Originally posted by: keysplayr2003
Originally posted by: Snooper
When you are trying to build an Uber part, some times you have to use components that are NOT wildely available. With the price of this thing, it IS a limited video card. Heck, the "performance" video card market is a tiny part of the whole video card market anyway. Something like the 512 is a tiny part of that tiny part. Which means it IS limited, one way or the other.
And blaming Samsung for not making enough 1.1 parts is silly as well. These are NOT hot dogs! You can't just set up a line that makes them and then drop them in the package and ship them to your customers! The 1.1ns parts are going to be a VERY small yield out of all their memory chips produced. Only a very small number of these things will pass all the sort and reliability tests at that speed. Most are going to be slower. Some, a lot slower (and end up in the trash at the end of the day). There are things they can do to increase the yeild of high performance parts (things like ... Uh. Sorry. Trade secrets.), but these steps always increase the cost, increase the time it takes to make the parts overall and often has negative impacts on yields. You tweak parameters during manufacturing to increase speed and it tends to push a lot of parts over the edge. Instead of being fast, they simply don't work.
As process improvements are made, the availability of these 1.1ns parts WILL increase. The price WILL come down. And then someone will start using .9ns parts instead for their top of the line card... And we start all over with this debate at the .9ns level.
So memory chips are not made on an assembly line then? Are you suggesting each one is made by hand? C'mon man, a HUGE A$$ company like Samsung? How else are these chips made? (Try not to give up too many of those trade secrets now.)
I would agree that chips are binned according to speed capabilites and stability at that speed. Just don't know why some of you think that it is not plausible for Samsung to be the culprit just as easily as nvidia could be. ::::shrugs::::
Grab your ears and give them a quick yank. Because you sure did miss the WHOLE point! The key there was "and drop them in a package and ship them to your customers".
When they build that hot dog, they have quality control (I hope) for the ingredients going in and for the (few) process steps required to assemble the dog (grinding, mixing, packing, cooking, cooling, etc.). Then the dogs head to packaging. They DO have a sampling scheme that is used to pull a dog out (1 out of a 100? 1 out of a 1000? 10000?) and verify it is up to specs (length, diameter, taste?, maybe a few others). The rest ARE NOT TESTED against performance specs! They do not sample each and every dog to make sure it has a certain taste. If it is long enough and weighs enough, it gets packaged and shipped.
Yes, IC ARE built on an assembly "line" of sorts. One of the most complex operations known to man, actually. The problem is that due to the extreme precision required at each and every operation (ok, manufacturing operation. The analytical operations to not impact the performance of THAT wafer, just the next ones up the line if it misses something!), there is no way that we can insure that each and every die on a wafer (and you can get anywere from a houndred or so CPU or GPU die on a 300mm wafer up to several thousand logic die per wafer!) gets exactly the construction.
The tollerances are just too small. Your build oxide layers that are only a few houndred atoms thick! The widths are measured in nanometers and you have to be able to align multiple process steps right on top of each other to build these transistors. Then we have to wire those transistors up into a functioning device with multiple metal and insulation layers tying it all together.
What you end up with a bunch of parts that all APPEAR to be absolutely identical. But they are not. Not were it counts anyway. When you start pushing things (faster clocks, closer timings, etc), they seperate out: functional failures (they partially or completely do not work), speed failures (everything actually works, but part or all of it the IC is too slow to sale. Trash), low speed (you can sell it in your value line), normal speed (the bulk of your parts), high speed (maybe 10% or so if you are luck. Goes into your high performance parts) and if you are lucky, maybe a very few very high speed parts (one, maybe two per wafer. If you are lucky).
Demand and company policy dictates what happens to that last group. Often, they are just rolled in the high speed parts and sold with them. Even at that "high speed", they are effectively underclocked. Companies tend to do this when they are not getting enough of the very high speed parts to create a new, high speed bin part.
That is often why you will see OCing getting better and better just before the company rolls out a new speed bin. The process improvements they have been making have made ALL the die perform better, so they have more of the "very high speed" parts off each wafer. Which means your average Joe has a much better chance of getting a part that will run much faster while remaining stable. Eventually, these become the new "high speed" parts and the new, faster parts end up in the redefinined "very high speed" bin.
The other thing companies do with these "very high speed" parts is set up a boutique line. Intel did it with the Extreme Edition parts. And it looks like Samsung is doing it with the 1.1ns memory parts. Availability will ALWAYS be low untiil sufficent process and part design improvements have been made to push all the part speeds up a notch.
Then it will start all over again with the 1.0ns or .9ns parts...
So, when you get right down to it, blaming EITHER Samsung OR NVidea is silly. This is NVidia's absolute top dollar card. They would sell as many of them as they possibly COULD at that high price. But they may or may not have enough very high speed G70s to fill the boards. And Samsung is getting top dollar for the 1.1ns memory. They would LOVE to sell as many of these parts at this premium as they possibly can! And I will promise you one thing: People at Samsung, NVidia, and TSMC are WORKING THERE BUTTS OFF right now trying to figure out how to make these parts faster. That is how the business works.
For those of you who didn't (or can't) understand the above, here is the "bottom line": It is NEITHER NVidia's NOR Samsungs "Fault". Neither of these companies are TRYING to hold back sales of the GTX7800-512. They are doing everything they can to make as many as they can. So chill out and quit your whining. Eventually, supply WILL catch up with demand and the price will start dropping (causing increased demand, amazing enough!). Until then, keep looking and maybe you will be lucky enough to get your hands on one and be an early adopter.
Facebook
Twitter