Geforce FX will cost $450

They're good at football too - they won the English Premiership last season...

Oh, hang on a minute - that was Arsenal. Oh well, much the same thing...

 

LOL.

Oh yeah, GO BLUES!!!!

 

If it is true that the high end one will sell for "only" $450 it's only because of the competitive pressure ATI will put on its release. Originally I saw nVidia's MSRP prices at $499 and $399 (US). I also wonder if selling it at a lower price reflects a lackluster performance in comparison with the current R300 products, which debuted for only $50 less. Be interesting to see.

 

One quick note about pricing:

The Street price of the FX will be whatever it takes to sell the quantity available. It doesn't matter what the MSRP is, nor does it matter how much the product itself costs to make.

The three biggest factors for the NV30 street price:
1) Relative performance to the R-300 products. (and/Or R-350 products)
2) Timing of product shipments (as this is directly related to the selling price of R300/R350 products)
3) Quantity of product shipments.

Because we don't really know much about any of the above factors in reality, we really have no idea what the FX will sell for on the street, so speculation is rather pointless....and quite frankly, nVidia probably doesn't know what they will sell for either. (And that doesn't sit well with OEMs.)

About all I would venture to say is that initially, whatever the NV30 price is, it won't be worth it to the vast majority...because initial prices will be 'artifically' high until there is more widespread availability. (This applies to any new product from any vendor.)

Given that we still have no idea when initial shipments will start, widespread availability won't likely happen until sometime in February at the earliest. By then, ATI's OEMs will be have dropped the price of R-300 parts even further. Currently, the 9700 non-pro goes for $230 on Pricewatch, and the 9700 Pro for $280. I suspect the 9700 non-pro may hit the $200 mark by Feb, and the 9700 will probably be about $250.

If NV30 is seen as "only on par" with Radeon 9700 Pro, I expect the selling price to be just slightly higer once it's available in quantity, as nVidia's brand will probably still help it move at slightly higher prices. If NV30 is accepted as an all around better product, the price will be a step higher than 9700 Pro.

On a related note I'm not convinced that ATI's 256 bit products will be viable at less than $200, so my hunch is that they would be phased out once competitive forces (from nVidia or from ATi's lower end cards) dictate the selling price needs to go that low.

Speculation: Radeon 9700 Pro will be phased out completely once NV30, R350 and RV-350 starts to ship. Radeon 9700 non-pro may still be a viable product however at the $175-200 price point. RV-350 will replace the RV-250 and sit in the $75-$125 price bracket. The $150 spot will be filled by either the Radeon 9500 Pro, or a high end RV-350 variant.

 

Hmm...in what time frame? I don't see the RV350 replacing the RV250 in the first half of the year...I see it more likely replacing the crippled R300 (9500) parts at the least, and that is accepting your 4 pipeline reasoning as a given...look at the life of the GF 4 MX...why would ATI need to discard the 9000 so quickly? (Presumable) cost savings from 0.13 should take a while to manifest at the very least, and I'd think this would hold true for the NV31 as well. I don't think the "DeltaChrome"(?), or anything else like it, will be very cheap, but if it pans out I guess this could change things..

In any case, I think for longevity and profit ATI would require an 8 pipe RV350...or atleast, like the R300, a chip that can be used both 4 pipe and 8 pipe. This would allow room for a smooth migration as 0.13 yields improved, and better profitability over a wider price range if they can manage a "low cost" design with 8 pipes (I really don't see why not...they seem to execute such goals well).

I don't think the 9700 Pro will disappear unless the R350 has an extremely marginal performance lead in some configuration...but that configuration would have to be more profitable for ATI somehow, and I don't see how that would come about. I've always felt the only sane way to release the R350 on the expected schedule is as a premium and/or workstation part...but even if the NV30 doesn't allow that, I still don't see the 9700 Pro disappearing as the designation is just a matter of yields, which I'd think would just improve over time...it is the non Pro I'd expect to disappear, along with other yield targetted R300 parts (as the RV350 replaces them). Seems to make sense as far as profitability and "performance crown" goals.

 

Hi demalion...Happy Holidays!

Well, I'm guessing a "Spring Launch" for both RV-350 and R-350. That is, product announcments in Late Feb, Early March, with wide availability sometime in April. Actual launch date and product ramp is likely tied a bit to NV3x product line availability. So the later the corresponding NV3x product ships, the later I expect RV-350 and R-350 to arrive.

Well, I don't see the 9000 as being "discarded". Only moved down a price notch. (To where the 7000 series is currently selling.) Remember, UMC will be producing the RV-280 (AGP 8X version of the 9000) chips at a cheaper price than TSMC is currently selling the RV-250 for, allowing this move to happen, and still allow ATI some profit margins at the reduced price point.

