AMD has coined a buzzword: mega-tasking. I’m still not positive what AMD means by it, but I’d use it to describe the work-style of those on the verge of needing second desktops or workstations to accomplish their heavy mix of foreground applications and background tasks. I estimate that a genuine power desktop is replaced on cycles of eight months to one year. This rapid cycle is exactly what motivated us to find server consolidation solutions, and we’re getting to the point where desktop consolidation is needed.
But client consolidation isn’t as easy as it is with a server because virtualisation can’t play as significant a role. A client system needs to maintain a constant state of low-latency responsiveness no matter how much is thrown at it. Hardware-accelerated virtualisation makes a big difference for execution speed, but peripheral throughput remains a choke-point. These will be addressed in future CPU and chipset architectures, but there is a solution right now.
I was given early access to a just-released AMD reference platform called Quad FX (code-named “four-by-four”), a dual socket, quad-core client platform developed in partnership with graphics and chipset shop Nvidia. The heart of Quad FX is the Athlon 64 FX-70 series CPU, an offshoot of the Revision F dual-core Opteron server processor. In the case of my Quad FX reference machine, the processors are 3GHz FX-74s, genuine (meaning two complete AMD64 CPUs on one die) dual-core CPUs equipped with 1MB of independent level 2 cache per core, independent on-chip memory controllers in a NUMA (non-uniform memory architecture) configuration, and independent buses reserved for traffic between cores and sockets. As I keep harping on, it’s all about the bus, and “independent” is the watchword for scalability on both servers and clients. If anyone thinks that 2006 put Intel ahead of AMD in any way other than hype, it’s my job to tell them they’re mistaken.
The FX-74 Quad FX platform running in a two-headed (two monitors) configuration genuinely obviates the need for a second machine. That’s hard to swallow I know, since Core 2 Duo desktops like the inimitable Mac Pro — the only Woodcrest (Core 2 Duo Xeon) desktop worth looking at — deliver the unmistakable feel of one extremely fast x86 computer. That’s satisfying as all hell; Mac Pro and OS X multi-task with an absolute vengeance, with all credit to Apple’s engineering. Mac Pro is an ideal balance of mean and green. But full independence of core, cache, socket, memory, I/O and inter-core and inter-socket communication allows constantly concurrent operation of all four — and soon, all eight — AMD64 cores in Quad FX and Opteron platforms.
In high-demand scenarios, Core 2 Duo will go round-robin, or sequential, with cores’ access to memory, peripherals and each other, a trait that suits Core 2 Duo more for multi-tasking than mega-tasking. I’d fetch up some numbers to support this take, but I feel I would just be driving Quad FX around with Vista, and you will, too, if you’re mega-tasking.
Quad FX points the way to the next generation of commercial client systems. Wait if you want to, but I’ll say this: if you take your first drive of 64-bit Vista on Quad FX you may move up to Opteron, but you will never go back to Intel.