Petz,
Good to see Intel fellow John Crawford still downplaying the 64-bit desktop.
When we compare the transition from 8086/286 -> 386 and 386 -> AMD64, there are basically 2 issues + some wild cars, the way I see it.
Issue #1 is the old segment - offset addressing. This was a real PITA, it introduced a lot of limitations, and thelimitations were real, visible, even to the end users in some instances, since the segment was only 64K. There are some paralles in 386 -> AMD64 transition, but the issue is not as visible and not as painful yet, at least it is not as broadly felt. It is a limit to increasing number of server apps, high end workstation apps and various content creations apps.
In a few instances, companies invest resources in the workarounds, such as PAE, but this will never catch on, since the writing is on the wall for 32 bit apps. The 64 bit processors are already in the marketplace, and it makes more sense to invest the money there, rather than in losing battle of 32 bit processors.
On the consumer side, where the bulk of the processors go, the issue is far less potent now than it was during the last transition. There are not that many consumer apps that are limited by the 2 or 4 GB limit of code and data segments.
Issue #2 is total memory addressing. In case of 8086, it was 1 MB, in case of 286, it was 16MB in protected mode (I believe). Back then, during the last transition, the memory prices were extremely high, and the 16 MB limit was more of a question of $$$s then bytes. The memory prices were between $500 and $1000, so one would have to spend $8,000 to $16,000 just on memory to reach this limit.
Back then, whoever needed that much memory was running a "real" machine (various minis, mainframes, high end propriatery workstations), not the "toy" x86 machines.
Even if there was a way to address more memory, there was no OS in sight to support that. So the memory limit was distant and not at all pressing.
The difference in these transitions is that the old limits were 64K per segment, 16Mb total addressable memory vs. 4 GB segment and 4 GB total memory (forgetting for now ways to address more than 4 GB).
Anyway, the 4 GB addressable limit is now more pressing then it was during the last trasition. There are 3 reasons, IMO:
1) memory is cheap, one can buy the limit of 4Gb for $600.
2) OS's able to take advantage of it are here (Linux and other flavors of Unix), or more or less here (Windows64)
3) The installed base of other "real", "big-iron" type machines vs. PC based hardware is a fraction of what it was during the last transition, and is dwindling (IBM Power and Sun are still there, who knows for how long). PC based hardware is more or less the only game in town now, and there is an intrinsic demand for more performance (memory addressability) from growing capabilities of existing software.
So on balance, we are approximately where we were when the last successful transition took place, or actually even further (due to more timely availability of OSs).
The wild cards are performance, price/performance and marketing. During the last transition, 286 outran 386 running existing 286 code (there was basically no 386 code), the clock speeds were higher, initially, yet, people still bought 386s to run 286 code, mainly due to marketing, even though the price was higher. Go figure.
AMD64 will not have much of a marketing push behind it (vs. Intel advocating status quo), but if AMD can match or exceed performance of current 32 bit Intel offerings, and successfully claim to throw in 64 bit capability on top of that basically for free, maybe AMD can break through.
I think there are 2 key words that should figure in AMD marketing of AMD64 vs. x86:
DEAD END
Joe |
