What a difference a “4” makes. The NDA on Nvidia’s new midrange GTX 460 (GF104) cards was lifted last night to a general reception of “right performance at the right price”. No doubt, after reading over many different reviews this card appears to be the real deal at the $200 price point (for now). Despite the glowing reviews however one nagging question remained (well other than the ‘odd’ lack of any tesselation bruhaha); considering that GF104 has an estimated die size of 320mm2 (which is nearly 96% of AMD’s high end Cypress part at 334mm2), can Nvidia actually make money on this part, or is it simply designed to put pressure on the market share laceration that’s been bleeding out lately.
It should be noted that the numbers illustrated below are mostly educated guesses, as AIB’s are not generally forthcoming with actual component pricing, and as such should be taken with appropriate spoonfuls of salt. The point of all this is to see how much wiggle room Nvidia has left itself by pricing these new GF104 parts so competitively.
Let’s start by working through this problem backwards from finished retail part to bare silicon. When you go to purchase a new video card, you generally do so at a retail, or e-tail shop such as Newegg. These places need to make money on the parts they sell (otherwise what is the point?) so the price you see advertised has been marked up some from the price that they bought it from the AIB (Add-In-Board) partners (such as Sapphire, eVGA, XFX, etc). Markups will vary depending on many factors, but retailers will generally add between 5 and 10% onto what they pay for the final sales price to end users. A look at Newegg’s financial statement from June 2009 shows an average markup of 10.9% (Cost of Sales / Net sales). For the purpose of this article we’ll be generous (to Nvidia) and go with the 5% markup figure.
AIB’s will markup their products when they sell to retail outfits by another 10-15% or so (call it 10% here), but here is where things start to get interesting. These board partners are the ones who compile all the various components into a working video card, and there are some considerable expenses incurred in this process. This is a business after all and if they cannot get all the parts they need to build cards that sell for a profit, they look to other suppliers that can offer profitable parts, or they go out of business. Knowing this, ATI and nVidia will adjust the pricing of GPU silicon on their end as much as possible to buoy their board partners, even if it means taking a loss on certain products to do so.
For the sake of brevity, we’ll simplify the part roster for a graphics card like the GTX 460 and break it down as such: GPU, PCB, RAM, heatsink, misc other passive components, assembly & testing, and packaging & accessories.
So we’ve got an estimated total of $71 to get the board put together, boxed up and ready to put on retail shelves. Let’s put all these numbers together now and see how much breathing room Nvidia has left itself for this part. Our AIB & retail markups of 15% come out to around $33, add in that $71 from above, and we’ve got $104, meaning that Nvidia is potentially selling each 1GB GF104 chip for about $116. Not too shabby, but whether or not this is a good number depends largely on yields of GF104. Without exact yield numbers to examine, instead, here is a “what-if” chart detailing approximate costs per die at given yield rates.
With the numbers thrown out there so far we can see that as long as Nvidia can get yields above 40% on these new parts they will at least be in the black, but remember that their profit margin must cover research and development, administrative costs, and marketing of the chip as well which typically ring up in the hundreds of millions of dollar range. As some food for thought, at 80% yield (more or less on par with ATI at this point) Nvidia would be banking about $77 per GPU sold. At that rate, they’d have to sell 1.29 million of these cards to recoup 100M of R&D investment. I’m not saying it’s impossible, but that’s a lot of video cards to move.
Without any solid numbers of actual yield on GF104 at this point it’s hard to draw any conclusions, except that this part is not going to make Nvidia a truckload of money, even if they are able to get near perfect yields. However, unlike GF100, the new chip offers Nvidia enough price latitude to put the squeeze on ATI’s midrange parts such as the 5770 and black-sheep 5830. The tempting combination of price and performance offered by the GTX 460 cards makes a powerful case for choosing the green team if you’re in the market for such a video card. The only problem with this line of thinking is that ATI has just as much price latitude as Nvidia when it comes to Cypress based cards (5830,5850,5870), and way more latitude when it comes to Juniper based cards (5750,5770) as the die sizes are all much smaller for the latter pair.
As SemiAccurate stated earlier, no matter how you cut it, or where your gaming allegiances lie, the GTX 460 is a good thing for consumers. We finally get a Fermi derivative card with reasonable temperatures and power draw, and ATI will have to drop prices on their mid-range to remain competitive as we wait for Southern Islands to launch later this year. While GF104 isn’t going to do wonders for Nvidia’s balance sheet any time soon, it does have the potential to slow their descent down the consumer market share ladder, and certainly whets our appetites to see what the lower end GF106 and GF108 have to offer in the future. S|A