If you cleverly store 60 MS/s as 8 bit binary files, you may get close to fast disk speed. If you convert them to single precision or anything else you well certainly have to get a raid and possibly multiple channels at that.
Further this will impact the PCI bus and you better have good bandwidth to memory. Running all 3 boards on a single system will certainly give it a work out! Even if you can stream this to disk, it is going to really tax the system if you ever try to get it out of the disk while acquiring. Adding in the GPIB and AO is just adding times when the system will block and that the data will not use the memory controller bandwidth efficiently. Now if you want to average a seconds worth of data to put in the fifo this is the wrong way to do it. Put an analog filter of 1 second on the thing and not try to average 1 million samples. Trying to fix in software what is broken in hardware (ie noise) is not the way to go. Get a good filter on the system and then a high precision analog board that is a lot slower. Heck, if you are just doing a running average every 2 seconds that averages 2 minutes of data then taking 1.2 billion samples to get what you could get with an RC filter and a 2 second digitizer this is the wrong way to go. But the requirements may need all that data stored. If you really want 3 boards and record all data the best way would be to install in 3 separate computers and route trigger signals and clocks between them as necessary. -Scott >This may sound naive, but I have a customer who wants to sample up to 6 channels >across 3 boards (2 channels/board) at a sampling rate of 10 Megasamples/sec/channel >(i.e., 60 MS/s in total). > >Three different units are being measured simultaneously, which is why 3 boards are >necessary (they don't all start or stop at the same time, but they could all be on at >the same time). > >Currently the system measures 50KS/s/channel with no problem (plus up to 2 channels >of AO). The data is averaged each second and put into a FIFO buffer which is usually >2 minutes (120 samples) long. Every 2 seconds, the FIFO is averaged (to get a >running average). I should add that a GPIB data logger is read every 2 seconds for >slower data. > >My NI rep has suggested the PCI-6115 whose specs could certainly handle this. What >I'm worried about is the computer keeping up with averaging 60MS/s. > >Anyone have any experience using these data rates over an extended period of time >(measurements can last WEEKS).
