Isn't it just getting harder for AMD?!
Yes, in more than one way, it is.
The laws of physics are the same for everyone, everywhere.
Involved in this "dispute", is a trio (for what's relevant): Intel, AMD & IBM. It's just not true that each of these have achieved what they have, all by themselves. A lot of contributors are working in the shadows...
Most of the 'tricks' which make a microarchitecture shine are, more often than not, real technological brilliant jumps; and, of course, engineering is also about economy & profit.
Both Intel & IBM are giants; AMD is about 7 orders of magnitude smaller than Intel, let alone IBM. This factor, implies all the dimensions involved, from real estate to "skunk works".
I'm not going to quote every relevant post in this thread, but there are some points worth mentioning:
a. AMD, together with IBM
et al, have an actual advantage, at the 90nm node. They managed to bring [comparatively] lower power consumption, lower frequencies, higher scalability & higher performance within a single pack: the (aka) K8 core. How? Apart the "skunk works", using IBM's SoI technology, a shorter & wider pipeline (à la RISC), an on-die memory controller and an open standard HyperTransport serial bus.
Will AMD be able to keep its current advantage towards Intel?
b. In the 19th century, James Clerck Maxwell, defined a rule of thumb, known as the 3D scaling rule, which states that, «any lossless circuit, when scaled down in its physical size by a factor of 10 in all 3 dimensions, will reflect a new circuit which is equal to the first, only 10 times faster.» There are no lossless circuits; so, was "NetBurst" a victim of MHz alone? Certainly not. Aside from Intel's marketing MHz hype, it was a fine technological approach to HPC; were it not for the old FSB, the lack of an on-die memory controller and the "mere" use of non-SoI strained silicon, "NetBurst" could be alive & well, even with fewer MHz... It hit the thermal barrier, however. Intel stubbornness? Short-sightedness?
c. IBM broke - again - the patent record number, last year.
Both POWER6 & Cell are to be working above 4GHz, within two years, max. Down with the MHz myth. Will IBM be able to "aid" the next AMD (aka) K8L core climbing up in 'speed'? Will AMD get by with HyperTransport v3.0 & Z-RAM? Will a programmable on-die memory controller be viable? What if the market trends change?
d. "Dry" scanning lithography will go down as far as the [commercial] 32nm node, probably. Double-exposure & Immersion lithography may go as far as the [commercial] 22nm node (these are ready, time-to-market aware technologies, not R&D); after, perhaps EUV & nano-robots (and what else?...). New materials, high-k metals, 3D gates, core-stacking, nanotech/molecular/quantum computing technologies are going to be pervasive, perhaps within the next 10 years. But, the laws of physics are the same for everyone, everywhere.
It doesn't make much sense to have a bunch of highly sophisticated features within a multi-core die, running at a snail's pace, does it?
MHz are here to stay & some of the highly sophisticated features are already implemented.
I believe that, if AMD holds to the MHz/TDP compromise with Intel, the next big fight will be feature's wise... And who gets there first. At this point, Intel has the lead.
Cheers!
Facebook
Twitter
Instagram