Why do they sell memory-crippled cards?

[Slaimus]

I do not really see how they save money by halving the memory bits. The GPU itself still need to have the same yield, as errors in the memory controller is not likely. Most manufacturers use the same PCB, but just leave half of the memory chips off, so there is no savings on the PCB side. Memory chips are sold by capacity, so there should not be any savings in using fewer chips either.

So why are there cards like that out there? The 9800pro 128-bit comes to mind as a recent example, but there have been 64-bit version of Radeon 8500, GFFX 5200, and more.

 

[Marsumane]

The chips are sold to them by another party. The price they pay isnt dependant on how much the company pays to make the chips, but rather what the company decides to charge them for it. So they typically have to pay more for a memory chip that doesnt cost as much as the difference that they are paying for them. As far as the pcb's go, I guess it must be a huge deal to provide more traces on the board. Im guessing that its the most complex, costly step in the production of the pcb and that is how money is saved? I'm really not sure on that one.

 

[alent1234]

they probably have some revenue models that tell them how many people will pay how much for a card. say they can produce 10,000,000 9800 level GPU's a year. Currently if you go to a retail store you will pay around $250 for that card. They know a lot of people won't do it, so they sell a bunch of crippled cards for less. They still have to pay a lot of fixed costs to TMSC or IBM whether they produce 10,000,000 chips or 6,000,000. So they try to break even on a small percentage of the crippled cards and make the profits on the higher end stuff.

 

[Keysplayr]

Originally posted by: Slaimus
I do not really see how they save money by halving the memory bits. The GPU itself still need to have the same yield, as errors in the memory controller is not likely. Most manufacturers use the same PCB, but just leave half of the memory chips off, so there is no savings on the PCB side. Memory chips are sold by capacity, so there should not be any savings in using fewer chips either.

So why are there cards like that out there? The 9800pro 128-bit comes to mind as a recent example, but there have been 64-bit version of Radeon 8500, GFFX 5200, and more.

Memory chips are sold by capacity, speed, fabrication process, type. It's a lot more involved than you make it out to be. It is cheaper for the card manufacturer to "halve" the data bus width than to create an entirely new "value" design to appeal to the "value" market. I guess the way they figure it, better to cripple our cards and deliver a certain price range card than to have our competitors snatch up all the money from that market segment.

They are crippled so they can sell at a lower price point. Its all about the bottom line and how much value, mid range, high end market value can they nail down. Its all about money and don't let anyone tell you differently.

 

[Peter]

And of course, with those memory-bandwidth-crippled setup, the very lowest speed bin of GPUs is used - faster ones would be memory bandwidth starved anyway.

Then there are plenty of applications where a certain rendering feature set is required, but where speed is quite irrelevant - and that's where you'd for example rather use a Radeon 9550LE than a 9200.

Finally, there's the growing "low profile" card market. You simply can't make an LP card without halving the memory bandwidth, simply because the reduced card size doesn't let you have RAM chips above the main GPU chip, only to the side of it.