Re: Timeline speculation
Originally posted by Rocketman
So, to do that you need a beta 10.5.64 and Metrowerks and some other tools, preferably a dual boot utility since it will likely be crash-o-matic.
Why dual boot if you are running beta hardware? What are you going to dual boot into?
Furthermore, the existing OS X will run on these systems with the addition of driver support for new motherboard components. Porting OS X to these new machines won't be intensive like porting OS X from PPC to X86 (and that isn't even that bad)
Lets further assume WWDC 03 (May) Apple releases the development call to software gurus to begin writing code for 64 bit, 4+ processes or threads, anf 2 or more processors plus farms and stacks.
You can write code RIGHT NOW that spawns dozens of threads. This is all you need to do in order to make good use of a quad processor system. The Kernel handles spreading these threads across the cpus. Mach already supports up to 32 processors and it could support more if they wanted it to. As a programmer, you don't code for CPUs (not specifically), you write multi-threaded code. If you average 4 threads on a dual processor system, Mach will 'time share' multiple threads on each cpu auto-magically.
Let us further assume they will be the most expensive chip and supplementary chipsets Apple has used for a decade.
This isn't a safe assumption. The .13 micron 970 is only slightly larger than the .18 micron G4.
.18 micron 7455 - ~106mm^2
.13 micron 970 - ~121mm^2
This means that gross yield for a given wafer will only be slightly less for a 970 (compared to a 7455). That is, one wafer can be cut into almost as many 970s as it could be cut into 7455s.
Now, the real yield per wafer is the number of chips you can make times the percentage of good chips. Motorola has had HORRIBLE yield rates, so even if they make more 7455s per wafer, fewer of them are useable. IBM generally has very good manufacturing processes so their yield should actually be higher per wafer. It is very likely that the 970 will cost LESS than the G4... especially if they have higher volume due to increased Powermac sales, Linux sales, and even embedded sales. Remember, this could also be going up against the G4 in embedded markets eventually (slightly more heat, but much better performance).
The cost of the support chips will likewise be determined by: die size (number of gates and process size), the yield at required speeds, and R&D. It is very difficult to guess whether or not the new support chips will cost more or less. They may not require significantly more logic than the previous chips, and if they are produced on a smaller process, their yield may actually go up (decreasing cost).
We're talking X-=serve not a desktop target market for about a year or two of the chip's life. It doesn't matter how long you "hold out" for a PowerPC 970, you will not save alot of money, and wait, to get real work done unnecessarily. Get G4's now. Plan to BEGIN shopping for a 970 a year or more after it is announced.
This isn't a good argument for buying a G4 now. There is NO reason why Apple won't introduce desktop 970s immediately if yield is good. Your arguments are not valid. The 970 is only slightly larger than the G4 and % yields are likely to be higher at IBM. Not only that, but just because the new bus will run at 900MHz doesn't mean that it will be significantly more expensive to make the chipset. Intel will be targeting 800MHz bus machines to the upper end of the mainstream market in a few months. If the IBM/Apple bus is in fact a quad pumped bus (4 ticks per clock cycle), it only needs to clock at 225MHz. Making an ASIC run at 225MHz in not a great technological feat!
The initial market will be server farms where the processor and bus speed is hyper-critical. And commands value.
This is Apple's first step into true enterprise space. It will be priced accordingly.
I completely dissagree at this point in time. It looks like IBM's yields are coming along very nicely. I expect that the only way that the xServe get's bumped well be for the desktop will be if IBM simply can't make enough chips for Apple. Considering they will be producing these at the massive Fishkill plant, this isn't likely to be an issue.
