VIA Chipsets Part Deux

Alright, first thing’s first. In Via chipsets the RAM clock runs asynch to the CPU bus clock and V-Link clock. In recent Via chipsts the PCI interface has been replaced with a 266 MHz proprietary bus called the V-Link, which convieniently matches to an even multiple of AMD’s 133 MHz EV6 bus (266 MHz data rate) and Intel’s 133 MHz AGTL bus (533 MHz data rate).

Onto Barton, it is simply a t-bred with double the cache using the same, slow, high latency cache bus. I suspect with good reason the performance increase will be minimal at best in all but server apps (look over at Intel’s Pentium III-S parts for examples, and remember they have four times the cache bandwith per clock at a two tick latency as oppsed to a 4-8 tick latency). And FYI: Intel has been shipping parts at 0.09 microns since AMD has been shipping parts at 0.13 microns. Intel is an entire process generation ahead of AMD.

Now then, onto other topics, Intel’s 64 bit processor is the Itanium, not the Titanium, and AMD’s forthcoming 64 bit processor is codenamed Hammer and will be marketed under the Athlon brandname with small cache sizes for single processor machines and under the Opteron brandname for 2, 4 and 8-way machines. Intel’s Itanium uses an explicitly parallel VLIW chip built from the ground up to respond to the computing needs for the next 20 years, and at 1 GHz is currently the fastest available or planned processor in floating point. It blows the doors off of the Alpha, PA-RISC, MIPS, UltraSPARC, Pentium, PowerPC, POWER3/POWER4 and Athlon line of processors not only in IPC on the FPU, but in sheer FPU performance, even when the other processors are given up to a 3x clock speed advantage (about the point at which the Athlon would overtake the Itanium is 3 GHz, the Pentium 4 would take to 4 GHz). It utterly and completely is a winner at floating point. In addition, it has the fastest IPC of any processor, ever, outside of speciaialty CPUs. Next is the AMD performance, which will hold the leader in ALU performance … unless Intel can get a 3.5 GHz Pentium 4 out of the door before AMD launches its Hammer at 2.0 GHz. And this is entirely possible. While performance is good on the Hammer, scalability is lacking and in addition adoption is lacking. Yes, Microsoft plans to put Longhorn on x86-64 … whenever it comes out (think Q2 2003 is where we’re looking at right now), but they also plan ports for PA-RISC and possibly PowerPC. And, did I mention the fact that they already have an IA-64 (Itanium and Itanium 2) port for Windows XP available that you can buy right now? Yes they do, and .NET server is only getting IA-64 support, not x86-64, which negates the largest advantage to AMD going to a 64 bit processor – getting around the 4 Gb limitation (Intel did this by using PAE, but AMD has yet to impliment it). The biggest winners from x86-64 initially will be the Linux world, but if you’ve seen any of Intel’s roadmaps you will find IA-64 on your desktop by around 2006 or 2008, with far more scalability than any x86 processor could hope for. Recall, most of the time on modern CPUs is spent getting around the limitations of the x86 ISA, and by eleminating that Intel has created a technically far superior CPU that will likely scale to 2.0 GHz on the next die shrink (from 0.18 to 0.13 microns) and then to 4.0 on the 0.09 micron process. Since Intel is close to production yields at 0.09 microns for certain parts (the Pentium 4 and Xeon processors will not be the first to get it), Intel can push these out of the door far faster than AMD. And recall that with equivelant clock speeds the Itanium will always win, and will still edge out the Opteron in FPU where the Opteron wins ALU with the Opteron at double the clock speed.