JHH Blows top, breathes fire...

[WaltC]

According to this snippet, http://www.theinquirer.net/?article=15984 , JHH is losing his cool presumably because people seem skeptical of his claim that R4x0 is "three years old"...

I have to wonder if his responses as reported here were especially wise, seeing as how they tend to underscore that the nVidia chief is anything but confident. It's got to be tough, too, I'd imagine, to go from recommending "three year old" technology as nVidia did for all of last year with nV3x and ps1.x, to pushing ps3.0 with "60 millions of transistors" this year for nV40.

The thing that may have JHH most rattled, though, is the belated realization that ATi may have looked at those "60 millions of transistors" and decided to forego them in R4x0 because of practical concerns such as actually being able to do more with their gpus than to market them, things such as being able to manufacture them with suitable yields and bring them to market so that ATi could sell them, as well as talk about them. If this account is accurate, it certainly seems as if R4x0 has managed to strike some kind of raw nerve inside of nVidia.

As far as proving what a non-gimmick nV40's ps3.0 support may actually be, it seems nVidia missed a golden opportunity to underscore their ps3.0 assertions by failing to include some revealing in-house demos of nV40's ps3.0 functionality in the review packages it sent to hardware reviewers along with nV40 when the product was officially launched. Seems a bit odd, I think, to hear JHH opine on what an important feature ps3.0 support in nV40 is, even to stating it "cost them 60 millions of transistors," while at the same time the company hasn't provided a single demo to illustrate what it does so much better than ps2.0 that is so compelling. With 3d technology, especially, "seeing is believing," isn't it?

 

[digitalwanderer]

"It's funny because it's true!!!"

Nice analysis of his reaction Walt and I tend to agree with you, this ain't a very promising sign for nVidia if he loses his cool so easily over that.

 

[PaulS]

The Inq is only two weeks late, I see.

From The Conference Call Thread:

- Talked about R420 being effectively a 3 year old architecture, and customers will pick the NV4x over that based on features such as Shader 3.0, since theirs is the only true next gen GPU
- Started arguing with one caller, who pushed them about whether PS3.0 will actually make any difference NOW. They went back and forth several times, and it got reasonably heated and amusing. They said they'd call the guy back after the conference call

 

[Evildeus]

http://www.beyond3d.com/forum/viewtopic.php?t=12575

 

[PaulS]

It's been locked all day, Evildeus

 

[PurplePigeon]

Is this referring to nVidia quarterly conference call? The description in the article sounds a lot like what happened towards the end of that call, although that happened some time ago, not last week as the article states. It had been some time since I'd bothered to listen to an nVidia call, so it was a bit of a rude shock to me to realize just how blustering JHH was.

After you listen to him talk, it's quite clear that his character is reflected in nVidia PR tactics. I have more sympathy for the marketing department at nVidia now.

Assuming that this is indeed the same call that I remember, the account is pretty accurate in the details. JHH did say something along the lines that it required 60 million transistors for them to implement SM 3.0. And if that's the case, I wonder just how much it will cost to get it implemented in the lower end parts.

I disagree about the assessment of JHH as rattled. I would actually describe his reaction as a sort of arrogant indignance.

The other thing that had me thinking was the JHH's discussion in the call about the one versus two molex issue, and how he claimed that only one molex connection was really necessary... And then lo and behold, not so long afterwards we start getting word that yeah, maybe only one molex is necessary after all so long as the power supply is sufficiently robust. That's when I really started wondering if JHH was more responsible for how nVidia markets its products than I previously thought.

 

[Evildeus]

It's been locked all day, Evildeus

Yeah because it's been talked in other threads....

 

[WaltC]

I have more sympathy for the marketing department at nVidia now.

I disagree about the assessment of JHH as rattled. I would actually describe his reaction as a sort of arrogant indignance.

The other thing that had me thinking was the JHH's discussion in the call about the one versus two molex issue, and how he claimed that only one molex connection was really necessary... And then lo and behold, not so long afterwards we start getting word that yeah, maybe only one molex is necessary after all so long as the power supply is sufficiently robust. That's when I really started wondering if JHH was more responsible for how nVidia markets its products than I previously thought.

