- Joined
- Feb 18, 2002
Story so far:
Athlon FX a.k.a.Thorton, unlike T-Bred, will have "Barton grade processor modules without extra L2 cache."
Graphic67's comments about there physically not being extra cache on the Thorton were pointed out to Adrian Wong, the guy who broke the news that Thorton's going to be called Athlon FX:
I asked him if he still stands by his opinion that Thortons will have 512 cache but that half of it will be disabled? He said yes:
http://forums.rojakpot.com/viewtopic.php?p=32063#32063
"Hello c627627,
A popular misconception of microprocessor manufacturing is that it is always cheaper to create a new line for different variants of the same processor, in order to squeeze in more chips per platter. Well, that's true, to some extent.
However, the bulk of the cost of manufacturing a processor does not come from the wafer. The bulk of the cost comes from the machinery used to make the processors and the cost of running them! The wafer is chicken feed by comparison. That's something many fail to take into account.
The cost of running a fab and the machinery is so high than the cost of manufacturing a Barton and a Thoroughbred is really not that much different. The Barton die is larger so it will be a little more expensive but not as much as people perceive.
Look at what you are paying for processors. Do you think it costs AMD more to make 2GHz Athlon XPs compared to 1.7GHz Athlon XPs? No... Cause it's really the same chip! The value of microprocessor is always about perception, never about the actual cost of the chip.
The same reason is why it makes more sense for AMD to make one single type of die (the Barton die) and use them to produce TWO different kinds of processors - the Barton and the Thorton. Because they are essentially the same processor, this gives AMD a lot of flexibility.
They don't have to worry so much about reserving capacity at their fabs. They can just churn out all the Bartons they want! And then they can package the Barton dies as Barton processors or Thorton processors according to the market forces.
Again, the cost of making a processor is NOT as expensive as you might expect. The bulk of the cost is in building and running the fabs. In short, the first chip costs millions to make. The rest are essentially free.
Now, regarding cache failures... Let's not forget the cache is ON-DIE. It's not a separate cache chip or die that is integrated within the processor package. Cache takes up a lot of space. That's why the Barton die is bigger than the Thoroughbred's die.
When a die is bigger, the yield naturally drops even though the defect density for the fab remains constant. The yield drops because of the larger die size. The larger the die, the more likely it is for contaminants to damage it. And due to its size, the Barton's large on-die cache is more susceptible to getting damaged by such contaminants.
As a new core with a larger die, the Barton will naturally have a lower yield than the Thoroughbred. Many of these will be due to damage to the L2 cache. So, what would be the logical thing for AMD to do?
Throw away the chips?
Nah... They are smarter than that. All they need to do is disable half the L2 cache (naturally the defective half!) and they will have a bunch of working processors! Of course, this doesn't mean they can get all their defective chips running. But it does allow them to recover a large portion of their "defective" chips.
And don't forget... Selling microprocessors is always about selling perception. If the market requires more Thortons, AMD is not going to say, "Sorry, we don't have any more defective Bartons to turn into Thortons." They will just disable half the cache of good dies to make them into Thortons.
If you think that's a waste of money, well, that's because you perceive the cost of the processor to be extraordinarily high just because it's a working Barton core. Well, no. It's all about marketing and perception.
If AMD gets to sell you one of their new cores as a Barton (at a higher cost), it would make them very happy. But they are just as happy to sell you a Thorton using the same core (or a defective one) at a lower profit to them. At the end of the day, profit is profit.
Hope that helps you some! "
How about that mod now to enable the extra 256 KB of cache that's supposedly there?!
Athlon FX a.k.a.Thorton, unlike T-Bred, will have "Barton grade processor modules without extra L2 cache."
Graphic67's comments about there physically not being extra cache on the Thorton were pointed out to Adrian Wong, the guy who broke the news that Thorton's going to be called Athlon FX:
I asked him if he still stands by his opinion that Thortons will have 512 cache but that half of it will be disabled? He said yes:
http://forums.rojakpot.com/viewtopic.php?p=32063#32063
"Hello c627627,
A popular misconception of microprocessor manufacturing is that it is always cheaper to create a new line for different variants of the same processor, in order to squeeze in more chips per platter. Well, that's true, to some extent.
However, the bulk of the cost of manufacturing a processor does not come from the wafer. The bulk of the cost comes from the machinery used to make the processors and the cost of running them! The wafer is chicken feed by comparison. That's something many fail to take into account.
The cost of running a fab and the machinery is so high than the cost of manufacturing a Barton and a Thoroughbred is really not that much different. The Barton die is larger so it will be a little more expensive but not as much as people perceive.
Look at what you are paying for processors. Do you think it costs AMD more to make 2GHz Athlon XPs compared to 1.7GHz Athlon XPs? No... Cause it's really the same chip! The value of microprocessor is always about perception, never about the actual cost of the chip.
The same reason is why it makes more sense for AMD to make one single type of die (the Barton die) and use them to produce TWO different kinds of processors - the Barton and the Thorton. Because they are essentially the same processor, this gives AMD a lot of flexibility.
They don't have to worry so much about reserving capacity at their fabs. They can just churn out all the Bartons they want! And then they can package the Barton dies as Barton processors or Thorton processors according to the market forces.
Again, the cost of making a processor is NOT as expensive as you might expect. The bulk of the cost is in building and running the fabs. In short, the first chip costs millions to make. The rest are essentially free.
Now, regarding cache failures... Let's not forget the cache is ON-DIE. It's not a separate cache chip or die that is integrated within the processor package. Cache takes up a lot of space. That's why the Barton die is bigger than the Thoroughbred's die.
When a die is bigger, the yield naturally drops even though the defect density for the fab remains constant. The yield drops because of the larger die size. The larger the die, the more likely it is for contaminants to damage it. And due to its size, the Barton's large on-die cache is more susceptible to getting damaged by such contaminants.
As a new core with a larger die, the Barton will naturally have a lower yield than the Thoroughbred. Many of these will be due to damage to the L2 cache. So, what would be the logical thing for AMD to do?
Throw away the chips?
Nah... They are smarter than that. All they need to do is disable half the L2 cache (naturally the defective half!) and they will have a bunch of working processors! Of course, this doesn't mean they can get all their defective chips running. But it does allow them to recover a large portion of their "defective" chips.
And don't forget... Selling microprocessors is always about selling perception. If the market requires more Thortons, AMD is not going to say, "Sorry, we don't have any more defective Bartons to turn into Thortons." They will just disable half the cache of good dies to make them into Thortons.
If you think that's a waste of money, well, that's because you perceive the cost of the processor to be extraordinarily high just because it's a working Barton core. Well, no. It's all about marketing and perception.
If AMD gets to sell you one of their new cores as a Barton (at a higher cost), it would make them very happy. But they are just as happy to sell you a Thorton using the same core (or a defective one) at a lower profit to them. At the end of the day, profit is profit.
Hope that helps you some! "
How about that mod now to enable the extra 256 KB of cache that's supposedly there?!