nVidia Conference Call Tomorrow

Actually, maybe they are talking about clockspeeds? The NV30 is clocked at 400Mhz, while the rumored speeds of NV35 is 800Mhz. I'm confident they are not talking about performance.

 

For what, $5 a board? I'm surprised TNT2s are even being produced still...

 

Just in case there's any remaining doubt about this, but Uttar you are incorrect in your assessment.

The X-Box arbitration caused NV to defer the X-Box revenues because they couldn't be certain of what the actual contribution to revenue would be at the end of arbitration... The expenses were not deferred because there is no question as to what those amounts were.

Expenses are reflected on an incurred basis (look up Generally Accepted Accounting Principles, or GAAP as it is more commonly referred-to, if you want to know more). Therefore, the related expenses have already been reflected in prior quarters, and the conclusion of the arbitration shows its full results on the bottom line for this quarter.

So the full amount was fully accretive to net income. [EDIT: before taxes]

[EDIT: wording]

 

800mhz is not the rumored spec for the Nv35. They cant even get 500mhz totally stable with an extreme cooling solution. Even With low-k, the first generation boards on that process are not going to hit anywhere near that high.

If they impliment the low-k, get decent yields, and continue to use the dustbuster they *may* hit around 625-650mhz

 

Maybe they are being creative with "256-bit bus" versus "128-bit bus" and will just sort of gloss over how that could apply to the competition. Even if it was still at 400 MHz, they could say the bandwidth had doubled and sell that pretty easily as "2x faster".

When was nv40 intended to be launched? Maybe they are referring to shipping that card this year (or atleast a bit more "within the year" than the nv30 was)...they did seem hesitant to give a time scale in association with the part they were saying was 2x faster, as well as shying away from the term "refresh".

 

Hmm, now that you say this, it actually makes sense

But... Well...

With this, you've got $51M net profit or 30 cents a share.
Without this, you've got 14 cents a share.

That means about $25M
But, err, wasn't the original number of $40M?
So they only retrieved $25M...

Any explanation?


Uttar

 

Damn, I really should login and stop double-posting...
Just after posting this, I realized what my mistake war.
There's taxes, and the $40M doesn't count taxes.


Uttar

 

Interesting. How did you calculate those numbers?

 

Why? Nvidia is not selling chips to the public but to board manufacturer. It's not impossible to ship 1.5 NV3*s in the quarter till their's always a gap in time between the chip being sell to their partners and the card being sell to the public/OEM.

 

I haven't actually looked over the numbers. I was just going off of the numbers that you guys had posted. But yes, this sounds about right.

 

I don't think it's the first time Nvidia has made dubious statements about 2x increases in speeds when neither the clock speed, the RAM or the benchmark speed showed any of it... Did something like that happen with the G2? Can't quite remember now... Anyway, they could compare NV35 with the optimised drivers at release vs. NV30 with the current crop of unoptimised drivers...

Or perhaps just have a 256 bit pathway and claim you doubled the speed.....

 

I think its perfectly feasible that by fixing a few design flaws, aiming for 600mhz, and doubling the bus on a currently bandwidth starved chip could actually provide double the performance on high-res AA situations. especially when taking into consideration that the drivers will probably be highly improved 6 months from now.

sadly I dont think that nVidia should have any problems doubling performance over the nv30 and deffinately not ATI

 

If doubling the performance of the FX were that easy to do in the near future, I should think they would have done it already. I don't see anywhere that the 2x performance design was even remotely related to NV35, and I think it's fairly safe to say that we won't see anything close to a 2x increase in performance until at least NV40.

Look at it this way, if raising the core clock to 600 and eliminating the memory bottleneck with even faster DDRII on a 256-bit bus could give twice the performance of the FX, that would imply the current FX design was so memory bandwidth limited, that you could clock the core back to 300 MHz while keeping the memory at 500 MHz and you wouldn't see much of a performance drop. I would bet that's not the case even at 1600x1200 w/ 8x AA and 8x Aniso.

As for the profit per share calculations, without the X-Box revenue I think it would be closer to 6 cents per share. Assuming net income = revenue - expenses, we know net income including X-Box settlement, and we know the revenue including the X-Box settlement as well as the ammount of the settlement, and expenses remain constant. So if we decrease revenue by $40.4 Million, we decrease net income by $40.4 Million as well, giving us $10.5 Million. Their shares were based on net income, not net profit, so 30 cents a share at $50.9 Million revenue would be about 6 cents a share at $10.5 Million.

 

1. That's not really the way to look at it...
Take a GFFX. Use 4x AA, 8x Aggressive Aniso - note the results
Take a GFFX at 500/300, note the results
Takea GFFX at 300/500, note the results

I'd bet the GFFX at 300/500 is faster than the GFFX at 500/300
And anyway, if you compared 6xS & 8x AF with 6x & 16x AF, which would result in identical quality, it would be perfectly reasonable to assume twice the performance because 6xS is simply abysmal due to the SSAA part.

2. No. You're not counting taxes and several other factors. nVidia said that without the XBox settlement, it's 14 cents.


Uttar

 

It might be, but I don't see why that would matter. If it's bandwidth limited at 500/500, it's going to be even moreso at 500/300. That doesn't have much to do with the statement I was adressing though.

The claim was that an FX at 600 MHz core with no memory limitation could be twice as fast as the current FX at 500/500.

My reasoning is that a 600 MHz core with no memory limitation is in theory going to be twice as fast as a 300 MHz core with no memory limitation. Therefore, if the 600 MHz core with no memory limitation is twice as fast as the FX, then the FX must be memory limited to the point that it's performing the same as something with a 300 MHz core and no memory limitation.

If that were the case, one should be able to bring the core clock down to 300 MHz on an FX and not see a difference in the performance. Whether or not it is still bandwidth limited at 300/500 doesn't matter with this theory. If it's still bandwidth limited, it shouldn't be any slower than it was at 500/500, since the memory bandwidth is kept constant and is limiting it in both cases. If it's not bandwidth limited at 300/500, then it still shouldn't be any slower, because according to the original theory a 300 MHz core that isn't memory bandwidth limited is what the 500/500 part is performing like to begin with.

My stance, however, is that this is incorrect. I don't think that the FX at 500/500 is so bandwidth limited that it is performing as if it were a 300 MHz part that isn't bandwidth limited. I claim that clocking the core back to 300 MHz would show a significant decrease in performance, even at the highest resolution, AA, and Anisotropic settings where it would be the most likely to remain memory bandwidth limited. Since I believe this to be the case, I therefore claim that producing a 600 MHz core with no memory limitation that is twice as fast as the current 500/500 MHz part (in the same tests, with the same settings) is not possible.

 

I didn't see where they sait it was 14 cents without the X-Box revenue. I only saw reports saying that it was 30 cents a share based on net income, (which is roughly revenue - expenses, maybe take 35% off for tax if their including that, but that shouldn't change the proportionality much). They probably know their finances a little better than I do

 

Maybe it's finally 8 pipelines!

And...

Mid January to mid February, mid February to mid March, mid March to mid April, yep, four. 8)

(Sorry, couldn't resist.)

 

BTW, thanks to all for summarizing the call! It was most interesting indeed...