GT300 to have an NVIO chip

Expensive is as expensive does

IT LOOKS LIKE the first GT300 chip out the door is not only big, but it will be expensive too. GT300 is going to have an NVIO chip like some of it’s predecessors.

If you recall, GT300 is a 530mm^2 or so chip that taped out a few weeks ago, and the family tree has five branches so far. Our moles on the great Taiwanese taiga tell us that only the biggest one will have an NVIO chip. This is a smaller secondary chip on the main GT300 package to provide the video out capabilities that Nvidia was not able to fit on the main die. This is consistent with the rumors about a ‘dedicated’ GPGPU SKU, Nvidia will likely make one without the NVIO and call it a ‘new chip’.

The problem is that as things currently go, the GT300 looks to about equal a Cypress in performance, but it is about 50% larger. If you assume Cypress is in the 350mm^2 range, a little less than 2x Juniper’s 180mm^2, that means at a bare minimum, Nvidia will be at a pretty hefty cost/performance deficit when it finally gets silicon out of the door.

How hefty? If you consider that Cypress is about 19mm on a side, GT200 24mm, GT200b 22mm, and GT300 comes in at 23mm, you have about 149, 89, 113 and 97 potential die candidates per 300mm wafer according to this calculator. The latest information I have is that a 40nm wafer from TSMC costs a bit over $5000, and a 55nm wafer runs nearly $4000. Lacking information on current 65nm wafer costs, we will guesstimate it will cost about $3000. With those numbers, we can make the following table of silicon costs, rounded to the nearest dollar.

GPU cost spreadsheet

One of these is different…….

Even if you ignore the drop in yields brought on by larger die sizes, Nvidia is at a 50% cost disadvantage. If you add in an NVIO – we are told it is about a $10 cost adder for silicon and packaging penalties – AMD can make a Cypress for about half the cost of a GT300.

Assuming that Nvidia can make GT300 perform roughly equally to Cypress, at this point a BIG assumption, then ATI can price its part at a profit while driving Nvidia into a loss. For derivative parts, the same holds true, but margins will be a bit tighter for ATI because the smaller four SKUs from NV lack the NVIO.

The fanbois are sure to say something stupid about Nvidia having better performance, but it would need to be twice as fast to make up for the cost deficit. Based on information we have about the architecture, that is not going to happen, no chance. Keep in mind these people are the ones who claim that GT300 taped out months ago, and is smaller than GT200b, and…. and…. and Nvidia must be sitting on the warehouse full of parts for some conspiracy laden reason. Heh heh, no cookie, tell mommy I said so.

There are additional factors that will drive up costs, but even without them, it is hard to see how Nvidia is going to make money on GT300. At the bare minimum, ATI can price the green team into the red while upping its own margins. Nvidia may have some spin about encoding for Ipods at 23x the speed of thought or some other pablum, but with GT300, it will lose on price. Badly. Again.S|A


The following two tabs change content below.

Charlie Demerjian

Roving engine of chaos and snide remarks at SemiAccurate
Charlie Demerjian is the founder of Stone Arch Networking Services and is a technology news site; addressing hardware design, software selection, customization, securing and maintenance, with over one million views per month. He is a technologist and analyst specializing in semiconductors, system and network architecture. As head writer of, he regularly advises writers, analysts, and industry executives on technical matters and long lead industry trends. Charlie is also a council member with Gerson Lehman Group. FullyAccurate