System on Module Performance

[Mads]

It is early days for the System on Module (SOM), and we are eargerly waiting for our first developers kit, but perhaps someone has had the fortune of testing the SOM already, and could give us an idea of the performance level we can excpect from it? In particular we are interested in how it compares to the cFP-2220 on numeric operations, but I do not expect anyone to have done a 1-1 comparison between those.

We have some extremely heavy number crunching to do (tomographic analysis), which takes the poor old cFP-2220 several hours to complete (another low-power SBC alternative in-house is even 10 times slower than that(!)). A modern desktop computer is able to finish the same job in a couple of minutes. I do not expect the SOM to get anywhere close to the desktop PC, but any rough idea on where in-between it might land?

[A.E.M]

Hi Mads.

The new SOM products uses the Zynq 7020 chip, which combines the processor and the FPGA on the same chip. The processor is a ARM Cortex-A9, with dual core and 667 MHz.

NI System on Module Development Kit

The cFP-2220 has a single core 400 MHz processor.

NI cFP-2220

So by just comparing the specs, I believe it is safe to say, that you will get a performance boost 🙂 I don't have any benchmarks to show you at this current stage.

[Mads]

I'm quite sure it's faster yes, but I guess we'll have to wait until the developer's kit has arrived. I'll report the results here, in case others are wondering on the same thing...

[Spex · Accepted solution]

The Zynq-7020 processor on the SOM is also used on the cRIO-9068 CompactRIO controller.  When that controller was released, we published some bechmarks relative to the cRIO-9074, which uses a 400MHz PPC processor similar to the cFP-2220.  Those benchmarks can be observed here:

NI cRIO-9068: Performance and Throughput Benchmarks

As a general guideline, we estimate typical applications will see roughly a 4x performance improvement on the cRIO-9068 (with Zynq processor) compared to previous generation systems with 400MHz PPC processors.

You may not see the exact same improvements, but expect similar results from your upgrade from Compact FieldPoint to the SOM. Unfortunately, we don't have a published benchmark on raw computational power; the benchmarks published demonstrate system performance improvements due to faster processing and other system-level improvements (such as faster DMA engines, faster memory access, improved Ethernet speed, more efficient OS thread scheduler, etc.)

Regards,

[Mads · Accepted solution]

Just received the SOM, and I have now run the tomograpic analysis on it. The result shows that the SOM is exactly 5X faster than the cFP-2220.

So even a bit quicker than expected, but that might be due to other additional resources, difference in OS and LabVIEW efficiency as well.

[Mads]

For anyone interested I compared the SOM with a cRIO-9030 as well.

The cRIO was about 2,5 times faster than the SOM (12 times faster than the cFP-2220).

My IntelCore i7-4770 Quad-Core desktop PC is about 10 times faster than the cRIO-9030...But that's not a fair comparison. I like the new targets...When it comes to performance per Watt the System on Module is especially good. Without the FPGA it only draws about 3 - 3,5 Watts when running our applications.