@tomas Yes; the FPGA I'm using right now is the Xilinx Spartan 3 device, XC3S1200E, 4ns delay.

The board I'm using is a Digilent Nexys-2, which unfortunately are no longer manufactured. However, I can't imagine this design would be difficult to port to a different FPGA board.

@tomas This computer is a stepping stone to another computer design I've been wanting to build for some time, which I've dubbed the Kestrel-3.

The Kestrel-3 will be color, have programmable video resolutions, maybe audio if I can figure out a good built-in solution, etc.

The Kestrel-3 will also be built using nothing but FPGA boards that can be targeted with Yosys, although it'll take multiple FPGAs to make everything work.

Yosos.. Is that the fool for the iCE40 series of FPGAs? I'd have thought those were a bit on the small side for a homebrew CPU, but I guess if you use enough of them it's fine. Break the CPU up, like the old Fairchild F8

Exciting times for RISC-V. Just youtube-dl:d the first part of that video series that's on hackaday now, should be perfect thing to watch during laundry
Follow

@tomas Surprisingly no; the iCE40HX4K and 8K are plenty big enough to house a 64-bit RISC processor as long as you don't go overboard with features.

My Kestrel-3 design is slated to use just two FPGA boards: one for the CPU, local RAM controller, and basic I/O (e.g., serial interface for hardware bring-up), and the other is for all the I/O that the Kestrel-3 will use (video, keyboard, mouse, etc.).

@vertigo Nice! I haven't dived fully into FPGA land yet, but I have one of those iCEsticks. I looked into it mostly for its antifuse features, which felt relevant for the harshness of space

I see neither your site nor hackaday.io has RSS/Atom, which is a shame. But hopefully subscribing on here is enough for updates on this :)

Paging @mikael
I had also forgot that RISC-V specifies a 128-bit version. I have a feeling that's going to be great for crypto stuff

@tomas The specification is not standardized, however, which is why instruction encodings for it aren't published yet. But, IIRC, it's supported in the Spike emulator, which is the reference (non-hardware) used to define the specifications with, so I'm glad it's there!

Yes, it'd make crypto easier, but lack of constant-time instructions is a big drawback. A task force exists now which is looking to add constant-time extensions to the spec. It's been slow-going though.

@tomas For me, the 128-bit specification is important because it opens up opportunities for finally realizing object-capability-based operating systems and applications. Both forward and inverted page tables starts to break down at 128-bit address space sizes (by which I mean a hardware walk is about as slow or slower than an optimal software TLB fill), so we'll start to see some nice innovations here, I think.

@vertigo This has some implications for distributed computing too. You can have an entire cluster share the same address space, making use of network DMA to send stuff around. Give every single byte its own IPv6 address
Sign in to participate in the conversation
Mastodon

Follow friends and discover new ones. Publish anything you want: links, pictures, text, video. This server is run by the main developers of the Mastodon project. Everyone is welcome as long as you follow our code of conduct!