Applied Micro (NASDAQ: AMCC) hyped its X-Gene low-power server chips as a product that would fundamentally disrupt Intel's (INTC 1.25%) position in the sever market. The company touted very high performance while promising much lower power consumption than the competition.
According to performance tests from well-respected technology site AnandTech, not only does Applied Micro's first-generation X-Gene not come close to Intel's top Xeon processors, but it is far less power-efficient than Intel's Atom processors at roughly the same performance.
Cutting to the chase
AnandTech's numbers show the Applied Micro X-Gene 1 delivers eight units of performance per watt. Intel's latest low-power Xeon E3 processors -- based on the company's Haswell core -- deliver between 20.7 and 22.4 units of performance per watt. Even the older-generation Ivy Bridge Xeon chip delivers 16.9 units of performance per watt.
While Applied Micro had positioned X-Gene as a Xeon competitor, its performance across various workloads isn't commensurate with Xeon. It does, however, deliver raw performance that is roughly in line with Intel's Atom C2750 part, which features eight Atom cores. The Atom part delivers 19.5 units of performance per watt, though, suggesting more than double the efficiency of the X-Gene part.
Does this surprise anyone?
What I find so disturbing is that Applied Micro's CEO repeatedly referred to X-Gene as "Xeon class" in interviews, conference calls, and other public comments. He even disparaged some ARM (ARMH)-based competitors for using ARM's own Cortex A-class CPUs, which are generally Atom-class in performance, rather than Xeon-class.
However, AnandTech's performance numbers don't lie: In single-threaded tests, the X-Gene at 2.4GHz performed roughly in line (actually slightly below in the LZMA single-threaded performance test) with the 2.4GHz Atom C2750. Guess what that means? The core inside X-Gene is Atom-class, not Xeon-class. I suspect off-the-shelf ARM cores would also have no trouble matching or exceeding the X-Gene.
Furthermore, as has been known for quite some time, the X-Gene is built on an antiquated 40-nanometer manufacturing technology while the Atom C2750 is built on Intel's 22-nanometer FinFET technology. This is one key reason Intel's Atom delivers so much better performance per watt than the X-Gene part, and why the first-generation X-Gene probably never had a chance.
The next X-Gene probably won't be much better
Applied Micro claims its next-generation X-Gene (known as X-Gene 2) will be 50% more efficient, according to AnandTech. If that's the case, X-Gene 2 is unlikely to be competitive with even Intel's C2750 in terms of performance per watt. Why? Adding 50% more performance per watt to the numbers AnandTech measured provides 12 units of performance per watt. That's still lower than what the Atom C2750 delivers; again, that's not surprising given that the X-Gene 2 will be built on an older 28-nanometer process.
Additionally, according to Intel's public road maps, its follow-on to the C2750, code-named Denverton, should be built on the chipmaker's 14-nanometer manufacturing technology. Few details of this chip are available, but the improved manufacturing technology alone indicates a solid power=efficiency boost.
In other words, I don't expect much success for X-Gene today or in the future.