
In a surprising advancement, we received word that Intel is apparently resorting to any sort of approaches to push its Xeon Phi GPU compute accelerator cards on the market, against AMD&rsquos FirePro and Nvidia&rsquos Tesla. The new cards are supposed to excel at DP FP64 calculations, but use tremendous quantity of energy as we reported right here.
Intel&rsquos new Xeon Phi cards are powered by a chip comprising all over 60 Pentium-like x86 cores and the primary feature is a theoretical x86 compatibility in which the software is much easier to port.
Ideal now, there are a good deal of applications that are utilizing regular x86 CPUs to compute specific tasks, but tons of these applications could advantage significantly if ported to Nvidia&rsquos CUDA or AMD&rsquos OpenCL.
GPU&rsquos from the two companies have enormous computing potential if the computer software is ported to a GPU compute compatible kind, but the costs of porting the application are substantial and numerous companies are even now not confident that the move would show worthwhile.
Intel&rsquos Xeon Phi cards are not speedier than Nvidia&rsquos Tesla or AMD&rsquos FirePro, nor are they additional efficient, but the company appears dead set on filling the market with them.
One particular feature that Intel heavily advertises about the Xeon Phi is a relative x86 compatibility.
The software program nonetheless demands to be ported, but Intel claims that the expenditures for this kind of a move are appreciably lowered as porting to the x86- based Xeon Phi is significantly less challenging.
On the other hand, Intel is famous for using any kind of tactics to undermine its rivals and now it appears that the business has performed it once again.
Experienced hardware analyst Theo Valich has reportedly located out that Intel has sold thousands of Xeon Phi accelerators for just $400 a piece beneath the excuse that people were pre-production cards.
The matter is that the cards in query are employed in a supercomputer in Texas&rsquos Sophisticated Computer system Center and in such systems, errors are entirely undesirable.
For that reason, Intel practically sold properly working and fully validated cards at charges that barely amount to ten % of the typical $3000 to $4000 value of a skilled AMD FirePro or Nvidia Tesla GPU compute card.
No comments:
Post a Comment