Oh I don’t know about that, a really scalable solution could still move mountains. My aforementioned PE had 1024 processors with extremely slow IPC, no floating point, and a local 256kB of RAM per CPU. Running it clocked to do a soft reset on each horizontal sync pulse on NTSC, it could do real time processing of video by having the front end quantize luma and chroma into 8 bits and then sending one pixel sample per line to each CPU. I ran an edge detection algorithm on it and then output the edges (with varying strength expressed as luma magnitude) overlaid on the video in real time. Granted - this would probably be a lot slower with the slow CPU clock (not really sure since the 6502 is more capable in some ways), but I’d love to see a massively parallel C64 just for the ridiculous glory of it. Justin > On Apr 20, 2020, at 15:00, smf <smf_at_null.net> wrote: > > SuperCPU works by having it's own faster ram, with the option of sending > updates to the 64k main ram so that vic2 can access it. Those updates > have to be sent at 1mhz and you have to avoid vic2. > > You can specify what areas of memory are mirrored, if you mirror all of > ram then it can have a severe performance overhead. > > Sharing multiple cpu's between 64k is kinda pointless, even if you end > up making it able to run two cpu's simultaneously then you're crippled > by the lack of memory to use that extra cpu power for. > > On 19/04/2020 21:24, tokafondo wrote: >> Well, time then to learn how C64 'Turbo' accelerators with external CPUs did >> work, then. >> >Received on 2020-05-30 01:34:39
Archive generated by hypermail 2.3.0.