I am sure Dave B will disagree with me on this one...

[Geo]

Well, just to be clear --I'm still on 16-pipes like grim death. The only thing that would move me (other than actual release from ATI of course!) is credible reports of >256-bit bus. I just think I understand the other sides reasoning, while not agreeing with it.

 

[ERK]

"Remember that the same people/scene were talking about EXTREME PIPELINES the same way at the launch of the R420". :?

Wouldn't it be ironic if this time it turned out the 'extreme pipelines' was correct?

Meaning the very high performance 16 "pipe" route that Dave is reminding us could happen.

 

[Headstone]

Maybe this is where the original rumor came from

 

[digitalwanderer]

"Remember that the same people/scene were talking about EXTREME PIPELINES the same way at the launch of the R420". :?

Wouldn't it be ironic if this time it turned out the 'extreme pipelines' was correct?

Meaning the very high performance 16 "pipe" route that Dave is reminding us could happen.

Yup.

Maybe this is where the original rumor came from

I'm almost scared to conceive of that possibility. :?

 

[Kombatant]

Having more and more pipelines is no panacea; if you don't work towards code being fed efficiently to your hardware (either on software level or, even better, on hardware level), all these pipelines won't actually contribute what they can in the card's performance.

 

[WaltC]

Maybe they're simply doing "a NV30"? i.e. they'll look at G70 numbers and maybe decide to do a last minute speed bump? Possible? Crazy?

Well, the 9700P was a shipping, retail product for roughly nine months before nV began "shipping" nV30 prior to cancelling it and going back to the drawing board.

Why should ATi be paying attention to G70 numbers reported for a G70 that itsn't shipping yet?... I mean in the gap between August '02 and spring-summer '03 when the 9700P reigned supreme I recall lots of "numbers" coming out of various nV-funded and manipulated web sites as to "privately tested, " pre-shipping nV30 performance versus the 9700P that proved not to be worth a plugged nickel when indeed nV actually did ship nV30. It was only when nV saw that no amount of whacky, zany, strange, RDF PR could save their bacon that it publicly cancelled nV30 production.

 

[MuFu]

I still can't believe people are fixated on "pixel pipes" after all the hints dropped by Dave...

Yeah I thought Dave had more pull than that A lot of people haven't taken notice but he will probably have his "I told you so" moment pretty soon.

I been taking that, WRT R520, as "don't get carried away with the *number* of pixel of pipelines" (i.e. don't assume quantity = performance). I still think they'll be pretty conventional, i.e. resemble R3x0.

RV530 and R580 sound a little different, however.

 

[trinibwoy]

I still can't believe people are fixated on "pixel pipes" after all the hints dropped by Dave...

Yeah I thought Dave had more pull than that A lot of people haven't taken notice but he will probably have his "I told you so" moment pretty soon.

I been taking that, WRT R520, as "don't get carried away with the *number* of pixel of pipelines" (i.e. don't assume quantity = performance). I still think they'll be pretty conventional, i.e. resemble R3x0.

Well if the number isn't as important and they are conventional pipelines then they can't resemble R3x0 too much. They would have to be vastly more efficient.

 

[Geo]

I still can't believe people are fixated on "pixel pipes" after all the hints dropped by Dave...

Yeah I thought Dave had more pull than that A lot of people haven't taken notice but he will probably have his "I told you so" moment pretty soon.

I been taking that, WRT R520, as "don't get carried away with the *number* of pixel of pipelines" (i.e. don't assume quantity = performance). I still think they'll be pretty conventional, i.e. resemble R3x0.

Well if the number isn't as important and they are conventional pipelines then they can't resemble R3x0 too much. They would have to be vastly more efficient.

Yeah, well, this is where clock speed comes in. MuFu has been publicy leaning 650-700mhz.

Assuming, as we've heard, that G70 is 430mhz, then G70 > NV40 by more than 50%. Assuming that R520 closely resembles R480, then 50% would require clockspeed of something like 810mhz. So MuFu is relying on *some* efficiency improvements to close that gap. I'm relying on more effeciency and less clock. We're two Jesuits arguing over just how many angels are on the head of that efficiency pin.

