[Oberon] Development boards

Skulski, Wojciech skulski at pas.rochester.edu
Sat May 2 04:53:45 CEST 2020

Peter asked:

> Is hardware really an issue? 

Yes it is, especially when it is absent.

> Is FPGA essential?

It depends for what. In this group we are talking of a neat System-On-Chip composed of the most standard peripherals (SPI, PS/2 driver, SRAM, and video) driven with a neat and very simple 32 bit CPU capable of up to 75 MHz. It is a small system. Why not replace it with ... (type the name of your favorite) which is of course much better, cheaper, and more popular (continue the list).

However, the FPGAs can also be used differently. Our 32-channel digitizer which we are developing at SkuTek is processing 32 parallel streams of 14-bit samples collected at 100 MHz on each channel. The FPGA is continuously processing 32*14*100 = 44,800 Mbits/s = 5,600 megabytes per second. (Neglecting the subtle difference between MB and MiB.) There are several dozens fixed operations in each channel for triggering, baseline removal, feature detection, and waveform storage. All operations run in parallel. The performance of this FPGA is staggering. And to impress you even more, the FPGA costs about $2k per chip. Just the FPGA. 

Is the FPGA essential in this case? Yes, it is. Can it be replaced with an Rpi or an array of Rpi's? No. Can it be replaced with a cloud? Perhaps. The power consumption would go through the roof, but people are trying. The idea of using CPUs rather than FPGA is lingering out there. Or perhaps GPUs. There are groups in our collaboration pursuing GPU farms for data processing. They still have some way to go.

One needs to know quite a bit about FPGAs to appreciate the technology. The FPGA is not the replacement for the RPi, or vice versa. These are very different domains.  

> ITX based systems are fairly small and low powered.

Very neat. How many engineers developed these boards? How many programmers spent how many man-years to program them? Any idea?

> S3 and A2 remain available.  That technology should provide an adequate basis for many embedded systems and control systems.

It depends what is the goal. If you want something small and easy then why bother with Oberon? Get whatever hobby system is out there and program a few "sketches" into it. Or some python scripts. It should work.

If you want to develop your own peripherals or IP cores like we are doing, then you need an FPGA to begin with. If this is the case then Oberon System suddenly starts looking very tempting. It would take about 1% of the resources in my Kintex-7.

If you ask why I have not done it yet, then the answer is that logistics has turned out a bit of a challenge in the small team.


More information about the Oberon mailing list