When Apple migrated from the IBM PowerPC G5 to Intel CPUs for the new Mac Pro, the new Mac Pro was hard pressed to show a significant speed advantage over its predecessor. Meanwhile, AMD was providing Intel with a lesson in CPU design with its Opteron, the darling chip for high performance servers, scientific work stations and clusters a few years ago. However, Intel has now caught on and caught up with the Nehalem series, and Mac-using scientists everywhere can rejoice.
Sometimes people, especially scientists, are anxious for Apple to jump ahead of the competition. That's a natural reaction -- Apple's ease of use and UNIX roots make it a natural for scientists on the desktop. However, Apple often ends up lagging. Apple was slow to move from PCI-X to PCI Express, and they were also, thanks to Intel, held back a little in CPU design.
Nehalem Quad Core CPU
When the AMD Opteron first came out, it heralded a new design that allowed the cores to access memory directly while Intel designs were, in a sense, glued together and data needed to travel through a bus to get to memory.
Intel has learned from the AMD design with the Nehalem design, and as Apple puts it:
Many quad-core processors are composed of two separate dies, which means some cached data has to travel outside the processor to get from core to core. That’s an inefficient way to access information. Enter the Quad-Core Intel Xeon “Nehalem” processor. Its single-die, 64-bit architecture makes 8MB of fully shared L3 cache readily available to each of the four processor cores. The result is fast access to cache data and greater application performance. Combine that with the other technological advances and you get a Mac Pro that’s up to 1.9x faster than the previous generation.
Of course, what Apple ddn't mention in the above note is that the term "many quad-core processors" refers to Intel's own previous generations of chips.
One of the problems with Intel's former designs, and a problem solved by the AMD Opteron, was that as more cores were added to the system, contention on that memory bus caused performance to drop. Perhaps Intel engineers didn't really believe it would happen years ago, or they weren't worried about it in dual core designs, or the "Not Invented Here" arrogance syndrome prevailed. Or all three. But it turned out to be the case.
This new architecture implemented by Intel has led to what looks to be a substantial increase in performance of the Mac Pro and finally distances the machine for good from the legendary IBM PPC 970, G5. With an estimated 2.4x increase in memory bandwidth, the NVIDIA GT 120 GPU and the coming OpenCL and Grand Central in Snow Leopard, the Mac Pro can finally be regarded as a full-fledged supercomputer on the desktop. The drool-factor has just escalated seriously.
Scientists and engineers: time to write up those purchase requests.
The Beastly Mac Pro, Now in Heft AND CPU Power