----- Forwarded message from Kragen Javier Sitaker <[email protected]> -----
From: Kragen Javier Sitaker <[email protected]> To: Jamie McCarthy <[email protected]> Subject: Re: Could a vacuum tube computer be fast? On Thu, Jul 22, 2010 at 07:25:48AM -0400, Jamie McCarthy wrote: > > I'm proposing they should have built *smaller* ones in > > which the tubes switched more often, and I don't know why they > > didn't. > > Hardware flaws? It was my understanding that tube reliability was > *the* major flaw in vacuum-tube computers. Hardware failures in > Eniac's 17,000 tubes gave it uptimes measured in hours. That would have militated against running the tubes so slowly and using so many of them. If you can use half the tubes for the same computational power, then the machine will run for twice as long. This was one of the major advantages of the LGP-30, with 113 vacuum tubes, over the ENIAC with its 17 000. (It's not quite a fair comparison. Stan Frankel, the LGP-30's designer, had the advantage of experience programming the ENIAC, and more than 1400 semiconductor diodes.) Also, if you can get more computational power out of the same tubes, your uptime won't change, but you'll be able to complete bigger calculations before the next crash. All of the subsequent tube machines were an order of magnitude smaller than ENIAC. ENIAC was just badly designed, that's all. > Economic pressures? What were the bottlenecks of the era? Did they > have many programs that were complex enough that runtime was the most > important consideration? Both runtime and memory size were major limiting factors. It wasn't until the late 1950s that the software problem really became known. > Heat dissipation problems? How did the wattage per liter, and energy > efficiency, change with smaller tubes? Tubes emit most of their heat > as radiation I would imagine, so the modern solution (conduct heat > away and force convection) might not be an effective offset as > densities increase. That's an interesting question, and Dave Long brought it up on kragen-discuss. Do you mind if I forward your email there, too? Kragen ----- End forwarded message ----- -- To unsubscribe: http://lists.canonical.org/mailman/listinfo/kragen-discuss
