| *** pietrushnic <pietrushnic!~pietrushn@2001:470:69fc:105::1:69c> has quit IRC | 02:30 | |
| *** pietrushnic <pietrushnic!~pietrushn@2001:470:69fc:105::1:69c> has joined #microwatt | 03:11 | |
| openpowerbot | [slack] <Benjamin Herrenschmidt> I was doing some back of the envelope math... Video will take a pretty good chunk of the system bandwidth.... | 16:35 |
|---|---|---|
| openpowerbot | [slack] <Benjamin Herrenschmidt> Full HD at 60Hz is half a gigabyte/s afaik | 16:35 |
| openpowerbot | [slack] <Benjamin Herrenschmidt> I could be interesting as a stress test of a non temporal DMA path :-) | 16:36 |
| openpowerbot | [slack] <Benjamin Herrenschmidt> I suspect letting DMA requests "sneak" in between cache line refill loads (between the loads of a single line) would kill things forcing the ram to pull a new bank and back and we might be better off stalling them until refill is idle and having a bigger read ahead FIFO for the pixels... It would be interesting to measure both options. I could even make some of this tunables registers | 16:39 |
| openpowerbot | [slack] <Benjamin Herrenschmidt> Plan B is a dedicated pixel fetcher in the L2 wrapper, but I don't think it would help much more and might cause placement difficulties adding more logic & muxes close to the L2 | 16:40 |
Generated by irclog2html.py 2.17.1 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!