backtop


Print 95 comment(s) - last by Pakman333.. on Mar 25 at 2:28 AM

Declining to follow the price cut plan

DailyTech revealed last week ATI's new pricing strategy to compete with NVIDIA's GTS 250 rebrand, also known as the 9800 GTX+.

The ATI Radeon HD 4850 512MB was to drop to $129, while the Radeon HD 4870 512MB was scheduled to be cut to $149. This was supposed to be accomplished primarily through the use of mail-in rebates, which ATI would help offset.

Most of the board partners took up the offer on the 4850. However, due to several factors converging at once, the 4850 can now be picked up for around $120 at several e-tailers, albeit with the mail-in rebate.

The story with the 4870 is something else entirely. The Radeon HD 4870 512MB was supposed to drop to $149 with a mail-in rebate, but some of ATI's graphics card partners are resisting this as they feel that since the card outperforms the GTX 260, it should compete against that card. They instead are positioning the 1GB version of the 4850 against the 1GB version of the GTS 250 at the $149 price point.

One of our sources in Taiwan told us: "The ATI lineup is very strong, and we feel the 4850 should go against the GTS 250 and the 4870 against the GTX 260".

ATI has been very aggressive with its pricing, with lower prices and higher performance in the same segments as its nemesis NVIDIA. This has led to declining revenues for board manufacturers, already hard hit by lower demand due to the global recession.


 

 

GTX 280

ATI Radeon 4870

GTX 260 Core 216

ATI Radeon HD 4850

GTS 250

Stream Processors

240

800

216

800

128

Texture Address / Filtering

80 / 80

40

72/72

40

64 / 64

ROPs

32

16

28

16

16

Core Clock

602MHz

750MHz

576MHz

625MHz

738MHz

Memory Clock

1107MHz

900MHz GDDR5 (3600MHz eff)

999MHz

993MHz GDDR3 (1986MHz eff)

1100MHz

Memory Bus Width

512-bit

256-bit

448-bit

256-bit

256-bit

Frame Buffer

1GB

512MB

896MB

512MB

512MB

Transistor Count

1.4B

956M

1.4B

956M

754M

Price Point

$349

$149

$199

$129

$129



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

RE: Huh?
By bankerdude on 3/11/2009 1:23:06 PM , Rating: 2
I think there are two different points being made in the article. 1. The card partners are resisting the opportunity to sell the product with a rebate (subsidized by ATI), and 2. The board manufacturers are hurt when they produce and sell a lower margin ATI board versus a higher margin Nvidia board.


RE: Huh?
By Operandi on 3/11/2009 2:05:35 PM , Rating: 4
Just because the nVidia board is more expensive doesn't mean it has higher profit margins for the board manufactures.

Someone can correct me if I'm wrong but I believe the AMD (ATI, whatever) GPUs are smaller both in manufacturing process as well as physical transistor count which would make them cheaper to produce than nVidia's GPUs (260 on up at least).


RE: Huh?
By afkrotch on 3/11/09, Rating: 0
RE: Huh?
By GodisanAtheist on 3/11/2009 3:58:03 PM , Rating: 4
quote:
Just because the actual gpu is larger, doesn't mean it costs more. If ATI's smaller gpu only had 50% yield and Nvidia's larger gpu had 100% yield, I can guarantee that Nvidia's would be cheaper to product.


-This is true, but given the smaller size and reduced complexity of the ATI GPU (if sheer transistor counts = complexity) and the fact that both GPU's are being produced at TSMC, I'd give the nod to ATI on this one.

All things being equal though, it doesn't look like either company is having any serious troubles as far as we know, although ATI can get ~2x the number of cores per wafer (given relative die sizes).


RE: Huh?
By afkrotch on 3/11/09, Rating: 0
RE: Huh?
By Targon on 3/11/2009 9:39:01 PM , Rating: 3
You have to look at it from the overall company perspective, not just how well the products are selling.

If AMD is able to produce an excessive number of 4870 GPUs, in order to avoid excess inventory, they need to sell more. Considering that AMD as a whole has been having a really rough time due to the economy, and until the Phenom 2, having parts that were not terribly competitive to the Core 2 series of processors, AMD needs money.

So, get more 4870 based cards sold. Even if the profit per card is less, the higher volume would more than make up for the lost profits per unit. It is sort of like selling 100 units at $10 each, or sell 1000 units at $9 each. AMD has not been in the position to sell products for top dollar, so their entire business plan revolves around selling more and more products.

Then, you also have to figure there is brand loyalty that builds up over time. If you have been buying a Geforce card for the past four generations, you will be more inclined to stick with NVIDIA. If AMD can get more people to buy a Radeon, and can stay competitive, those people will buy Radeon products going forward if they are happy with the product. So, AMD is thinking ahead on this one.

The whole idea of rebates is really annoying in general, because of all the complaints about rebate processing. You have the receipt, you have the product, with the box, you fax or send it over, and still end up without the promised rebate money. AMD could drop the price by 10 percent across the board and it would have the same overall effect that the rebates would have, without giving customers the headache.


