Where does this leave the competition?

[Geeforcer]

For this thread, I am ignoring Nvidia and the unannounced NV30 for obvious reasons, and would rather focus on the chips introduced earlier this summer: Perhelia and P10 and only one market: consumer graphics enthusiast (high-end).

Matrox. For all the years of anticipation as well as early hype and enthusiasm Perhelia can only be considered as "competition" in light to its recent release. Due to the rather disappointed performance highlighted earlier, I had major concerns regarding Perhelia’s ability to compete even with GF4ti - for all intends purposes the card is not a competition for R300 in the specified market. The Triple Head is hardly a selling point to high-end gamers, and even if the importance of that feature was to skyrocket, it would not negate R300s full DX9 support and overwhelming performance advantage. Considering that the two cards are at the similar price points, the competition Perhelia provides to R300 is akin to the threat a World War One era gunboat poses to Seawolf-class submarine.

This does not necessary signal the end of Matrox, businesses that were buying G550 might still embrace Perhelia is spite of its hefty costs. However, Matrox has invested heavily in the "gaming" portion of the chip, and at this stage Perhelia the gaming card is a failure. Matrox has lost its relevancy in high-end market in 99 and has failed to successfully re-establish itself in that market.

Moving over to (C/3D)labs. Comparing what we know about P10 and R300 at this moment, it seems that P10 might be more flexible is certain areas. That could translate into potential advantages in professional field where this flexibility can be takes advantage of, but its largely irrelevant in high-end consumer market. What IS relevant is full DX9, which P10 does not have. ("We think you need at least .13 for that" - famous last words). OF course, there are no apps that would sell DX9 cards (hell, even the API itself is not out yet), but I doubt that having choice between a non-DX9 and DX9 card many would opt for the letter, especially considering the name recognition advantage ATi hold over 3Dlabs. The absence of a consumer P10 card does not help the matters.

For some time now, we begin the year with hopes for healthy high-end completion between multitude of vendors, but it looks like that for a third year in a row only 2 companies will matter: Nvidia and ATi

 

[hughJ]

considering how giant of a leap the performance gap is now, nvidia will wind up having to make an equally giant leap in price cuts to stay competitive until the nv30 is out...

how cheap will the Gef4Tis get, I wonder?

 

[Geeforcer]

Nvidia could perhaps combat 9000 is they were to drop 4200 prices, but I don't see how the price drops could significantly increase the ability of Ti line to compete with R300: the performance and features delta is just too great.

 

[Mize]

considering how giant of a leap the performance gap is now, nvidia will wind up having to make an equally giant leap in price cuts to stay competitive until the nv30 is out...

how cheap will the Gef4Tis get, I wonder?

A $200 Ti4600 would be nice

Mize

 

[Joe DeFuria]

Yes, performance and feature wise, the GF4 Ti just is not in the same competitive league as R-300. So to remain a viable product, nVidia and its partners will have to reduce pricing of the Ti cards to sell them through.

The question then, can nVidia and its partners make money on the Ti at those prices, or will the GeForce4 Ti orders pretty much stop once the current stock sells though?

Having said that, at this point ATI lacks one critical part that some web sites did note: ATI is lacking that $150-$200 "mainstream gamer" part in their line-up.

Presumably, nVidia could sell a boat load of Ti-4600 boards if they price it at $200. R-300 would be the best card, but a Ti-4600 at $200 would arguably be the best bang-for buck deal.

So, ATI is in a similar position that nVidia was at the GeForce4 launch. The GeForce4 MX was "worse" than their previous Ti-200, and the Ti 4400 and 4600 parts were on the expensive side, out of reach for most. nVidia's solution was by getting out a new SKU a couple months later: the Ti-4200 to compete in the $200 price point. This was apparently an "unplanned" part, but ATI forced it by aggressivly pricing the Radeon 8500.

We have read that ATI will release some sort of R-300 variant in a few months time to compete in the "much less than $400" price point, but higher than the RV-250 price point. Some are speculating a 4 pipe version of the R-300. If true, I would expect it might loose the 256 bit interface as well, to keep the same bandwidth / pixel rate ratio.

The $200 price point will be an interesting segment to watch. It may potentially be a 4 pipe R-300 competing with the Ti-4400 or 4600.

 

[Geeforcer]

This is a good point: despite late introduction Ti4200 is the best-seller, for obvious reasons. The fact that it belonged to fastest GPU family currently available certainly help.

To get some insight in Nvidia's most probable (and effective) response, we need 2 dates: 1. Wide availability of 9700 and 2. NV30 launch. Much of what is likely to happen price-wise will depend on these 2 dats.

Slightly off-topic, I wonder if anyone here has any isghit in Creative plans.

 

[Dave Baumann]

http://www.beyond3d.com//articles/p10tech/index.php?page=wvp3.inc

 

[mboeller]

The $200 price point will be an interesting segment to watch. It may potentially be a 4 pipe R-300 competing with the Ti-4400 or 4600.

IMHO no, an 4Pipe RV300 would compete with the rumoured NV31.

Maybe the R200 will have an longer live than we think at the moment

 

[Mize]

The question then, can nVidia and its partners make money on the Ti at those prices

Since they can make money on the Ti4200 at those prices I'd say the can do it with the 4600 as well.

Mize

 

[Geeforcer]

http://www.beyond3d.com//articles/p10tech/index.php?page=wvp3.inc

Thanks Dave.

From the article:

...One thing that Creative are keen on is getting a product to the 'SoundBlaster price point', i.e. roughly about $99.

<snip>

...In the interim its not 100% clear exactly what Creative will be doing with 3Dlabs technology, and they may even continue producing boards for other chip vendors to meet market segments that their own technology currently does not fit into, while more parts are being readied.

