ATI Availability at Anand's
- Thread starter Geo
- Start date
http://world.altavista.com/babelfis.../ce/0,2000062982,20095625,00.htm&lp=zt_en
Anyone seen this? I dont know what to believe these days. Anyway Babelfish's translations are the best engrish you'll ever see. However is it mentioning to bridge chip below?
Anyone seen this? I dont know what to believe these days. Anyway Babelfish's translations are the best engrish you'll ever see. However is it mentioning to bridge chip below?
as the man said
Jerry Sanders, said: “Only real men have fabsâ€
well if they had them, more control would be possible, since they don't have them... pfft... and it was only a year or more ago, when the Taiwanese fabs were not running at full capacity, and there were fears of long term overcapacity, IIRC.
Jerry Sanders, said: “Only real men have fabsâ€
well if they had them, more control would be possible, since they don't have them... pfft... and it was only a year or more ago, when the Taiwanese fabs were not running at full capacity, and there were fears of long term overcapacity, IIRC.
DaveBaumann wrote: "modern processes take 3 months to turn around, such that if you have a popular part and want to order more wafer for that part it'll be three months before the extra supply is seen"
Fodder replied: "Can you expand (in layman's terms ) on why it takes quite so long? Or do you just need to book (a long time) in advance?"
20 years ago I worked in a design group in a building with a fab, so I know a little bit about how it used to be done. Even then, it took many weeks to physically move wafers through the fab line, even after the masks were all made, simply because of the number of steps required in fabrication. Today, everything is much more complicated, but let's think about back then.
Consider what is needed to put one layer of metal on the wafer. First you need to cover the whole wafer with a thin layer of metal. The thickness must be precisely controlled, so you can't do it quickly -- maybe you use vapor deposition. Then you need to physically move the boat-loads (that's a technical term) of wafers to another machine where you deposit a "resist" layer. Then another machine where you use the mask to expose the resist. Then another machine where you etch off the part of the resist where you don't want metal. Then another machine where you etch off the metal, except where there is resist. This is another step that must be very precise (and therefore slow), or else you over-etch or under-etch and whole boat-loads of wafers are ruined. Then a step where you wash off the remaining resist.
Multiply that times the total number of layers, which I think these days goes into the dozens. Add in the fact that putting on a metal layer probably requires more steps these days than I described above. And then consider that the boat-loads probably have to wait at all of the machines, since keeping all of the machines working all of the time is more important than speeding the wafers through, at least for production. There are ways to get small quantities more quickly for testing and debug, but that is costly.
Add on top of all of that any scheduling issues at the fab, plus vendor fears of having excess (and potentially valueless) inventory, and it takes even longer. Perhaps the mystery is not that it takes 3 months to turn around new parts, but rather that it can be done that quickly!
Enjoy, Aranfell
PS: Throw in the incredible capital cost of all of these machines, plus the incredible expertise required for the research necessary to keep shrinking the dimensions (plus the capital cost to buy new machines for each new process) and it becomes clear why few companies other than specialists try to run fab lines these days.
Fodder replied: "Can you expand (in layman's terms ) on why it takes quite so long? Or do you just need to book (a long time) in advance?"
20 years ago I worked in a design group in a building with a fab, so I know a little bit about how it used to be done. Even then, it took many weeks to physically move wafers through the fab line, even after the masks were all made, simply because of the number of steps required in fabrication. Today, everything is much more complicated, but let's think about back then.
Consider what is needed to put one layer of metal on the wafer. First you need to cover the whole wafer with a thin layer of metal. The thickness must be precisely controlled, so you can't do it quickly -- maybe you use vapor deposition. Then you need to physically move the boat-loads (that's a technical term) of wafers to another machine where you deposit a "resist" layer. Then another machine where you use the mask to expose the resist. Then another machine where you etch off the part of the resist where you don't want metal. Then another machine where you etch off the metal, except where there is resist. This is another step that must be very precise (and therefore slow), or else you over-etch or under-etch and whole boat-loads of wafers are ruined. Then a step where you wash off the remaining resist.
Multiply that times the total number of layers, which I think these days goes into the dozens. Add in the fact that putting on a metal layer probably requires more steps these days than I described above. And then consider that the boat-loads probably have to wait at all of the machines, since keeping all of the machines working all of the time is more important than speeding the wafers through, at least for production. There are ways to get small quantities more quickly for testing and debug, but that is costly.
Add on top of all of that any scheduling issues at the fab, plus vendor fears of having excess (and potentially valueless) inventory, and it takes even longer. Perhaps the mystery is not that it takes 3 months to turn around new parts, but rather that it can be done that quickly!
Enjoy, Aranfell
PS: Throw in the incredible capital cost of all of these machines, plus the incredible expertise required for the research necessary to keep shrinking the dimensions (plus the capital cost to buy new machines for each new process) and it becomes clear why few companies other than specialists try to run fab lines these days.
You are very welcome. By the way, one old trick is to hold back some wafers before the metal mask stages during debug. That way, if a bug can be fixed with just a metal change, the revision can be made a lot more quickly and cheaply than if fabrication has to start from the beginning. So debug cycles aren't (usually) the full three months. Still, it has always been the case that serious bugs can *really* mess up the schedule.
Enjoy, Aranfell
Enjoy, Aranfell
Thanks from me as well. That one just made something click in my head!
FYI:
Firm delivery dates for X850 cards are showing up on UK web-sites now:
Komplett.co.uk have a confirmed 27/01/05 delivery date:
http://www.komplett.co.uk/k/ki.asp?sku=305921&cks=PRL
OcUK have a firm delivery date of next week (product in transit)
http://www.overclockers.co.uk/acatalog/X850_Series.html
HTH
P69
Firm delivery dates for X850 cards are showing up on UK web-sites now:
Komplett.co.uk have a confirmed 27/01/05 delivery date:
http://www.komplett.co.uk/k/ki.asp?sku=305921&cks=PRL
OcUK have a firm delivery date of next week (product in transit)
http://www.overclockers.co.uk/acatalog/X850_Series.html
HTH
P69
Komplett.se has 39 Powercolor X800 256 Mb cards in stock. Though it's currently priced rather close to the vanilla 6800.
Much cheaper Sapphire X800 256 Mb versions are on their way though, confirmed for 2005-01-27. Same goes for their 256 Mb XL cards which are in fact cheaper then Powercolors 256 Mb X800.
Much cheaper Sapphire X800 256 Mb versions are on their way though, confirmed for 2005-01-27. Same goes for their 256 Mb XL cards which are in fact cheaper then Powercolors 256 Mb X800.
x850 xtpe in Spain arrives where its buyers tomorrow
http://www.forospc.com/forophp/viewtopic.php?t=104242&postdays=0&postorder=asc&start=20
Anand :? :?
The Radeon X850 XT will be available at the end of January with the X850 XT PE available a week later in February.
http://www.forospc.com/forophp/viewtopic.php?t=104242&postdays=0&postorder=asc&start=20
Anand :? :?
The Radeon X850 XT will be available at the end of January with the X850 XT PE available a week later in February.
Too wordy, try:
CJ said:
Bjorn said:ANova said:
Is very easy to find 6800 PCIE's. The 6800 GT/Ultra PCIE versions are another thing though
I seriously doubt that RAM modules are causing the problems. Why would the AGP versions of the 6800's have such good availability then ? (There are lot's of 6800 GT's and even Ultras in stock at komplett.se f.e)
It seems like the native agp cards are easy to find, and the native pci express cards are easy to find, but not the crossovers.(pci-e ati cards are possible to find, the agp ones pretty much aren't)