In a surprising development, we got word that Intel is apparently resorting to any kind of methods to push its Xeon Phi GPU compute accelerator cards on the market, against AMD’s FirePro and Nvidia’s Tesla. The new cards are supposed to excel at DP FP64 calculations, but use tremendous amount of power as we reported here.
Intel’s new Xeon Phi cards are powered by a chip comprising around 60 Pentium
-like x86 cores and the main feature is a theoretical x86 compatibility where the software is easier to port.
Right now, there are a lot of applications that are using normal x86 CPUs to compute certain tasks, but lots of these applications could benefit greatly if ported to Nvidia’s CUDA
or AMD’s OpenCL.
GPU’s from the two companies have huge computing potential if the software is ported to a GPU compute compatible form, but the costs of porting the software are considerable and many companies are still not sure that the move would prove profitable. Intel
’s Xeon Phi cards are not faster than Nvidia’s Tesla
or AMD’s FirePro, nor are they more efficient, but the company seems dead set on filling the market with them.
One feature that Intel heavily advertises about the Xeon Phi is a relative x86 compatibility.
The software still needs to be ported, but Intel claims that the costs for such a move are significantly reduced as porting to the x86-based Xeon Phi is less complicated.
On the other hand, Intel is famous for using any kind of tactics to undermine its competitors and now it seems that the company has done it again.
Experienced hardware analyst Theo Valich has reportedly
found out that Intel has sold thousands of Xeon Phi accelerators for just $400 a piece under the excuse that those were pre-production cards.
The thing is that the cards in question are used in a supercomputer in Texas’s Advanced Computer Center and in such systems, errors are completely undesirable.
practically sold perfectly working and completely validated cards at prices that barely amount to 10% of the usual $3000 to $4000 price of a professional AMD FirePro or Nvidia Tesla GPU compute card.