The other reason to move the 9000 down in price point "so quickly", is because nVidia is suppossed to also launch a DX9 "$100 price bracket" part this spring as well (NV34 or NV31.)

We'll just have to agree to disagree here. a low cost 8 pipe design on 0.13 is one thing...but a DX9 compliant 8 pipe design is another. It's my opinion that an 8 pipe, DX9 compliant part has too large a transistor budget on 0.13 to be viable in the value segment (and also too power hungry for the mobile segment.)

My reasoning is this: the "problem" with the 9700 Pro come this spring, will be its ability to have decent profit margins. Once it no longer is the "top performer" (either replaced by the R350 and/or NV30), it will become sort of like the GeForce4 Ti-4400...kind of in no-mans land. No longer the "best", but likely not cheap enough compared to the 9700 non-Pro and 9500 Pro.

The R350 can still probably a money maker if it MSRPs well above $300 and assuming more than "marginal" performance increase over 9700 Pro. Enthusiasts are typically willing to pay the higher premiums to have the "fastest", as long as there is some tangible performance increase.

I don't really think there's going to be much room at all for a $400 MSRP card come this spring. With 9700's around the $200 range, only the most die-hard enthusiasts would be willing to spend significantly more for "the" fastest card. I think there was only a brief period where $400 was viable, because the R300 offered SO much more than the GeForce4. But now that window has passed, I don't see a new card (from either ATI or nVidia) being successul at that price range.

So the way I see it shaping up this spring (Street prices): (Note: I'm going with the assumption that the NV31 is a "midrange" chip, and the NV34 is the "value" chip, as shown in an nVidia road-map somewhere...that might turn out to be backwards or completely wrong, though.)

* $350+ Bracket: Not viable anymore. However, might see some "built for show, extremely limited quantity" boards. More for mind-share than anything. An NV30 "ultra" or R350 "Ultra", perhaps equipped with 256 MB Ram.

* $300 bracket: High-end R350 variant, and possibly NV30 variant.

* $250 bracket: left-over 9700 Pros, possible NV30/NV34 variant in here. I don't see this price bracket being very successful though. (See GeForce4 Ti-4400). Enthusiasts will pay the Extra bucks for the top-of the line, and everyone else will get the "better" bang-for buck" cards at the $150-$200 price point.

* $200 bracket: 9700 non-pro. Possibly a low-cost variant of R-350 though, if one is made. Also, the NV31 should try and fit in this price bracket.

* $150 bracket: Possibly 9500 Pro, or high-end RV-350 variant. Either low-end NV31 variant, or high-end NV34 variant.

* $100 bracket: RV-350 and NV34

* Sub $100 bracket GeForce4 AGP 8X and Radeon 9000 AGP 8X

 

likewise!

...

Oh, I misread. OK, I don't disagree then...maybe I shouldn't have skipped breakfast.

Oh, I thought my RV280 assumptions went out the window with what I've been hearing about the 9100...I guess it is just the model name I got wrong, or am I just behind on the current "9100" gossip?

Hmm...is it really supposed to target that price bracket right away? *poof* to the GF 4 MX? That just seems odd without a full DX 9 lineup...if the GF 4 cards remain in the marketplace, that is one confusing lineup I'd think. I'd expected a higher price target, bolstered by "high end" mobility targetting, for the NV31. Am I missing an NVxx designation...what is going to be in the middle...are there going to be speed bins of the NV30 "non-Ultra"?

Of course you could be right...but out of curiosity, how many transistors do you think the "DeltaChrome" part has? I think ATI has routinely excelled in execution in terms of power consumption, and I'd think them outdoing S3 for low transistor count isn't too far-fetched...where would that leave us as far as power-consumption and design complexity? I guess we'll see.

Could be...but as I said I think the R350 will leave room for it...and unless the yields don't improve, I don't see why the 9700 wouldn't disappear instead of the 9700 Pro. I do agree that the cheapest part perceived to be connected to the high end will be a good seller, but that to me would be the RV350. That's why I think it is important for the RV350 to be an 8 pipe card atleast as an option, to eventually replace the R300 as the R350 moves down.

There is the R300 used on a 256-bit bus board...the clock speed of that to me seems to just be a matter of chip yields that will just improve. As long as any 256-bit bus R300 card is out there, I don't see why clock speed is a factor (i.e., if there is a 9700, I don't see why a "Pro" won't be offered).

There is the R300 used on a 128-bit bus board...I see the RV350 replacing these and offering improved margins. To do this over the widest range, I think 8 pipelines or the option of really high clock speeds is necessary. I guess we'll have to wait and see if my guess is a .... <Merry Christmas, pun omitted >

I think you under-estimate the importance of perceived performance leadership. The Ti 4600 was pretty expensive for a good while. It is the NV35 and R400 that will likely cause prices to drop...either that or price wars but I don't see that being in the interest of ATI or nVidia for their very best parts (I think another player would be required...3Dlabs....PowerVR?... ).

