From IBM's chief technology officer's speech at the International Electronics Forum:"Somewhere between 130-nm and 90-nm the whole system fell apart. Things stopped working and nobody seemed to notice." He added, "Scaling is already dead but nobody noticed it had stopped breathing and its lips had turned blue." We are always seeing stories about "the end of Moore's law" and for years these stories have consistently turned out to be untrue. But this seems a little more specific and a lot more believable.
In a possibly related story, Intel has recently announced a complete change in their future processor roadmap, dropping their massive, and monolithic, P4 flagship in favor of a more energy efficient dual core design.
So perhaps CPUs have hit the wall in some sense. But does it really matter? My amateur understanding is that we will still continue to see total system performance increase, but more and more those increases will come from other links in the chain (from mass storage speed increases, from bus speed increases, etc...) as well as from redesigning toward parallelism.
In other words, while we might not see 6 ghz processors, we will for sure see dual core 3 ghz processors (very soon,) and for most applications this will amount to the same thing. So expect more breathless "Moore's law is invalidated!" stories (even though it's not really a law and can't be invalidated,) but don't get too worked up. There is plenty of room still for innovation.
There are a number of things that happen at higher speeds/smaller geometries. The power density thing that Intel is facing is something huge. Extrapolating from energy density trends, going from "hot plate" to "surface of the sun" happens real fast.
Here's some speculation on things engineers will do ...
- Much faster memory access. This really isn't so much a microprocessor chip problem as a microprocessor system problem. Caches, predictive execution, multi-core processors, etc., etc., are increasing complex bandaids. It took a while to get from 32 to 64 bit. How long until data busses are 128?
- Task specific acceleration. The GPUs, the SIMD instructions in Intel (MMX --> SSE2), etc. are approaches using hardware to accelerate very specific, yet very common compute tasks.
|
In a possibly related story, Intel has recently announced a complete change in their future processor roadmap, dropping their massive, and monolithic, P4 flagship in favor of a more energy efficient dual core design.
So perhaps CPUs have hit the wall in some sense. But does it really matter? My amateur understanding is that we will still continue to see total system performance increase, but more and more those increases will come from other links in the chain (from mass storage speed increases, from bus speed increases, etc...) as well as from redesigning toward parallelism.
In other words, while we might not see 6 ghz processors, we will for sure see dual core 3 ghz processors (very soon,) and for most applications this will amount to the same thing. So expect more breathless "Moore's law is invalidated!" stories (even though it's not really a law and can't be invalidated,) but don't get too worked up. There is plenty of room still for innovation.
- jim 5-10-2004 7:10 pm
There are a number of things that happen at higher speeds/smaller geometries. The power density thing that Intel is facing is something huge. Extrapolating from energy density trends, going from "hot plate" to "surface of the sun" happens real fast.
Here's some speculation on things engineers will do ...
- mark 5-11-2004 6:37 am