23 Sep
2023
23 Sep
'23
7:03 p.m.
On 2023-09-23 12:36 p.m., Paul Koning via cctalk wrote:
I say get the memory first, who knows what you need will be around
later. I wish you good luck, as FPGA software does what it thinks is
right, not what you think is right.
Ben.
On Sep 22, 2023, at 9:30 PM, Martin Bishop
<mjd.bishop(a)emeritus-solutions.com> wrote:
Paul
I endorse your point regarding Lattice's gouging. Support for anything prior to the
XO parts now costs a significant premium. Their XO2 parts are the most useful to this
community - free tools and 0.5 mm pitch, e.g. 100p & 144p - not dense but usefully
large, 3v3 IO and agricultural assembly.
The Xilinx free tools no longer have license files, which was how Lattice cut us all off
at the pass. The current Vivado ML Standard Edition (tools to normal people) are free up
to the XC7Z030 - which is a fairly serious device. I have a PDP-11 and space to spare
running in the markedly smaller XC7Z010; 16b / no MMU, most of the 45 instruction set.
FPGA are (organically) memory poor - perhaps because the access time is ~3 ns. I should
think you would be in with a chance of fitting the 6600 logic, however on a '30 you
have 265 x 4 ki by BRAMs = ~1 Mi By, if more is required either a dedicated external
memory device or DMA to/fr DRAM would be required.
The 6600 model I'm building is a gate level model, so it is cycle-accurate, but also
large. I'm figuring several hundred thousand gates, which makes sense if you consider
the module count for a 6600. A large enough FPGA for that seems to have enough on-chip
memory for both PP and CP memories, leaving only ECS as off-chip. That's helpful
because both PP and CP have tightly constrained cycles; DRAM would be nearly impossible to
make work, though SRAM is doable.
paul