I thought the NV34 was NV31 as integrated video? Perhaps that is the NVxx chip I am missing from my equation.

When I'm talking about that range, the FX Ultra and the corresponding top end R350 are exactly what I'm talking about. The "non" (whatever) versions of the cards is where I perceive the price wars and value.

I'd say quite likely the NV30 and R350 "non" versions will be here. The thing is I don't see this being necessary for the R350 quite so early...it all depends on how much the parts cost. I do foresee it replacing the R300 in the mid eventually, just not in the spring.

That's why I think the RV350 needs to be 8 pipe capable, so the line up can migrate to RV350, R350, R400 later in the year. I just expect there to be 9700 Pros as long as there are any R300s being manufactured. I think this should be the R300 inventory clearance price bracket, and the RV350 should be capable of being profitable below this point.

Yes, it was when you were saying $100 and below initially for the NV31 that I was disagreeing. I think the RV350 and NV31 make sense to occupy here and below, if they are capable. I think both likely are. I'm not sure why you expect one (NV31) to target here and not the other (RV350). Both are targetting mobile markets eventually, aren't they?

Maybe my view of the nVidia lineup is bonkers...I guess I missed some info concerning the NV34.

Well, I guess we'll see.

 

I think we've both fleshed out our speculation, so I'll leave it mostly at that...thout I will clarify a few things / questions you had:

I'm not sure what you've been hearing about the 9100. Rumor has it that the 9100 is simply a re-branded 8500. (In other words, the 9100 is using the R-200 chip.)

The RV-280 chip is suppossedly an AGP 8X version of the RV-250. (The chip that is used in the current Radeon 9000 products.) There has been no indication of what the board level products using the RV-280 will be called. It may just be the "Radeon 9000 AGP with 8X AGP"

Things are verry unclear about the mid-range nVidia market. Here's the line-up I was speculating on:

NV30: high-end
NV34: value (replace Geforce4 MX)
NV31: Mid-range

Note that the NV34 is actually LOWER in performance than the NV31. Seems counter-intuitive, but the reason I put the line-up this was is due to a leaked nVidia road-map that showed the NV34 as the "value" part, and the NV31 as something between the NV34 and NV30. Again, that road-map could be out of date, or it could be completely wrong.

EDIT: Here's the link to that road-map:
http://www.nvmax.com/Articles/Previews/NVIDIA_GEFORCE_FX/supp/roadmap.gif

You'll have to cut and paste that URL...clicking on the link won't work.

In short, it appears that nVidia will use 3 different chips to cover low, mid, and high end. Not clear at all if there will be different clock speed variants of each chip. Though it's widely expected that the NV30 will have at least two clock-speed variants.

That's a good question, and I have no idea. Unfortunately, there's too much unknown about the DeltaChrome parts for me to take them into consideration at this time. Good point though....it is advertised as an 8-pipe card, and also advertised as a notebook solution.

I'll try and explain my reasoning...

Right now, the 9700 Pro offers say a 15% performance advantage over the 9700 non-pro. Profit margins on both parts (board level products) are probably similar, maybe slightly higher than the 9700 non-pro at this time.

Let's say the R-350 offers a 30% performance advantage over Radeon 9700 Pro. (This would assume 400/400 clocks, and slight increase in efficiency vs. the R-300, which I think is a reasonable guess.)

Let's assume the following street prices this spring:

Radeon 9700: $200
Radeon 9700 Pro: $250
Radeon "9900": $325

I just don't see many 9700 Pro sales in that line-up. Enthusiasts would probably be willing to spend the extra $75 for the 9900 over the 9700 Pro for the notable performance increase. More budget concsious would not spend the extra $50 for the 9700 Pro over the non-Pro for the "marginal" performance increase. OEMs would be forced to lower the 9700 Pro prices to move them, which would (my speculation) probably cut into profit margins too much.

Solution? Discontinue the 9700 Pro. (Just sell current left-over stock). Let the 9700 non-pro sell at $200 in relatively high volume. Let the 9900 sell in the $300-$325 market for low volume, higher margins.

In a nut-shell, I don't like the idea of a $250 product sandwiched between a better performing one, and a lower cost one. (I use the relative "unpopularity" of the GeForce4 Ti 4400 as the basis for my opinion.) And that's where a 9700 Pro would have to sit, IMO, for it to be reasonably profitable.

 

I think the 9700 non-pro will also be discontinued rather quickly. It's the 9500's that will last longer.

 

Very possible, if ATI releases a "low end" R-350 variant.

And while I see the 9500 Pro sticking around for quite a while, I don't see the same for the 9500 non-pro...at least in its current incarnation of using an R-300 chip with 4 pipes disabled.

