Software Tuning, Performance Optimization & Platform Monitoring
Discussion regarding monitoring and software tuning methodologies, Performance Monitoring Unit (PMU) of Intel microprocessors, and platform updating.

why is Atom 3826 much slower than Celeron M 440



I was told to join some software forum for my question better than a hardware related I started with: So I hope to be right here.

I am trying to find a new plattform for an app I am running. A plattform that give at least the same performance.

Up to now the app is running on a Celeron M 440 under XP.

So now I'd like to run it as it is on a newer plattform using Win7 and an up to date processor that matches this app.

So the first test is an Atom E3826 that might give the same performance than the "old" Celeron M 440.

But the result is for me very astonishing. It is running 3 times slower.

As I wrote the app by myself I could cut it down to the main problem.

The core of the app is very simple; it is mostly running image filters.

So it is just running over a long byte array, do some calculation, store the result back, finish.

Looks like this:

1.    Initialisation of an BYTE array; length is 10.000.000 :
          const int LEN = 10000000;
          unsigned char *arrPtr = new unsigned char[LEN];
          for (int i=0; i<LEN; i++){
              arrPtr = (unsigned char)250;
2.    Run the filter
          int sum=0;
          for (int i=1280; i<LEN-1280; i++){
              sum = (int) (4*arrPtr + arrPtr[i-1] + arrPtr[i+1] + arrPtr[i-1280] + arrPtr[i+1280]);
              arrPtr = (unsigned char)(sum / 8);

When I run this small filter algorithm (step 2.) on the "old" Celeron M 440 it needs around  70ms.

When I run the same app on the Atom 3826 it needs around 210ms.

Why that big difference? It is not just 20%, it is 3 times. There must be some basic problem.

As the app is just crunching data, the much faster memory interface of the AtomE3826 should give a big advantage.

By the way, as you see this app uses just one thread and has no need for multicore performance but just single core performance

And it is a x86 app made for XP and I run it in XP compatability mode on Win7.

Could this be a problem how effective the processor is used ?

But on my Desktop PC (also Win7) it runs really fast.

I could send you my small cut-down test-app if it is of any help.

Or would you recommend another processor type  to replace the Celeron M 440 ?

Thanks for any tipps and help



0 Kudos
0 Replies