uselton@nas.nasa.gov (Samuel P. Uselton) (05/15/91)
In article <268@rins.ryukoku.ac.jp> will@rins.ryukoku.ac.jp (will) writes: >In article <1019.282B28FD@nwark.fidonet.org>, Samuel.P..Uselton@p0.f13.n391.z1.fidonet.org (Samuel P. Uselton) writes: >> The NAS project at NASA Ames regards pushing industry into producing >> a teraflops computer by the year 2000 as a "Grand Challenge" problem. >> It'll be quite a while longer before that capacity finds its way >> into workstations for the broad market. >> > > From what I have seen and heard (from the people doing this research) > this technology should be availible to the government and big > industries by 1997 (and maybe by 1995) and into the home a few years > later. The NAS plan specifies tera-FLOPS sustained performance on particular real problems (Fluid Dynamics Simulations and other Aeronautical applications). The core of this work is inversion of large sparse matrices (standard heavy duty scientific application) where large means on the order of 10**6 rows (and 10**6 columns). *PEAK* performance of tera-FLOPS by 1997 I believe. Maybe even 1995. Sustained performance on real problems is going to take several years of experience and several more of software development (to learn how to maximize use of highly parallel machines). We got one of the two Touchstone Gamma prototype machines, (like Intel's current iPSC2/860) with 128 processors. These processors have high peak ratings, but between needing a bigger cache (for OUR application) and needing a better compiler (when it was delivered - over a year ago now), intial sustained performance figures were dissappointing to those who had been looking at the theoretical peak numbers. It is going to take years of that kind of experience to get what we need. > As I have been told, the biggest problem now is not making > teraflop computers, but manufacturing techniques must be updated for ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ This isn't a problem in "making" ? ( :-) ) > production and quality control standards must be updated for mass prod. ^^^^^^^^^^^^^^^^ Meaning the yield has to be brought up to a useful level, meaning that it is hairy edge right now.... ok. > The reason cited was that these computers use components that require > more advanced manufacturing technologies and that current facilities > must be redesigned to meet these requirements. > > A little side note: I was once told that another reason for the delay > of the teraflop technologies was that: > All of the Corporations that make computers and have these > technologies currently cannot just release it at this time even ^^^^ > if they could. The reason is "Economics". They and thier ^^^^^^^^^^^^^ (Which implies they can't.) > customers have invested billions in current technolgy. To make > it absolutly obsolete over night would destroy there companys. (Certainly true. In fact that had something to do with the demise of Denelcor makers of the HEP computers - at least according to some of their potential customers I spoke with at the time.) > Not to forget that many of these manufacturers have warehouses > full of new equipment ready to be sold. Worth billions. Not the warehouses full, so much as the factories ready to make them and not paid off yet. Replacing this equipment with new & better to make the smaller faster thing before the cost has been amortized over the predicted number of units, is where they lose big. > > This is one reason that companies do incremental scaling of > computer technologies. To get as much money as possible with > as small an investment as possible. It's all "Economics". > And that's why NASA announces "Grand Challenges" and timetables to meet them. By telling manufacturers to EXPECT something, it gives their planners info on how many units or how many years they will get from a generation of manufacturing equipment. > > > Will... Sam Uselton uselton@nas.nasa.gov employed by CSC working for NASA (Ames) speaking for myself
will@rins.ryukoku.ac.jp (will) (05/16/91)
In article <1991May14.170317.23649@nas.nasa.gov>, uselton@nas.nasa.gov (Samuel P. Uselton) writes: >Not the warehouses full, so much as the factories ready to make them and not >paid off yet. Replacing this equipment with new & better to make the smaller >faster thing before the cost has been amortized over the predicted number of >units, is where they lose big. > Thankyou for clarifing this. This is what I really wanted to say but could'nt get the right words. Will...