Intel® TBB 3.0 in Intel® Manycore Testing Lab

A few weeks ago I had an opportunity to use Manycore Testing Lab (MTL) server which was not loaded by academic tasks that time. MTL is free for the Intel Academic community. Information how to get access can be found on MTL page /en-us/articles/intel-many-core-testing-lab

The manycore servers do have the latest Intel compilers and libraries installed. New version of threading building blocks (3.0) has also been installed after its release. Access to the systems and setting the environment is as simple as access to the local linux boxes. Not sure about xserver part but ssh, scp and console tools work just fine.  MTL forum contains several hints how to use servers more efficiently like assigning jobs to cores, not using HT, setting the environment and so on.

When our team was working on releasing version 3.0 of TBB, I was conducting performance measurements on MTL server. Results were included into official marketing documentation available in “Intel® Threading Building Blocks 3.0 In-Depth” paper available at /sites/default/files/m/3/5/8/Intel_TBB3_InDepth.pdf. I have run almost all examples from TBB 3.0 package. Depending on a workload examples show different scalability results: from expected very bad for Fibonacci which does not scale at all and shows basic usage scenarios to pretty well scaled examples which have good scale for 64 threads and especially for 32 threads which match to physical cores. But I’d like to write a bit about Sudoku example which was introduced in TBB 3.0 to show new task_group feature.  This example is out of official performance data because it is not well optimized comparing to the best serial versions of Sudoku solvers but it shows amazing scalability which can be seen on the picture below. input4 initial Sudoku table was used for calculations with default options for compilation and running the example including default scheduler initialization. The command line was following:

$ for i in {1..64};do ./sudoku input4 $i 2; done

After TBB 3.0 was released it was installed on the servers in the lab and MTL users can get an access to it.

For more complete information about compiler optimizations, see our Optimization Notice.

1 comment

Mike P. (Intel)'s picture

I'm glad to hear that your use of the MTL was sucessful - I think you managed to catch it in a somewhat free moment or two. Thanks for taking the time to include the MTL in your testing scheme.


Add a Comment

Have a technical question? Visit our forums. Have site or software product issues? Contact support.