Hi Gerry and Jon, This memo gives a quantitative analysis of multiplier, adder and memory utilization for the F engine
https://www.cfa.harvard.edu/twiki/pub/SMAwideband/MemoSeries/sma_wideband_utilization_1.pdf Some recent optimizations are not taken into account, however it is still valid by and large. The link should be freely accessible from anywhere on the net but if you have trouble, let me know, or if you have any other questions. Cheers, Jonathan On Aug 24, 2012, at 7:07 PM, Gerry Harp wrote: > Hi Dan > > Your numbers look somehow backwards. > > Did you mean: > 32 inputs at 8K > 16 at 4K > 8 at 2K > 4 at 1K? > > Presumably, by narrowing the input bandwidth could permit proportionally > smaller channels at the output? So at 2.5 MHz input one could do 1 Hz > channels? > > Trying and compiling is the right thing to do. Hopefully we can arrange to do > this pretty soon. > > Gerry > > On 8/24/2012 8:55 AM, Dan Werthimer wrote: >> >> hi john, >> >> to get a real number, it would be best to compile the design . >> you can re-compile zaki's octal F engine designed for roach I >> (8 inputs at 250 Msps), or Hong's F engine (32 inputs at 250 Msps, >> designed for roach II). >> >> rough guess: you can get 32 analog inputs at 1K channels, >> or 16 analog inputs at 2K channels, or 8 analog inputs at 4K channels, >> or 4 analog inputs at 8K channels. >> >> best wishes, >> >> dan >> >> >> On Fri, Aug 24, 2012 at 7:54 AM, Jon Richards <[email protected]> wrote: >> All, >> >> I need some advice. We are looking into developing an FX correlator using >> the Roach2 s as the "F" and GPUs as the "X". >> >> We will have 4 inputs to the Roach2 ADCs, at 250Msamples/sec, 8Gb/sec >> flowing out to the X boards. How many channels should the Roach2 be able to >> process? How would I guestimate that? >> >> Any ideas appreciated! >> >> Thanks, >> Jon Richards >> >> <Mail Attachment.png> >> > > -- > ---------------------- > Gerald R. Harp, Ph.D. > Director, Center for SETI Research > SETI Institute >
