AMD beats Apple to a real 64-bit PC

THE 64-BIT PC running a 64-bit OS finally, officially, became a reality last week. And no, it is not an Apple PowerMac G5, simply because MacOS is not yet available in the 64-bit form. As expected for months, the new kid on the block is AMD's Athlon 64 platform.

There were dozens of Athlon64, Athlon64 FX and Opteron boards on the show around, from almost every mobo vendor except SuperMicro, compared to only one Itanium2 offering - a dual-CPU board from, yes you've guessed it right, SuperMicro. While of course we couldn't find any PowerMac G5 clones - cloning Macs is almost as ethically unacceptable as human cloning - the 64-bit tension was palpable.

There have been many failed attempts to bring 64-bits to the desktop before, but does this latest iteration of the game justify those demanding such a shift?

Failed attempts
The history of efforts and initiatives to jumpstart 64 bits into the mainstream goes back some 12 years to the ACE consortium when several key vendors, including Compaq, Acer, MIPS and Microsoft, among others, gathered to create an alternative, non-Intel, CPU-independent Windows platform to run the then-future Windows NT operating system. At that time, the MIPS R4000, the very first 64-bit mainstream processor appeared, and soon after, Acer launched a nice high-end 64-bit PC based on that CPU.

The arguments among the vendors, as well as Intel's fight-back with its "preferential" vendor treatment, various veiled threats and quick Pentium chip announcement in 1992 helped can the effort - one of those sorry events in the PC history that turned the tide for the worse, it seems. MIPS was, of course, the biggest casualty in the whole mess, ending up submerged by Silicon Graphics - a move which proved wrong for MIPS, and wrong for SGI, in the long term.

Alpha was the next platform to attempt the 64-bit PC launch with its 21164PC series of inexpensive low-power Alpha processors running Windows NT in 1995 to 2001. These systems ran what was probably the most stable NT flavour ever - a proof that at least a part of Micro$oft OS quirks come from the idiocies in the X86 architecture - and were available for as low as $1,500 - a very low price for a stable, fast 64-bit machine with memory and I/O bandwidth matched by X86 PCs only several years later. In a sense, they were the first 64-bit true PCs, because you could boot them under 64-bit Linux and have a fully 64-bit system on your desk, supporting quite a few standard apps and peripherals, for those few bucks you spend on it.

Too bad Microsoft withdrew the NT support for it, and the excellent Alpha architecture itself fell victim to the tripartite Compaq-HP-Intel conspiratory powerplay, making Compaq more edible and leading to the eventual merger. Anschluss, or what?

Software support
The key problem then is still valid now, and of course scratches the hull of Intel's good ship Itanic as well - the lack of native, optimised 64-bit applications to fully make use of the platform and justify extra expense, therefore increasing the user base. And without that user base present or firmly committed, most software vendors are not willing to port their software - a clear chicken-and-egg situation, isn't it?

Could it be that Apple also wants to be ready with enough native 64-bit titles when they finally launch the 64-bit version of MacOS X?

Last century, Alpha had the greatest and clearest chance, with its stable NT support, however disagreements between DEC/Compaq and Micro$oft derailed the widening of software support for the platform.

AMD, which absorbed quite a few leading Alpha engineers, attempted to solve the problem by evolving its successful Athlon to the full 64-bit level, while keeping the native 32-bit application performance and adding Alpha-style superior memory and I/O architecture.

The Athlon64/Opteron platform is exactly that, in a nutshell.

But AMD seems to have taken for granted something that Alpha always had: good compilers. When the murder of Alpha was pre-announced by the INQUIRER in June 2001, I alerted AMD at a very senior level to quickly get a team of good compiler guys from the Alpha team, to help strengthen the Hammer base - since they didn't have the balls to simply take over the Alpha and not let it go to Intel. It seems my pleas fell on deaf ears - and today AMD has to depend on a half-baked PGI compiler suite, as well as perennial GCC cycle of "improvements and fixes". And yes, using Intel compilers for the best SPEC2000 results.

Now it is clear why AMD has nothing against Intel making Prescott, Tejas etc binary compatible with AMD64 - Intel can get the compiler job done for AMD hah.

Unused advantages
At present, the benefits of 64-bit platform on the desktop compared to the 32-bit platforn are far less clear than the benefits of 32-bit vs 16-bit platforms some 15 years ago. The current high-end games and authoring programs are pretty much happy with around 512MB to 1GB RAM.

The two key benefits that 64-bit platforms bring are crucial for workstation and server applications mostly, not for the desktop users: larger direct memory addressing - 16 EB (exabytes) vs 4 GB - and more precise and/or faster 64-bit integer number processing with 64-bit chunks at a time, important for many commercial and technical or multimedia applications.

Initial tests on the AMD64 platforms show a between 5% and 20% performance gain by running native 64-bit binaries vs 32-bit binaries compiled and optimised for the AMD64. Few codes could have an even greater benefit, but there also could be applications that have a slight performance dip in the 64-bit mode, due to the more cache space spent on large 64-bit address pointers in the code - leading to a few more cache misses.

To move or not to move
The benchmarketing war between Athlon64 FX-51 (a rebadged Opteron) and Pentium4 Extermination Edition (a rebadged XeonMP) is in full swing since last week - by now, it does seem that the Athlon64 side beats the 3.2 GHz Pentium4 EE side in a few more tests than vice versa, whether talking about the 32-bit or 64-bit code on the AMD side. However, if running the P4 Exterminator at 3.4 GHz, the tide turns slightly to the Intel's favor. So, don't be surprised to see the P4 EE thingy get announced at 3.4 and maybe even in 3.6 GHz versions as well, besides the standard 3.2 GHz flavour.

After all, in its 32-bit mode, and according to some early unofficial tests, Prescott is also only about 10-20% faster than the normal P4 at the same GHz on the usual score of current benchmark apps, compared to 5%-15% benefit from the P4 EE vs P4 on the same clock - just by slapping the 2MB L3 cache on it.

It seems pretty obvious that Intel will ultimately move to the 64-bit X86 arena, quite possibly with something that does run the AMD64 code directly, maybe even "extend the extensions" to add 32 registers per integer and SSE3 FP units, compared to "only" 16 on the AMD64 (not to mention just eight on the standard X86, and the huge 128 per set on the Itanic). Of course, due to the possible "iceberg" effect on the Itanic ship hull, Intel will not be advertising any such effort early.

Therefore, both parties might soon be knee-deep in the 64-bit PC mudfield battle - and you will then be anyway dragged by each of them to their side of the mudfield, either hypertransported or hyperthreaded. Take your pick.

So, the question is not whether to move to the 64 bits on the desktop, but when - now with the AMD Athlon64, or a bit later with Intel's "XXX" CPU? Well, hehe, that choice is up to you all - we're not paid by either party to promote them, but to bring you "truth and nothing but the truth" [Objective journalism is an oxymoron, Ed.] µ.