manaman said:
You don't blame your CPU because your mother board fails do you? Think of it more that way. nVidia and ATI make the equivalent of a CPU for the graphics card, someone else makes the board and buys the other components that are used. Buying a cheap graphics card would be the equivalent of buying a cheap motherboard, then blaming your CPU manufacturer when it fails. Its not a perfect comparison, but it's the best I can think of that is easy to follow.
The fact is, the chipsets are never almost never the failure points in either companies product.
Oh and I think that you will find both nVidia and ATI custom chipsets working smoothly together in the Wii and 360. You will also find a Sony and nVidia co-designed GPU in the PS3.
Obviously you had one bad experience that you are going to stick with as the model of all things to come, hell or high water, but that doesn't seem to be the case for the rest of the world. You can't even really generalize the products into more expensive, or higher quality because of how varied they are cost wise, and power wise. You can just know that sometimes one will have an advantage over the other. Power wise nVidia was leading for a long time. ATI has closed in and have value cornered, but that may have all changed in the months since I was last looking at video cards.
You are partially right here. AMD and NVIDIA indeed only design GPUs and then outsource their production to, like, TSMC. After that they design PCB (which is a board to which GPU, memory and cooler are attached) and outsource production of cards, based on it (and PCBs themselves) to the companies.
By the way, all high-end video-cards are designed by AMD and NVIDIA at first and manufactured on the single company's fabs (AFAIK in NVIDIA's case it's Foxconn) at least in the first few months, so you can't blame for malfuntions those who just glue their brand sticker to a card and sell it (ASUS, MSI, Gigabyte, EVGA, BFG, you name it - everyone). Custom designs for high-end cards are allowed much later.
There are a couple of things however.
First, GPU chip basically defines what PCB and resulting card will look like - if the chip is complex and big as hell (like GT200 was some time ago and like GF100 is right now) it'll emit a lot of heat, and you just have to draw that heat off somehow.
Current air cooling technology can't be significantly improved, which means that you can't make air cooler MUCH more effective without raising its price (using expensive materials for radiator, adding heat pipes, using some sofisticated thermo-interface like liquid metal) and/or size (bigger radiator dissipates more heat than small one on the same technology, bigger fan results in better air flow).
There are more advanced cooling solutions, but they are all more expensive and usually less reliable (water, liquid metal, phase transition, thermoelectric cooling etc), so if you have massive chip with high TDP, you've got to either make the board (and cooler) big or tolerate very high temperature. High temperature definitely lowers life expectancy for any card, and any small production defect is more likely to result in permanent damage to such a card.
Oh, and high TDP means high power consumption, so you have to use more expensive components like capacitors and make PCB design more complex. More such components further increase risk of malfunction (high temperatures and all), like leakage (or you must use solid ones which are expensive).
Second, heat isn't everything - there are other factors. If you make a GPU with 512bit memory bus, than you have to provide it with enough memory chips and complex circuitry. It's expensive.
GPU defines video-card design (and thus reliability + price), and with high-end models it's too expensive for most companies to design their own cards even after GPU maker allows to do it (you can see ASUS with it's MARS cards or, earlier, Sapphire with its weird stuff like Radeon 3850x3, but it's done mostly for image of the company).
So yeah, most malfunctions' direct cause is some manufacturing defect, but minor manufacturing problems are unavoidable (you can't yield 100% of product), while poor chip design results in more complex products, which are harder to produce, so there'll be more problems (and higher price).
That's why first GeForce GT200 (GTX280/260) were very unreliable and hugely expensive (I bought GTX260 at launch and it burned out in a week. A guy in the store told me a lot of people returned these cards at the time), while Radeon 4850/4870 were cheap in relatively durable.
Sorry for such a long posting - I just mean that GPU designer is mostly responsible not just for the card's performance, but for its reliability too. Long gone are days when any Chinese sweatshop could produce it's own castrated design and sell cards cheaply under "No Name" brand.