Nvidia is going to report its financial results for the Q4’2008 today, and most of the conference call will be dedicated to explaining the reasons for bad results and finding the way forward. But what is not going to get discussed is the actual reason for these results: since 2005, Nvidia continuously raised the cost of its production by making more and more complex PCBs, coming up to the point that almost all of “enthusiast” (high-end) and “performance” (high-end mainstream) has to be manufactured by the likes of Foxconn and Flextronics, with partners doing nothing but putting stickers on cards manufactured by less-liked mega-companies. It all culuminated with the introduction of GeForce GTX260 and 280 cards, cards built on 12-layer PCBs of enormous complexity.
Every time Nvidia was given the choice of reducing the cost of manufacturing, the company opted to look the other way. With the resurgence of ATI, who chose to optimize its designs to the maximum, Nvidia’s bubble bursted spectacularly and the profit margins were gone. Enter 2009 – when designing the 55nm parts, the company decided to aggressively reduce the cost of manufacturing. First changes came on professional Quadro CX/FX boards, followed by 55nm GTX260 and GTX285. GeForce GTX295 will probably go down in history as the most complexed part that the company ever manufactured, since you can expect that even a dual-PCB GT212-based part will be less complex, even if 512-bit memory controller would be used. Chinese site Expreview.com came up with a story detailing further refinements of the design on GeForce GTX 260 parts, but the same procedures are being implemented across the line, from cheapest to most expensive 55nm and 40nm cards. Under the condition of anonymity, insight of the design of 40nm performance and enthusiast parts was disclosed to me. According to this manufacturing expert close to the company, complete 40nm line is being designed with cost-effectiveness and scale of economics in mind. Number of PCB layers went down even on the GT212 part, and late P8xx and early P9xx designs are a clear testament of changes that are going on in Nvidia’s engineering departments in APAC region and of course, in the US.
Many designs are finalized and are now in the process of refinement, but this should lead to dramatically lower cost of manufacturing, thus giving flexibility in pricing. On the example of upcoming performance part, we were cited 20% cheaper cost of production. 20% is a large number given the fact that Nvidia will implement GDDR5 memory, but this is a normal result – the moment Nvidia started to think with their wallets, instead of ego – improvements were under way. So, we will have GT212 with more complex memory controller than ATI’s RV790 and RV890, yet cheaper to produce due to all the advantages of GDDR5 memory combined with the optimized design of the PCB itself. We’ll see if that will be enough or not.
In a way, Nvidia’s A.D. 2008 was identical to Ferrari’s 2005 – they believed in being unbeatable and got their arses handed to them. We will see will 2009 end for Nvidia like Ferrari’s 2007 or 2008.