The INQ's Microprocessor Forum Report

This year, the fact that there were no new, fresh first-time X86 PC processor announcements for the first time on the event, was overshadowed by quite a few new "other" CPUs that will be of great interest to the computing community.

In his keynote talk, John Crawford, Intel Fellow, talked about the new generation of transistors needed to allow next performance levels for CPUs, and the issues that arise when you reach the molecular or atomic-level semiconductor.

Nanotechnology materials are one of the solutions proposed to allow the miniaturisation required. Other key problems for next generation CPUs are huge design complexity, associated test & debug issues, as well as power and heat dissipation problems that come as you cram more gigahertz and hundreds of millions of transistors in an ever smaller space - according to Crawford, the billion-transistor desktop CPUs will be there by 2007.

A typical problem is that the "logic" portion of CPU, the one that has all the execution units and logic for computation, is much hotter than the area with the on-chip cache memory, for instance - it is a "hot spot".

As a proof or a need for a billion-transistor CPU, he also described a hypothetical ultracomplex future 64-bit Itanium chip with 4 (four!!) Itanium2-like CPU cores on a single chip (120 M transistors total for CPU logic), 16 MB on-chip shared cache (850 M transistors for it), and leaf-like internal bus structure, exceeding slightly the billion-transistor figure. Such an approach also spreads the power "hot spots" more evenly across the chip, following the increasingly important "balance performance and power" approach.

Now the real news: The first 64-bit Macintosh PowerPC CPU, by who else but IBM. The PowerPC 970 technical details were unveiled, and this little fast 64-bit CPU will enable Apple to overtake the X86 PC in the 64-bit migration. Why? Isn't Hammer going to appear a quarter earlier at least?

Yes, but the 64-bit Mac migration strategy is firmly within the decision of one company, Apple, which controls both system hardware and software - while 64-bit PCI migration will involve the tussle and consensus among many more disparate hardware and software players.

The PowerPC 970, or GigaProcessor Lite (Power4 Lite), brings the proven high-speed Power4 64-bit technology to the desktop - together with desktop "ClawHammer" Athlon version, it should be the lowest-power consumption 64-bit general-purpose processor. The CPU can process up to 8 instructions per clock, has 64 KB primary instruction cache, 32 KB primary data cache, and 512 KB secondary cache on chip. It easily supports dual and quad configurations. There is a very powerful dual 128-bit SIMD unit that is supposedly compatible with Apple "Velocity Engine" extensions for Macintosh. It is at a low 42 watts (or 16 watts for the more "mobile" 1.2 GHz version) power consumption level, lower than Xeon or Athlon, for instance. This little thing could make a 64-bit PowerBook - imagine the tomorrow's 64-bit mobile battle: ClawHammer by AMD vs PPC970 by IBM: where's Intel there?

Intel described more details on its recently-discussed future mobile PC processor, codenamed "Banias". This is the first X86 CPU designed from ground-up specifically for low power mobile computing, not a "recycled" desktop CPU version. Its features look like a mix of Pentium-III, Pentium-4 and new unique stuff. But nothing new there vs the Fall IDF.

On the server side, AMD covered more details about its 64-bit Opteron "SledgeHammer" X86-64 processor. For the first time, the benchmark results were out there, including the SPEC2002, where the new CPU achieved very high numbers even in the 32-bit mode using Intel compilers. At 2.0 GHz, the numbers are 1202 int2000 and 1170 FP2000, which is the best integer and one of the best FP figures for any processor. Keep in mind, though, that the shipments are still some six months away. Again, AMD "Hammer" family are the first PC X86 processor extended to the full 64-bit architecture, and compiling the SPEC with 64-bit compiler should improve the benchmark results by another 15-20%, according to AMD. We had a chance to play with a Dual Opteron 1U server running both SuSe Linux 7.3 (64-bit) and Windows Advanced Server 32-bit in a dual-boot fashion. The CPUs ran at just 1.4 GHz at this stage.

Intel showed its updated roadmap, confirming that the next "Madison" IA-64 processor will also be called Itanium2 and will ship early next year.

The monster with a huge 6 MB on-chip cache is expected to reach an unbelievable 477 million transistors, 8 times more than the first Pentium 4. Next year the Pentium 4 and Xeon processors will be updated to reach the 0.09 micron process - the codenames are "Prescott" and "Nocona". The new processors are, among other new features, expected to run at between 3.2 and 4 GHz in the first one year of deployment, and use a sped-up 667 MHz front-side bus (you'll need dual-channel DDR333 memory to feed that one well).

Fujitsu showed the best-ever SPARC chip, the SPARC64 V. This superb processor takes away the Solaris hardware platform lead from Sun - it runs faster (1.35 vs 1.2 GHz Sun USIII for early next year), processes almost twice as much per cycle due to out-of-order execution, wider buses, and twice the number of computation units, and has much larger 2 MB on-chip cache combined with almost triple the outside bus speed. Furthermore, it has mainframe-like features for real-time error recovery even on the internal unit level. Too bad it's only going initially into their large PrimePower servers.

In Part 2, we'll cover the "special" processors - there is even more interesting stuff there...µ