OSDev.org

Interesting.

It's none of those actually, rather a combination of radio astronomy and software defined radio. I will use a broadband radio receiver to feed the ADC, but then I will want to remove all the man-made stuff by subtracting the strongest radio stations. I don't think I can do this in realtime, but if I have 128GB of memory to record the signal into, then I can do analysis of maybe half or one minute of data at a time. The realtime stuff is for the next stage when I know what to look for.

So, yes, I might eventually be able to construct a reduction algorithm, but it likely won't be anything standard and so ready-made designs will be of little use. I could use some DSP module, but a PC solution is more flexible and easier to use & program.

Microcomputers didn't represent any sort of conceptual advance. They by and large reimplemented well-established concepts on smaller scales and just plain ignored many conceptual advances throughout the history of computing. Home electronics hobbyist projects had been around since the turn of the 19th-to-20th century, so they weren't new. One interpretation of the hobbyist computers / microcomputers is that the byproducts of industrial computing achieved some sort of critical mass of usefulness leading to greater marketability as opposed to any sort of genuine innovation being involved. The standard post-WWII model of not-just-basic research being carried out at universities and by the military only to be given away to cronies in the private sector applied to computing as it did for aviation, medicine, and really everything else e.g. the internet. There is now a moderate secondary design field in existence, but even as they are distinct designs, they typically represent technically inferior solutions involving well-established concepts specialised to smaller systems with a strict subset of the technical developments of the history of computing as opposed to any sort of originality. FibreChannel, InfiniBand, enough RAM to make 64-bit meaningful above physical addressing extensions for 32-bit, etc. are all beyond the reach of the "mainstream consumer" beyond just multi-socket boards, or, for that matter, multi-board systems, or meaningfully powerful non-x86/RISC systems. So colour me disgusted regarding how technically inferior products are foisted on the mass market instead of best-of-breed designs and implementations actually trickling down.

I decided against using a two processor motherboard. Mostly because it wasn't cost-effective. The server components are much more expensive with poorer preformance, and I'd need more memory chips at considerably higher costs. The AMD server processors also are clocked at a considerably lower frequency, and Threadripper processors are much more cost effective.

So, I'll build a new computer based on a 12-core (24 thread) Threadripper processor that has the additional PCI lanes required for the data acquisition board (it needs 8 lanes). I've also bought 8x16GB DDR4 RAM chips (the limit of the motherboard). The FPGA evaluation board I'll use have hundreds of DSP cores, and so could be used for realtime analysis.