17:19 <mtrbot-ml> [mattermost] <pathfinder49> @rjo I've had a play with your example https://github.com/m-labs/artiq/issues/1373#issuecomment-577793300. As far as I can tell, the sinusoid updates at 1.3 MS. When I try increasing this towards 2 MS, an incorrect waveform is produced. Is your update frequency the limit?

17:27 <rjo> The link data rate is a constant and isochronous 2.55 MS/s. Half that rate is the limit of generating RTIO events by the CPU. "incorrect waveform" suggests a misunderstanding of what a sample is, how the Fastino data link works, and when the samples are converted.

17:29 <rjo> It's on the list of things to document in the coredevice driver. For now read the FPGA (Fastino and ARITQ gateware) code and documentation/comments.

17:37 <rjo> you might get full rate out of DMA. for how many channels is something tom wanted to test. i'll add a additional optional "wide" RTIO interface where one event contains samples for multiple/all DACs later.

17:37 <rjo> but that'll be pretty costly in terms of BRAM and logic usage for the RTIO fabric.

17:47 <mtrbot-ml> [mattermost] <pathfinder49> Thanks, I'll try DMA.

18:50 <mtrbot-ml> [mattermost] <pathfinder49> @hartytp DMA works fine at 2.55 MS with a single channel. I've tried up to 32 channels. This works for a short time only. Updating all channels at 2.55 MS I get an RTIO underflow in the DMA playback after 8 samples per channel (longer with less channels). With 2 channels DMA underflows after about 10 000 samples per channel.

18:53 <mtrbot-ml> [mattermost] <pathfinder49> @hartytp DMA works fine at 2.55 MS with a single channel. I've tried up to 32 channels. This works for a short time only. Updating all channels at 2.55 MS I get an RTIO underflow in the DMA playback after 8 samples per channel (longer with less channels). With 2 channels DMA underflows after about 1 000 samples per channel. Edit: corected cycle number typo

18:53 <mtrbot-ml> [mattermost] <pathfinder49> @hartytp DMA works fine at 2.55 MS with a single channel. I've tried up to 32 channels. This works for a short time only. Updating all channels at 2.55 MS I get an RTIO underflow in the DMA playback after 8 samples per channel (longer with less channels). With 2 channels DMA underflows after about 1 000 samples per channel. Edit: corrected cycle number typo

19:00 <mtrbot-ml> [mattermost] <hartytp> @pathfinder49 cool! Glad you've got things working well enough to gather statistics...

19:01 <mtrbot-ml> [mattermost] <hartytp> I think it's more long-term useful if you post this info on GH though.

19:01 <mtrbot-ml> [mattermost] <hartytp> nice job!