 

[MuFu]

Well, they are going to have to revamp the ALU and reg structure anyway so you'd expect a greater pixel crunching throughput per clock for starters; not really an "efficiency improvement", as such (same could be said for adding VS units; it's just "more"). There are plenty of other areas that efficiency improvements could come from - internal bus topology (strongly hinted at), memory interface, dispatch/scheduling, drivers etc. Talking very generally of course, but I can see a lot of minor enhancements in conjunction with a big clockspeed bump providing the neccessary performance. Massive efficiency increases are Xenos territory. 8)

R4x0 was one thread of R3x0 enhancement, R5x0 is the other. Given twice the development time and the requirement to meet FP32/SM3.0, obviously the latter is going to be the more dissimilar, but I still wouldn't expect anything too drastic, given the common ancestry.

 

[Geo]

Well, I was lumping anything that increases IPC as "efficiency" (at preserving ticks anyway! ).

 

[ondaedg]

Maybe they're simply doing "a NV30"? i.e. they'll look at G70 numbers and maybe decide to do a last minute speed bump? Possible? Crazy?

Well, the 9700P was a shipping, retail product for roughly nine months before nV began "shipping" nV30 prior to cancelling it and going back to the drawing board.

Why should ATi be paying attention to G70 numbers reported for a G70 that itsn't shipping yet?... I mean in the gap between August '02 and spring-summer '03 when the 9700P reigned supreme I recall lots of "numbers" coming out of various nV-funded and manipulated web sites as to "privately tested, " pre-shipping nV30 performance versus the 9700P that proved not to be worth a plugged nickel when indeed nV actually did ship nV30. It was only when nV saw that no amount of whacky, zany, strange, RDF PR could save their bacon that it publicly cancelled nV30 production.

Which Nvidia "funded" and "manipulated" web sites are you referring to? Proof or linkies please.

Someone should buy Walt an NV30 so he can talk to it and tell it what an awful video card it really is!

 

[Bouncing Zabaglione Bros.]

Why should ATi be paying attention to G70 numbers reported for a G70 that itsn't shipping yet?...

So they don't end up getting raped the same way that Nvidia did when they didn't pay attention to what R300 was going to be? When Nvidia were suprised by R300, they were forced down a route of trying to adapt and market the NV30 against something that was completely superior, and ended up cancelling NV30 and all the associated R&D costs.

Everything is judged against what your competitors make, and if you are too far off the curve, you end up having wasted your time and money. Besides, I'm sure that Nvidia and ATI have a better idea of what their competitors are up to than most of the rumours flying around. No company dares get complacent in a market that moves as fast as this one does.

 

[WaltC]

Which Nvidia "funded" and "manipulated" web sites are you referring to? Proof or linkies please.

Someone should buy Walt an NV30 so he can talk to it and tell it what an awful video card it really is!

I'm very sympathetic to your obvious lack of memory cells and synapses, and note with sadness that it's apparent that you weren't a regular reader of B3d articles and forums during '02 and '03 as I was, else you'd recall us beating that issue to death for months and months and months...

We regularly digested and critiqued such tripe as [H]'s "Doom 3 Preview Benchmarks," which [H] reported running on sealed boxes furnished by nVidia running only nVidia-furnished benchmarks--not to mention the fact that the "Doom 3 Preview Benchmarks" setup by nVidia ran on [H] about a *year* or so before the game was released. Even Carmack recommended ATi over nV for Doom 3 early on, and ran an early build of it on R300 while stating that nVidia just had nothing comparable to test. (Sigh...I guess you need a link for that, too?) I just can't fill you in here as I don't have the time or inclination. Sorry.

And you apparently do not also know that nVidia publicly and formally cancelled nV30 production shortly after it attempted to start it, and that even nVidia's CEO JHH is on record saying "nV30 was a failure." So even if someone wanted to run out and buy me a spanking new nV30 card they couldn't--since nVidia canned it long ago.

Geeze--I'm appalled sometimes at the incredibly short memory spans I encounter. Trust me when I say that a good memory is a Good Thing...

 

[RussSchultz]

Your accusation of "NVIDIA funded and manipulated" goes well beyond whatever facts you just recited.

