[TUHS] Re: Reaction to the 3B2 at Bell Labs

On Sat, Nov 26, 2022 at 06:47:52PM -0500, Dan Cross wrote: I read the Wikipedia page on the 9000. It's sad that the 9000 wasn't cancelled when they had better alternatives. It reminds me of a CPU I got cancelled at SGI. It was in the mid 1990's and the project was code named "Beast". It could do tons of data movement, to make that happen, the packaging had pin recepters on all sides, top, bottom, and all 4 sides. The packaging, at volume, was going to cost $1,200 each. No CPU, just the packaging. On top of that, SGI did a flip/flop design cycle, flip focussed on integer performance, databases, while flop focussed on floating point and catered to the super computing market. This was when almost all of the processor architects were talking to me because I had written LMbench which was a set of microbenchmarks that measured bandwidth and latency of everything. The architects loved them because they were small and could be run on a simulator. So I had a pretty good idea what Sun was coming up with (I had left Sun for SGI), knew what Intel was doing, HP yep, DEC less so but had some idea about the Alpha roadmap, IBM was the one place that I didn't have good intel. I looked at what Beast was claiming, looked at the past predictions of MIPS chips and when they shipped vs when they would claim to ship, and looked at the exploding internet / database market and gulped. Beast was the wrong answer and we were gonna get crushed. I started shopping my theory around, eventually got a meeting with Wei Yen (who was someone high enough up). I went over all the info I had, Wei Yen was super impressed and asked me if this was my job and I said "oh, no, this is just a hobby I play around with" and he replied "Keep playing, you play nicely" and went off and cancelled Beast. DEC needed someone with that sort of data. I had actual performance results from every processor in the mainstream and they were no bullshit results, anyone could reproduce them. Then I had some idea of almost everyone's roadmap. Without the data, they would have wasted a ton of money on Beast. In the end, it didn't really matter, x86 killed SGI just like Sun and IBM killed DEC. And x86 killed Sun as well. -- --- Larry McVoy Retired to fishing http://www.mcvoy.com/lm/boat