RE: Huh?
By SiliconDoc on 3/13/09, Rating: -1
RE: Huh?
By GodisanAtheist on 3/13/2009 2:01:24 PM , Rating: 2
Because the corporate heads saw a hole in their line-up and realized they could make more money selling crippled 4870 cores in a lower bracket than they could sell fully functional cores at full price?

Just because its the same die with fewer SPs doesn't mean it was a yield problem, they could intentionally be crippling cores. If I'm not mistaken that's exactly what AMD was doing with the Phenom II x4's as x3's a little while ago (not even burning off parts of the chip there).

All ATI R7xx cards are actually made with 900 SPs, but only 800 are ever activated when the card is shipped out. The processor is built with huge amounts of redundancy, so it ultimately doesn't matter if they get a core with 890 sp's or 810 sp's, cause only 800 are ever actually going to be activated. As you can imagine this works wonders for yields...


RE: Huh?
By Etsp on 3/11/2009 5:57:48 PM , Rating: 2
If all other variables are equal, yes, a larger core costs more.

Let's say we have two platters that are equal in cost, and are similar in every way. On one of them we produce the larger nvidia chip, on the other we produce the smaller ATI chip. Which platter is cheaper? Neither. Their costs are about the same. But, since the ATI chip is smaller, we can fit more of them on the platter, so we have more of the ATI chip than we do of the nvidia chip.

So, for the same cost, we have made more ATI chips than nvidia chips.

On top of that, we have the fact that the ATI chips will have less wasted platter space because of yields. If these two platters have a similar number of defects, it stands to reason that the two platters would have a similar number of defective chips. So, the nvidia and ATI platters have a similar number of defective chips, but the ATI platter has more chips in total. This means that the platter with the smaller ATI chips on it has better yields

(Of course, GPU's and CPU's usually have some redundant circuits to help improve yields, but lets ignore that for the sake of having less variables in this example...)

So, if all else is equal, the smaller ATI chip is cheaper to produce, AND has better yields.

The ATI RV770 has 956,000,000 Transistors.
The nvidia GT200 has 1,400,000,000 Transistors. The ATI chip is about 2/3 the size of the nvidia chip, so ATI produces about 30% more chips per platter. This difference is HUGE.


RE: Huh?
By Alexstarfire on 3/12/2009 6:42:10 PM , Rating: 2
Well no one is denying what you said. The problem is that in real life it's not like what you said at all. A straight 30% better yield is great, but if it takes 2 chips to compete with one nVidia chip then nVidia still wins. It would actually make the nVidia chip about 27% smaller if 2 of the ATI chips are needed. BTW, even though it's 30% smaller the yield difference could be bigger or smaller depending on wafer dimensions.

I'm not sure if you were talking about the GTX260 vs the HD4870 or GTX280 vs HD4870 though. I'm talking more about the latter.


RE: Huh?
By SiliconDoc on 3/13/09, Rating: -1
RE: Huh?
By Etsp on 3/14/2009 12:14:10 AM , Rating: 3
ATI is in a good position for the near future. You make references to a past fiscal year, before this core was released, and presume that those numbers imply the success, and/or profitability of this chip? Talk about foolishness...

Not to mention the fact that nothing in my entire post was even closely related to any rumor, or comment made by ATI/AMD, or even any press release, advertisement or any other form of marketing for that matter. It was conjecture based on the analysis of how silicon chips are made, and the given transistor counts of these cores. My entire post made no mention of profits, which was the focus of your attacks. Only costs.

I made no mention of the price point at which these cores are being sold because it was irrelevant to my statement. My whole point was that a smaller core is cheaper to produce when all other factors are equal.
You seem to think that this statement makes me an ATI fanboy... This is not the case, had the roles been reversed, and NVIDIA had been the one with the smaller chip, my post would have been almost identical.

On a side note, I must mention your incoherent diarrhea of the keyboard is quite distracting from the points you try to make. Try to get some help with that.


RE: Huh?
By meepstone on 3/12/2009 6:56:11 PM , Rating: 2
why bother bringing up an argument on pure speculation, rather pointless.


RE: Huh?
By glitchc on 3/18/2009 10:55:15 AM , Rating: 2
quote:
Just because the actual gpu is larger, doesn't mean it costs more.


On the contrary, that is precisely why it will cost more. Given the NRE costs are amortized over the high quantity of chips produced, the Die Yield becomes the primary benchmark of cost.

Take a look at the following formula:

Die Yield = Wafer Yield x (1 + (Defects per unit area x Die Area)/a)^-a

Assuming a is constant, the defects per unit is fixed for a particular process, and is not dependent on the design of the chip. So if both AMD and NVidia are using a 65nm process from TSMC (for example), their defect rate is the same. Given that the wafer yield is also tied to the process, the only variable left is the Die Area, which is why larger dies typically experience lower yields (all other things being equal, of course, as stated above).

quote:
If ATI's smaller gpu only had 50% yield and Nvidia's larger gpu had 100% yield, I can guarantee that Nvidia's would be cheaper to product.


Sure, but if they're both produced on an identical process, it is physically impossible for the larger die to achieve better yields than the smaller one.


"Spreading the rumors, it's very easy because the people who write about Apple want that story, and you can claim its credible because you spoke to someone at Apple." -- Investment guru Jim Cramer














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki