19:01:53 From Ron Schreiner : You look very comfy Scott :) 19:05:56 From Paul Rak : we can talk offline 19:06:04 From Ron Schreiner : Pay no attention to the bouncing man. Argonne's centrifuge is just a little out of balance or something like that. 19:06:34 From munch : I know the IT crowd has been hybrid the last few sessions 19:08:06 From Ron Schreiner : I'm in the final Lab for IBM's Quantum Challenge 2023. I have 2 more exercises to complete. 19:19:45 From Ron Schreiner : Where does the Handford Site in WA state fall ? 19:19:59 From Neil Ormos : When it's time for questions for the presenter: In the relatively early days of non-monolithic supercomputers, the nodes were relatively small, and results had to be shared among nodes, so a key factor of performance was the interconnects among nodes. What are the interconnects used in modern supercomputers? How fast are the interconnects? 19:22:19 From Ron Schreiner : A100's being NVIDIA GPUs ? 19:23:28 From Paul Rak : https://www.alcf.anl.gov/polaris 19:27:23 From Michael Potter : Question: Why do they retire the machine? Why not continue to us it? 19:28:02 From Paul Rak : https://www.hpe.com/us/en/compute/hpc/slingshot-interconnect.html 19:33:02 From Ron Schreiner : So no 50K toilet seats ..... nice 19:33:39 From munch : Given the heat envelope of a machine like this, does it feed into any kind of district heating system in the winter? 19:33:50 From Michael Potter : Can you show some calculations on that? Seems strange that a 3YO machine would be that inefficient. 19:43:36 From munch : Sounds like the interconnect is somewhat analogous to the physical and data link layers out here in the quotidian computing world? 19:45:52 From Scott Nemec : Questions can be typed in here on Chat 19:46:00 From YDY : Does each new system like the Aurora include a whole brand new HVAC system? 19:47:11 From munch : Quotidian - day to day, like enterprise support 19:49:11 From Neil Ormos : When it's time for questions for the presenter: In the relatively early days of non-monolithic supercomputers, the nodes were relatively small, and results had to be shared among nodes, so a key factor of performance was the interconnects among nodes. What are the interconnects used in modern supercomputers? How fast are the interconnects? 19:51:07 From Michael Potter : how can it drop it in 5ns. That is only 5 feet. 19:52:20 From Michael Potter : 5 feet at light speed. 19:53:24 From Michael Potter : maybe it is 5ns in one rack. 19:58:54 From Ron Schreiner : Any future plans for Quantum Computing hardware ? 20:00:15 From YDY : What major revelations have you learned over the years from each revision of the supercomputers. Things that surprised you/everyone that wasn't known before and maybe changed the trajectory of future implementations? 20:05:38 From Ron Schreiner : Which Jetson ? 20:05:58 From Ron Schreiner : Oh looks like one of the bigger ones 20:21:59 From Salvatore Saieva : How do scientists typically get large datasets into the computing environment? 20:24:26 From munch : That 30% sounds like a quantified case of murphy's law 20:28:15 From munch : The 'Lessons Learned' stage of a project 20:31:04 From Michael Potter : where could we find a list of projects that were run? 20:31:12 From Michael Potter : Which ones are open source? 20:38:32 From Paul Rak : https://www.silicon.co.uk/workspace/ibm-us-pulls-plug-on-federal-supercomputer-project-36548 20:40:44 From munch : So, the sunspot nodes are the phase 1 you discussed earlier? 20:42:30 From Michael Potter : What are you perception of using the Apple M chips in super computers? they are very power efficient but given they are RISC does that limit them running in parallel 20:44:23 From Neil Ormos : How sensitive is the software to the organization of the machine and the various hierarchies of processors, GPUs, and interconnects? Are there problems that perform well on the old machines but perform poorly on the new machine until they are rewritten? 20:45:47 From Neil Ormos : Also, how many GFLOPs/W does the new machine provide, and how much better is it than the machines ALCF has used in about the last decade? 20:47:52 From munch : So, that baby version would be phase one or two of the scale-up 20:50:00 From Michael Potter : Regarding “everyone is RISC” Please describe how the intel instruction set chips are “RISC”. 20:50:20 From Michael Potter : s/everyone/everything/ 20:54:07 From Michael Potter : I did not have problem with it. 20:54:34 From Michael Potter : what is an example of a CISC processor then? 20:55:02 From Michael Potter : Is the under the covers RISC instruction set documented? 20:55:23 From Salvatore Saieva : Isn’t x86 considered CISC (Complex Instruction Set Computer)? 20:57:30 From Philip Rynes : He is referring to microcode as being RISC 20:57:46 From Michael Potter : https://stackoverflow.com/questions/13071221/is-x86-risc-or-cisc 20:58:49 From Michael Potter : CISC instruction set built on a RISC core. 20:59:07 From munch : I suspect many of us had an early experience with dennard scaling when we had to turn off the turbo switch to play games 20:59:24 From munch : or, games from the previous generation of hardware 20:59:55 From Salvatore Saieva : Reacted to "He is referring to m…" with 👍 21:00:00 From Salvatore Saieva : Reacted to "CISC instruction set…" with 👍 21:03:36 From Ron Schreiner : History is full of examples where hardware advances came with out the software required to fully exploite the new performance. 21:06:31 From YDY : Why don't you centralize power and remove the powersupply from the nodes? 21:08:28 From munch : Politics is the eighth layer of the OSI model 21:19:43 From Salvatore Saieva : Thank you, your presentation was interesting and enjoyable. Good night all. 21:20:17 From munch : Is 77 still the most current version of FORTRAN?
 21:21:33 From munch : Feel free to laugh if it isn't 21:21:34 From Ron Schreiner : I once worked on a team building a C vectorizing compiler written in Fortran ... don't ask. 21:21:39 From Michael Potter : It is a good thing that it is stable. The developers are destroying Java by adding all kinds of crap instead of just fixing security issues. 21:22:40 From Michael Potter : can you make a transpolar that looks kinda like Java but transpires to Fortran? 21:23:00 From Philip Rynes : There is an old program, 1980’s, to convert FORTRAN to c 21:24:10 From Ron Schreiner : Doesn't gcc support Fortran as a front end ? 21:24:51 From Michael Potter : yes. gcc compile fortran. It will soon also compile cobol. 21:26:55 From Michael Potter : This group to watch a presentation about paint drying and ask questions. 21:27:16 From Michael Potter : we are known to ask questions. 21:27:50 From Michael Potter : nite people. 21:28:59 From Ron Schreiner : If you can get gcc to stop after building the AST, you could extrude the AST to a different language. 21:30:03 From Neil Ormos : What is it about FORTRAN that makes it more suitable than other languages for parallel computation? Is it just the libraries already written that are cluster-aware? 21:32:20 From Neil Ormos : I had a question. 21:33:32 From Neil Ormos : So it imposes a discipline that make the code easier to parallelize? 21:33:56 From Ron Schreiner : That's exactly what I worked on, building a C complier that could unroll a loop and turn it into a single vector instruction. 21:34:11 From Ron Schreiner : Fortran loops are dirt simple 21:34:26 From Ron Schreiner : C loop a different story 21:34:27 From Carey Schug/USA:IL:Chicago area : telling college kids that fortran is the language of cutting edge super comnputers des not get them interested? 21:36:16 From Carey Schug/USA:IL:Chicago area : fortran 77? I started with fortran II 21:36:56 From Ron Schreiner : You could transpile Fortran to something else 21:37:11 From Ron Schreiner : AST Abstract Syntax Tree 21:38:37 From Carey Schug/USA:IL:Chicago area : wuerdityon 21:38:55 From Carey Schug/USA:IL:Chicago area : maske a C-- that is C++ with additional restrictions to make it more fortran like? 21:39:19 From Ron Schreiner : Call it C dum dum 21:39:47 From Neil Ormos : Thank you, Bill. Very interesting.