 

The 9500 non-pro should be a truly low-cost part. Here's to the low-cost NV3x being similar.

 

Good posts here and I admire the reasoning behind them. I'd like to suggest that one more factor that's worthy of consideration regarding NV30 pricing is the cost of the memory being used. Using history as a guideline it may be acceptable to speculate that nVidia can't count on dramatic declines in that cost. Even leaving yields and board costs aside the dollar value of such high performing memory will enforce the "stubborness" of the break even point for the NV30 from having all that much give.

This part could become a legacy sore point for nVidia and they might see it as worth their while to fudge its true costs. There are many pages from the movie industry they could take for that endeavor and we may never get a good sense for what financial impact this card had.

 

Good point Babel-17. DDR-2 in production quantity at those speeds is approximately 50% more costly than DDR-1. I have a pretty accurate idea of the BOM cost for a 256MB NV30 board, and the memory makes up the vast majority of the cost. In any case, my guess is that a 256MB DDR-2 board cannot have an MSRP lower than $400. In all likelihood, it's probably going to be US$450-500. Until DDR-2 prices come down substantially, this is going to be a huge factor in the MSRP for any 256MB DDR-2 card from any manufacturer.

 

If DDR-II is that expensive, then might it be possible that nVidia's mid-range and especially low-range chips will use DDR-I memory?

 

I'd say it's about 99.9% certainty that nVidia's "value" line ($100) will not use DDR-II. (Just as I'm about 99.9% that ATI's "value" line will not use a 256 bit bus.)

I'd say mid-range (Sub $200) is up in the air, but I'd guess that will also be DDR-I. IMO, The $200 market will likely be seing DDR speeds at about 300-350 MHz max by spring time. (Currently, the mid-rage is at 275-300 Mhz.)

One thing nVidia might do, is try and push a 64 MB DDR-II based card in the mid-range, along side of 128 MB DDR-I versions in a similar price bracket. Sort of like what they did with the 4200. (Some cards have 128 MB of 'slower' memory, and other cards have 64 MB of 'faster' memory.)

 

Timely info. 8)

http://www.digitimes.com/NewsShow/Article.asp?datePublish=2002/12/26&pages=05&seq=21

http://www.sapphiretech.com/VGA/9100.asp

http://www.ati.com/technology/hardware/radeon9100/index.html

Not much info at the ATI site yet...seems to kinda be in "pseudo launch" mode.

Anyway, it does appear to be in fact the R-200 core "plus FullStream" support. Though I expect Fullstream support is a driver level addition,and the chip itself is no different than the R-200 chip on the Radeon 8500.

EDIT:

Note that the clock speed (according to Digitimes) is the same as the Radeon LE variant, (250) not the "retail" Radeon 8500 (275). Here is an example of ATI "keeping the lower performance" model alive, and getting rid of the higher performance model. (Much like I'm speculating they will get rid of the 9700 Pro, and keep the 9700 non-pro around for a little while...)

 

Its possible that the lower speed requirements would allow them to:
1) use a slightly different process that would increase the yield
2) reduce test time by removing burn-in or cold and/or hot testing.

Both these would decrease their cost of the part.

 

Hmm....

My thoughts were that "9700 Pro" chips would be produced anyway, and a greater percentage of them over time...if significant costs could be saved by cutting down on testing and/or cheaper manufacturing (more info would be appreciated, I hadn't considered that would factor in for a complex chip at all and some education would be nice) I guess it does make more sense that the "Pro" would be dropped as the R350 price squeezed downwards.

 

From my understanding of the ramp/qualification process, you do a 'split' lot characterization of your chip on the process: hot, cold (how it performs at hot and cold temperatures), fast and slow (which are process variations which change the gate width--I think). This makes a spreadsheet of data with hot, cold, fast, slow in the 4 corners. The data in the spreadsheet is maximum frequency (or something like that).

From this chart, you can determine the maximum speed that you don't need to test at elevated temperatures, or if you need to adjust the process to obtain that.

Testing at elevated temperatures is (not suprisingly) expensive. And yes, testing can be relatively expensive--remember each chip has to go through this same test.

Burn in is another expensive test. If you can tweak the process to reduce infant mortality at the expense of max operating speed, you can cut another large expense out.

I'm sure I've misrepresented some of the facts, as I'm only repeating what i've heard in meetings and picked up in conversations.

Oh, and an interesting tid-bit: I went to compusa today and the 9500 pro is no longer given shelf space. (Its on their website still, though).

 

Actually, I had thought that only a few chips out of each batch were tested (Heard this from a teacher who apparently did this exact testing, at Intel if I recall). This would be a good reason why chips are overclockable (particularly CPU's), and why every once in a while a bad chip is sold.