.0+ represents frivolous length. It's arguably as significant a feature as PS 1.4 over PS 1.1-1.3. If the GFfx still has worse shader performance than R300...well, same with the 8500 compared to GF4! Indeed, while 2.0+ is probably less relevant for gaming within the GFfx's life span than 1.4 is within 8500's, it should offer very significant benefits for DCC. Same with FP32 (which incidentally can be written using exactly the same number of characters as "FP24").
As for PR featureset documentation such as
[url=http://nvidia.com/view.asp?PAGE=geforcefx_features said:
...
Concerning the Athlon64, I hear they are having trouble getting it up to speed at the moment.
OpenGL guy said:
The "microprocessor technology" was unveiled in 2002.
It's a "Analyst's Choice" award, what did you expect? I can totally understand why they have chosen NV30. No big deal, and certainly not worth a thread full of rolling eyes and dismissive jibes. IMHO, it's just another reminder of how disappointing the product is.
MuFu.
Hellbinder
Banned
No its another reminder of how powerful Nvidias PR department is. No one that really matters seems to be psting anything about the GFFX being anything other than the Greatest thing since sliced bread.
Beat me too it.
Now after reading that I'm going to eat some humble pie - I thought they had blindly award NV30 based on its "promise" (as per the description of the award), but to do so in light of the recent fiasco regarding performance, price, heat, noise etc is absurd.
*munch, munch, munch...*
I'm going to join in the eye rolling too, heh.
MuFu.
Now after reading that I'm going to eat some humble pie - I thought they had blindly award NV30 based on its "promise" (as per the description of the award), but to do so in light of the recent fiasco regarding performance, price, heat, noise etc is absurd.
*munch, munch, munch...*
I'm going to join in the eye rolling too, heh.
MuFu.