 

[WaltC]

So they don't end up getting raped the same way that Nvidia did when they didn't pay attention to what R300 was going to be? When Nvidia were suprised by R300, they were forced down a route of trying to adapt and market the NV30 against something that was completely superior, and ended up cancelling NV30 and all the associated R&D costs.

Everything is judged against what your competitors make, and if you are too far off the curve, you end up having wasted your time and money. Besides, I'm sure that Nvidia and ATI have a better idea of what their competitors are up to than most of the rumours flying around. No company dares get complacent in a market that moves as fast as this one does.

The problem is, though, that the G70's aren't being shipped and so ATi cannot run out and buy a batch of them for in-house testing. From late summer '02 through spring '03, ATi was *shipping* the 9700P (I bought one 9/02) and nVidia could buy and test 9700P's to its heart's content. nVidia got burned *not* by pre-shipping marketing numbers surrounding R300 prior to R300 shipping (Ati did not circulate any that I recall), but nVidia got burned by the *R300 itself* during the nine months it was shipping before nV got nV30 out of the door--only to turn around and cancel nV30 production afterwards and call it a failure.

See what I'm saying? It's one thing to react to a product your competition is shipping--quite another to react to benchmark numbers being circulated for your competition's product before it has shipped and before you can verify the veracity of any of those numbers. The nV30 benchmark numbers circulated widely pre-nV30 but post R300 were *bogus* to the point of being absurd. nV30 "in the flesh" was much worse compared to R300 than the nV30 pre-shipping benchmark publicity numbers indicated.

So, unless ATi enjoys shadow boxing and ghost chasing, I'd be surprised to see the company doing anything major which is based purely on rumor and wild speculation. I know I wouldn't worry about it if I was them. Instead, I'd concentrate on making my upcoming products everything I'd already know my markets expect. Then, when the G70 ships, ATi can pick up a few and it will soon become apparent what they'll have to do, or else what they won't have to worry about.

 

[WaltC]

Your accusation of "NVIDIA funded and manipulated" goes well beyond whatever facts you just recited.

"nVidia funded" could mean anything from sites paid to run nVidia advertising to sites paid to slant their accounts so as to present a nVidia bias. Or it could mean a bit a both. Granted, I don't know the precise details but I have no clue as to how you can recall the sites which ran such glowing pre-shipping accounts of nV30 *without* developing such logical suspicions. What you call "accusation" I call recollection...

"nVidia manipulated" simply means that some sites were talking a lot to nVidia PR types who were leading them down a merry goose chase... I only submit the sites so affected at the time were quite ready and willing to be led down those blind alleys as their glowing, pre-shipping nV30 prose and "benchmarks" indicated.

But you are right in one respect, I suppose--maybe they were just drooling morons who simply didn't know when their chains were being yanked? Is that how you'd prefer to see it?

Anyway, it is an historical fact not of my making that the pre-shipping nV30 benchmark-number publicity generated by nVidia and several web sites at the time turned out to be completely bogus. If it pleases you to "accuse" these parties of nothing more heinous than stupidity, well, I won't argue the point further... I just don't see anything at all positive or enlightening that came out of the glowing pre-shipping nV30 press that popped up all over the Internet after ATi began shipping the 9700P.

What's the old saying...? "A lesson forgotten is a lesson to be relearned"...? That's pretty much where I'm coming from here.

 

[incurable]

And you apparently do not also know that nVidia publicly and formally cancelled nV30 production shortly after it attempted to start it

Quite understandable if he doesn't "know" that, I doubt anyone can keep up with everything you make up in your head ...

 

[Jawed]

But you are right in one respect, I suppose--maybe they were just drooling morons who simply didn't know when their chains were being yanked? Is that how you'd prefer to see it?

I think it's the only way to take it. NVidia spanked most "technical" websites with its PR at this time, whether it was over NV30 or 3DMk03. There are plenty of editorials admitting as much.

Jawed

 

[WaltC]

Quite understandable if he doesn't "know" that, I doubt anyone can keep up with everything you make up in your head ...

OK, excuse me for lying when the truth is that nV never officially cancelled nV30 production and JHH never declared nV30 a failure in a public statement. Thanks for enlightening the world that I made it all up...

Uh, do you have any more fiction spinning around in your head?....Unbelievable.