Will build about 125 of Mexican border wall or an Exacomputer.
I know which is most likely to be beneficial to the US.
Intel will, as expected, provide the processors for the US government's exascale-grade Aurora supercomputer, due to be deployed in 2021. The contract to build the 1,000 peta-FLOPS beast – that's a machine capable of crunching a quintillion floating-point math calculations per second – will run to $500m, with Chipzilla …
“Aurora and the next-generation of Exascale supercomputers will apply HPC and AI technologies to areas such as cancer research, climate modeling, and veterans’ health treatments.
Yeah, right - that climate modelling would be of changes that the current administration doesn't even believe are happening, presumably? I think we can count on this being used for quite different benefits to society, primarily "better" nukes.
Yeah, right - that climate modelling would be of changes that the current administration doesn't even believe are happening, presumably?
Fortunately Trump doesn't actually run the government. It's far too complex a beast for his idiocy to completely inundate it in the 4 years of his Presidency (the only way he gets a second term with his approval ratings is if the Democrats are complete idiots and nominate one of their lunatic fringe). Even if by some infernal miracle (or DNC stupidity) he gets a second term he STILL won't have enough time to completely take over all the stuff our government does. There are a lot of better heads between him and whoever gets to decide what simulations are run on the governments supercomputers.
For the record, I hold an equal amount of disgust for the GOP and the DNC. And slightly more for select fringe lunatics on both sides, of whom Trump is one.
" (the only way he gets a second term with his approval ratings is if the Democrats are complete idiots and nominate one of their lunatic fringe)"
Don't worry, Hillary has recently ruled out a comeback in order to facilitate a comeback by popular [ media ] demand, and will be installed to fulfil the Neocon Dream. They cannot permit disobedience as happened when voters went off-script in 2016 ever again.
"The MSM are the men that will not be blamed for nothing"
Worth remembering: the original Intel supercomputer was supposed to be delivered on the same schedule as the two Summit supercomputers. One of the contract requirements was that the same architecture could not win all three labs. The IBM/Nvidia/Mellanox group delivered both its machines, Intel did not.
Oh, and those IBM machines run on Power.
One of the other contract requirements was that everything in the solution had to be commercially available; the labs did not want to be out on an island with some unique configuration with no future and a rarity of skills.
So yeah, I know it's popular to rag on IBM, and I know I'll get a bunch of downvotes for saying this, but the reality is that IBM's coalition got it done, Intel's did not.
Mentioned more than once: "The super will be Uncle Sam's first publicly known exascale computer" - c'mon Chris - tell us more.
I can see that various US agencies have placed various contracts over the last 6 years or so - do the US already (probably) have an Exascale Super as your cryptic comment(s) suggests?
$500M? That's the true cost of maybe half a dozen F-35s. Possibly fewer as Lockheed has been under pressure to get the sticker prices down and players in the US Military Industrial Complex are nothing if not cognizant of government wishes. They'll recover any paper losses one way or another. That's the way the game is played.
Anyway, a supercomputer seems to me a better investment than a few copies of an extraordinarily expensive, and likely unnecessary fighter aircraft. And it's good to hear that Cray is still around in some form. Seymour Cray's mid and late 20th century computer designs (CDC 1604, 6600, 7600, etc) were really quite remarkable and were a joy to work with.
When at University, an assembly language course was required. The standard option was PDP-11, but we EEs were second to the CS students, whose department owned the PDP-11s. In practical terms, this meant third shift (12M - 8A) time slots...not an appealing option. However, due to the fact that the U had just purchased a CDC Cyber-74, which came with a 1 year deployment of an Applications Engineer, there was a second option: two sections of CDC Cyber assembly language, and I jumped at the chance.
Nominally, input/output was via punched cards. I did this once to say I'd done it, then used my personal Teletype for the rest of the assignments. Hardware floating point, 60 bit words, and someone who knew all the tricks to guide us. It was a wonderful experience.
// no blinkenlights icon?
“Aurora and the next-generation of Exascale supercomputers will apply HPC and AI technologies to areas such as cancer research, climate modeling, and veterans’ health treatments. The innovative advancements that will be made with Exascale will have an incredibly significant impact on our society.”
and although discovered and, at least partially, developed by US Gov the medicines will bankrupt many and just be completely affordable to many more.
the bandwidth and power consumption. . .
The top end of the top500 HPCs has always been (and remains) something of a technological pecker-measuring contest. The marginal efficiency (that is the Rmax/Rpeak) for the HP Conjugate gradient benchmark for the top 3 in the Top 500 is interesting. The point is that if you really need the horsepower, you build a real supercomputer with a high bi-directional bandwidth, like RIKEN.
So the key questions are: What is Exascale computing really going to be able to do, and who in the world might want to do that?
Hint: The current population of the US is approaching 330 Million. Do the math and see how many MegaFLOPs per man, woman and child an exaFLOP comes to.
Then consider the aggregate computing power that folks like Google, Microsoft, Facebook. . . the list is damned nigh boundless, are devoting to what they do.
What could possibly go wrong?
Biting the hand that feeds IT © 1998–2020