 The T2 tile project is building an indefinitely scalable computational stack. Follow our progress here on T Tuesday updates. I'm Dave Ackley, a retired computer science professor, and I've loved computer programming since using FORTRAN on punch cards in the 1970s. But over the years I started to see that what makes programming so much fun, the power that you get to control, everything the machine does, is also what makes the resulting system so incredibly fragile and breakable and ultimately limited and untrustworthy. But I also saw that to do computing right, to build scalable, resistant, sensible systems, was going to take much more than new buzzwords on business as usual, much more than patching apps and changing passwords and blaming users because they don't. We need to reinvent the whole computational stack, hardware and software both, to really do it right. Now, a conventional computer is a world in a box, fixed and finite with totally centralized control and the programmer and then the program is the omnipotent master of that little universe. We could break that box open, we could basically turn it inside out and have hardware that could be expanded as needed, like tiling out a bigger ballroom floor after the dancing has already started. A whole new stack is admittedly too big an ask for most folks and the grant proposals I wrote about it pretty much got bad reactions and never went anywhere. But even though such a tile based indefinitely scalable machine would eventually need new chips and new circuits to make it fast and cost effective, we could get started by building prototype computing tiles using just off the shelf technologies. Even though those tiles would be huge, slow and expensive, we could try to make a hundred or more and as prototype hardware floor upon which to develop the software choreography and grow this new computational stack. So that's what I decided to do, I'm mostly a software person and I've always been fascinated and terrified by hardware design but I worked on this T2 tile project sporadically for a few years until in August 2018 I retired from academia to focus on it full time and in October 9th 2018 I started the T Tuesday updates YouTube channel to document my progress to find like minded folks out there if there were any and to keep me from slacking off too much when it had become extremely clear that there was a tremendous amount still to do. It's now been a year of T Tuesday updates, 52 episodes and even though we're not as far as I'd hoped we'd be by now, the truth is Dave, there's really been incredible progress across the board. Check it out. I mean in the pilot episode I asked the world for help and the world answered. People watched the videos and made comments and suggestions and asked questions and some folks went way beyond that, building out on social media for education and outreach. There's now t-shirts and stickers and other merch. We've live streamed public meetings, there's now a T2 tile subreddit and a wiki and dedicated chat rooms with lots of Q&A and lively discussions, it's great. And there's an online simulator to make playing with these new ideas as easy as pointing your browser at mfm.rocks. And now there's even early support for running simplified versions of the software on GPUs, graphics processing cards and getting incredibly high speeds out of it. Thank you folks for all that incredible support and to everybody for caring enough to check back in week by week for the updates. For my own efforts at education and outreach I gave conference talks and workshops. I wrote a paper about our latest developments in artificial cell membranes which are destined to be I hope to be a high level software organizing principle once the hardware tiles actually start working in quantity. And at the same time over the weeks and months I pushed forward on those T2 tiles themselves as best I could. I refreshed and revised the circuit board design I had started in 2017, got new PCB samples made, I fleshed out the bomb, the bill of materials and really started to appreciate just how deep the hardware rabbit hole goes. This part or that part, this 20 digit part number or that 20 digit part number that differs in one digit. What's the difference? Who knows? It's buried in the data sheet somewhere, better just get samples of both and try them. I hand built a few boards to test and for you bet to debug and deal with oscilloscopes and all kinds of incredibly terrifying hardware stuff and although I was and I still am completely terrified that I've missed some totally obvious hardware problem that's going to show up and bite my butt later, eventually the T2 electronics design started to settle down and seem plausible. I worked on 3D design for the tile cases with all the holes in just the right places so that the inter tile connectors could plug in between them when you put the tiles next to each other in the proper orientation. My 3D printer which I got started flaking out but eventually I fixed it myself even though it was supposedly still under warranty or eventually I also went and did designs for power zone frames not just for cases for a single tile but now to hold a group of tiles so that they can be handled as a unit and keep the power connections all internal. You know I called myself software guy lost in hardware land and for sure that's true but looking back now really I am scientist lost in engineering land just as much and on top of all of that I'm academic guy lost in commercial land. When it came to acquiring enough parts to manufacture over 100 tiles it was all about quantity limits and lead times and often negotiation with vendors like halfway around the world and I am not a good negotiator but gradually the parts came in and I finally sealed a handshake deal with the owner of an albuquerque electronics manufacturer to stuff the board to solder all the components in place using their pick and place machine and their reflow ovens and so forth and now we have over 150 T2 tiles in hand waiting for software. Yeah software you know that part that I'm supposed to be good at but of course you know for a whole computational stack software is not just one thing there's the coprocessors coded in assembly language and Linux kernel modules to manage communications and locking and a common data manager to distribute the system software out over a grid of tiles visualization program to see what's going on and all of that before even beginning to face the user level programming and languages that we have called the MFM and Splat Ulan and Splat and that was really the main focus of all of the previous work so all of that software effort since then has been dedicated to getting intertile events working where some user written code would just do its thing making local changes to its environment but those changes would end up partially on this tile and partially on the neighboring tile so that the whole computation could ultimately be movable thus the name movable feast machine and as year one of these T Tuesday updates ended we're getting close so I claim and hope and usually believe to having intertile events working but we're not quite there yet will we really get them working if we do how fast will the tiles go will they deliver one air one AER one an average event rate of one event per site per second across a whole grid no matter how many the tiles are involved will we get to see those hundred plus tiles all put together and running code running code written not just by me but hopefully by other folks out there in our tiny rebel alliance as well to find out follow our progress here on T Tuesday updates welcome to year two