I've long suspected that much of nVidia's PR tone was set at the top, in a sort of, "It's that way because I say it's that way," kind of approach to marketing. As to the dual molex PR spin, I think that's anything but clear at the moment as I can't see how using two molex connectors with an insufficient PSU would be any better than one... I mean, the PSU doesn't care or know what's connected to each of its plugs--it either supplies the power for proper operation of every device connected, or it doesn't. That's the thing--the PSU has to support everything else in the system as well as the gpu--so it really doesn't follow that anyone would ever want to use a 6800U with only one molex connected--which I guess is why nVidia engineers decided it needed two of them?...

 

[Pete]

"Incandescent with rage" sounds like an exaggeration if you've listened to the conference call. He was becoming somewhat exasperated with that last caller's rather confused line of questioning, though I didn't hear him lose his cool at all.

Maybe I missed it, as I wasn't paying full attn throughout the call, but did someone ask how that extra 60M transistors will trickle down to the rest of the NV4x product line (particularly, the budget chips)? I know he covered that for the 6800U with 12" vs 8" fabs and "only 9% greater die size."

I'm also surprised no one asked him about speed relative to ATi's $500 card. Sure, as J-H said, I'd rather pick the $500 card with SM3.0 support, but no one asked if that involved a performance trade-off (and, again, if that would trickle down to their cheaper NV4x GPUs).

(OT: I'm curious if those analysts read B3D and other web reviews, as they referred to "reviews" a few times.)

 

[PurplePigeon]

"Incandescent with rage" sounds like an exaggeration if you've listened to the conference call. He was becoming somewhat exasperated with that last caller's rather confused line of questioning, though I didn't hear him lose his cool at all.

He was getting exasperated, but he also wasn't doing much to try to understand the caller's question, which I thought was actually pretty clear; and the tone of the answers also started to have a faint whiff of borderline-polite intimidation (maybe just my imagination). But yeah, "incandescent with rage" is not how I would describe JHH, although it is the sort of florid exaggeration that I've come to expect from The Inquirer.

Basically, JHH was claiming that customers would (in his opinion) obviously choose their SM 3.0 product over ATI's product because SM 3.0 was such an important feature. The caller was trying to ask if the fact that ATI's upcoming product would be native PCI Express (instead of a bridge) would offset the attraction of SM 3.0. He was attempting to assess the importance of one over the other. In the end, they all just ended up talking past each other...

 

[jvd]

Well hey. It was nvidia after that taught us if you can millk the tech for a ton of years then milk the tech.

The x800 may be based of 2 year old hardware (wtf does 3 years come from? in august it will be 2 years old) but as seen it is the fastest hardware out .

 

[ZenOps]

Well... There are some tech demos that use Pixel and Vertex 3.0.

Nalu the mermaid uses what 19 passes, 3.0 water effects and spanning technology for her hair?

But then again, any 3D programmer in their right mind would probably gasp at the idea of code so "bloated" that it would require that many passes to start with. I'm all for visual quality, but you don't just add things because they are there, you add them if they look better or have a definite performance delta.

3.0 effects can be made longer, so that certain effects can be done in one pass... Now honestly, it can also be a "bad" thing as the longer you make the shader length, the longer it takes to go through the shader. It might come down to: Do you want the effect in PVS 3.0 at 400Mhz done in 19 passes, or would you rather have the exact same effect done in PVS 2.0 in 25 passes at 600Mhz.

It would look the same in either case, and PVS 2.0 can run faster if the chip is run at higher clockspeeds than the 3.0 counterpart. (clock for clock though, 3.0 is faster.)

 

[BRiT]

Why would Nalu demo ever use more than 1 geometry pass since a single pass could be of infinite length, no?

 

[BetrayerX]

Why would Nalu demo ever use more than 1 geometry pass since a single pass could be of infinite length, no?

What will you swallow faster?
A giant mammoth sized 3 foot hot dog......?
......or 3 or 4 Ballparks?



Seriously, there's a point where the engine will choke, even if it can process a huge shader in one pass.

 

[BRiT]

Why would there be any differences between 1 pass or multi-pass (aside from inefficiencies of having to resend the geometry data over again)? Would not a decent compiler be able to resolve the register references and dependent textures just the same on the single-pass program?

Or are you merely stating that their current compiler is not up to snuff to handle longer shaders?