What types of programs would perform better on a Xeon Phi as opposed to, for example, a macbook CPU?
tclarke
Programs that would perform better on a Xeon Phi are programs that can take advantage of highly parallelizable code and data. Especially code like matrix multiplication, vector addition, and other scientific processing. A Macbook might have a faster clock rate and a larger memory, so it would be more amenable to large amounts of serial code or many serial programs running at the same time.
ferozenaina
Most common GUI/desktop software are not optimized to run on > 4 cores. Many were designed to run only on a single core. These would do really well with a high clock rate CPU.
The Xeon Phi has a 1.1Ghz clock rate which is super slow. Unless the program was designed and parallelized to run on 72 cores, it would be better to run on dual/quad core CPUs with 2+ GHz clock rate. I think only research/scientific computation would come in this category.
Video processing applications (which can be easily parallelized) are designed to work using GPUs.
ferozenaina
What is the advantage of Xeon Phi vs. a NVIDIA 1080? For both, the code would have to be parallelized heavily. Programming on Phi is easier than CUDA but Phi is 2x cost of the 1080. Any reason to pick the Phi over a GPU?
I think the problem type (compute/bandwidth bound), cache size would play a big role here. The 1080 GPU has more raw computing power.
Are there any popular applications that can use the Phi? Where are they typically used?
cuiwei
@ferozenaina: The most obvious advantage of Xeon Phi over Cuda is that, you can port your code originally written to be compiled and run on a X86-64 machine almost for free, while porting them to Cuda might requires a significant part of the code re-written. This is unacceptable in many cases. However it appears that high-end Nvidia cards indeed outperform Xeon Phi. Here is an article for your reference: http://www.ds.ewi.tudelft.nl/fileadmin/pds/homepages/fang/papers/iwbbio2k14a128.pdf.
What types of programs would perform better on a Xeon Phi as opposed to, for example, a macbook CPU?
Programs that would perform better on a Xeon Phi are programs that can take advantage of highly parallelizable code and data. Especially code like matrix multiplication, vector addition, and other scientific processing. A Macbook might have a faster clock rate and a larger memory, so it would be more amenable to large amounts of serial code or many serial programs running at the same time.
Most common GUI/desktop software are not optimized to run on > 4 cores. Many were designed to run only on a single core. These would do really well with a high clock rate CPU.
The Xeon Phi has a 1.1Ghz clock rate which is super slow. Unless the program was designed and parallelized to run on 72 cores, it would be better to run on dual/quad core CPUs with 2+ GHz clock rate. I think only research/scientific computation would come in this category.
Video processing applications (which can be easily parallelized) are designed to work using GPUs.
What is the advantage of Xeon Phi vs. a NVIDIA 1080? For both, the code would have to be parallelized heavily. Programming on Phi is easier than CUDA but Phi is 2x cost of the 1080. Any reason to pick the Phi over a GPU?
I think the problem type (compute/bandwidth bound), cache size would play a big role here. The 1080 GPU has more raw computing power.
Are there any popular applications that can use the Phi? Where are they typically used?
@ferozenaina: The most obvious advantage of Xeon Phi over Cuda is that, you can port your code originally written to be compiled and run on a X86-64 machine almost for free, while porting them to Cuda might requires a significant part of the code re-written. This is unacceptable in many cases. However it appears that high-end Nvidia cards indeed outperform Xeon Phi. Here is an article for your reference: http://www.ds.ewi.tudelft.nl/fileadmin/pds/homepages/fang/papers/iwbbio2k14a128.pdf.