To anyone expecting a high-end product this year, neither of these statements is particularly encouraging.

 

[Joe DeFuria]

Since they can make money on the Ti4200 at those prices I'd say the can do it with the 4600 as well.

Um, the 4200 doesn't nearly have the same core or memory speeds of the 4600, hence it's lower cost.

IMHO no, an 4Pipe RV300 would compete with the rumoured NV31.

Possibly...but then where does the rumored NV28 fit in?

 

[Mize]

Since they can make money on the Ti4200 at those prices I'd say the can do it with the 4600 as well.

Um, the 4200 doesn't nearly have the same core or memory speeds of the 4600, hence it's lower cost.

Clearly...but many (most?) 4200s are capable of hitting 4600 speeds so perhaps a $200 4600 that rarely overclocks well is what might happen. Beats me.

Mize

 

[jvd]

how about a 200-250 mhz r300 ... they can get that out for about 200 ... it should perform a little worse than a ti 4600 but it will be full direct x 9 and they can have it ready in a two omnths. all they have to do is bin it .

 

[Joe DeFuria]

JvD,

That's a good point. I don't expect a reduction in core speed alone to reduce costs that much, but the fact that ATI would also be able to use "slow" 200-250 Mhz DDR Ram may indeed lower the cost.

I could see a 225/225, 8 pipe R-300 pricing around $200...especially if they cut the memory to 64 MB.

 

[KimB]

Clearly...but many (most?) 4200s are capable of hitting 4600 speeds so perhaps a $200 4600 that rarely overclocks well is what might happen. Beats me.

Mize

It's not all that common for a Ti 4200 to hit Ti 4600 speeds.

Additionally, most Ti 4200's use a different (cheaper) board design.

 

[RussSchultz]

Knowing what I've learned over the past two years working at a fabless semiconductor, I don't think that it is economically viable to speed bin parts _that slow_.

In general, yield is hurt by particulates, human fabrication error, slop in the machines, etc. Generally, while there is some process variation, it is not so extreme that a part is fully functional but only at half speed. Considering acceptable yield is up at 60-80%, and most of that are due to completely failed parts and not due to not making the speed grade, they'd have to produce on the order of 20 fast "highend" parts for each "low end" part.

And granted, while each "low end" part would be "pure" profit because it would have gone in the trash bin anyways, I don't think ATI could make enough parts to harvest enough failures to make it economically viable.

Plus, thats quite an inversion to where you want to be. Sell your mainstream part for X, charge a premium for "fast" parts.

Cost to ATI is silicon area + packaging + test time. These "slow" parts aren't going to be any smaller than the normal part, which is where the majority of the fabrication cost comes from--silicon.

 

[Joe DeFuria]

Knowing what I've learned over the past two years working at a fabless semiconductor, I don't think that it is economically viable to speed bin parts _that slow_

I guess nVidia did the economically "non-viable" thing with Geforce4 Ti then.

they'd have to produce on the order of 20 fast "highend" parts for each "low end" part.

Is that wht nVidia does with the 4200 relative to the 44 and 4600? I would have guessed that they sell a lot more 4200 boards...

Cost to ATI is silicon area + packaging + test time. These "slow" parts aren't going to be any smaller than the normal part, which is where the majority of the fabrication cost comes from--silicon.

Hmmm...the PCB and total packaging of the Ti-4200 (ram packaging) is singificantly different than the 44 and 4600 parts....

Look, the bottom line is, that nVidia put together a "slow" version of the $400 GeForce4 Ti in order to compete in the $200 segment, so why does it seem so economically non-viable for ATI to do the same?

There's a big difference between economically "ideal" and economically "viable." I would agree that the 4200 is not an economically "ideal" part, nor would a "slow" R-300 be economically "ideal." Nvidia's "ideal" part for the $200 segment was the practically still-born MX 460. Competition from ATI is what made the MX460 economically non-viable, so the less ideal, but at least viable, 4200 was born

So, how "economically viable" is it for nVidia to sell the 4600 boards at 4200 prices?

 

[RussSchultz]

They're most likely selling the exact same parts at a reduced cost and making less margin on it. Remember how the 4600 was "immediately available", and the 4200 has just recently shown up? Its because NVIDIA saturated the high end market and has to offer a mid priced product to continue moving product.

Also, those few months in production allowed NVIDIA/TSMC to refine their production and test program to more precisely determine the edge between good part/bad part and hence bring yield up, allows lower cost per part, and then makes marketting the low end product more feasable.

ATI can do exactly the same things. Could they market a low end part? Yes. Cost half as much at retail? Maybe. Now? No. In a few months, probably.

 

[3dcgi]

I don't know any numbers so this is all conjecture, but I'm sure Nvidia has been making a very high margin off the 4600. And decent margins off the 4200. I believe a 4600 will have a higher profit margin at $200 than a slow binned R300. The extra 47 million transistors in the R300 means it's not a cheap part no matter what the clock speed is.

 

[MuFu]

AFAIK, both of the new boards are relatively cheap to manufacture. Considering the fact that the 942 board accomodates the expanded memory bus and uses an 8-layer PCB, it doesn't carry a huge premium over the R200 boards it succeeds due to the level of integration in the R300 ASIC (although I guess if some sources have to step up to 10 layers to maintain SSTL_2 integrity it will be more costly). Of course, the ASIC itself is costly.

The RV250 is a masterpiece of economical design; the 964 boards cost 25% less to manufacture than nVidia's equivalent and the ASIC is obviously cheap to produce. Understandably, the margins are going to be great. Even 958, with its BGA-packaged RAM, should be much cheaper than any R200 board ever was to manufacture. The DCs are cheap too; just 4-layer PCBs that actually cost less than the Ripper 2 chips themselves.

MuFu.