IBM debuts 2nm device with 50 billion transistors (Page 3): And this type of computing power changes the way we do science. Rather than the classic hypothesis/experimental...

TriX Loc: Raleigh, NC

Doc Barry wrote:

Most specifically, a FLOPS means a floating point operation per second. PetaFLOPS = 10^15 floating point operation per second. Now that we are seeing exaFLOPS (10^18 floating point operation per second), I expect that we will achieve zettaFLOPS (10^21 floating point operation per second) perhaps by the end of this decade or the next.

And this type of computing power changes the way we do science. Rather than the classic hypothesis/experimental design/one-experiment-after-another model, we can model every possibility in near real time.

PaulBrit Loc: Merlin, Southern Oregon

I meant to say when I wrote: "who can’t image this processing speed” who can’t imagine this processing speed.

Doc Barry Loc: Huntsville, Alabama USA

TriX wrote:

And this type of computing power changes the way we do science. Rather than the classic hypothesis/experimental design/one-experiment-after-another model, we can model every possibility in near real time.

A real possibility TriX, but I fear even more massive code bloating. 😁

TriX Loc: Raleigh, NC

Doc Barry wrote:

A real possibility TriX, but I fear even more massive code bloating. 😁

Yep, it’s a way of life. Except when you get a paradigm shift such as moving from HDs to SSDs, aps rarely get faster because as HW speed increases, the SW designers write larger code to take advantage of it for more features. With high performance computing, where multiple users are scheduling the machine, there is an incentive to write tight code because the time is so expensive. These machines often have hundreds of thousands or millions of cores and cost hundreds of millions of dollars (if not billions) and consume enough power and A/C to require a substation. Their lifespan is rarely more than 5 years before they become obsolete, so the cost per minute to operate is quite high. The main uses are particle/quantum physics, nuclear weapons modeling (since there is no more testing), weather modeling and drug/genetics modeling

TheShoe Loc: Lacey, WA

wham121736 wrote:

For the uninitiated a petaflop is a computing speed of one thousand million million mathematical operations per second!

Isn't the flop specifically a floating point operation? They are inherently slower than the addition (including subtraction) and multiplication (including division) operations, making a petaflop a much loftier goal.

Doc Barry Loc: Huntsville, Alabama USA

TheShoe wrote:

Isn't the flop specifically a floating point operation? They are inherently slower than the addition (including subtraction) and multiplication (including division) operations, making a petaflop a much loftier goal.

It is flops or FLOPS. A FLOP or flop is the second state of a flip-flop. 🤔😉. Sorry, but could resist.

TriX Loc: Raleigh, NC

If high performance computing interests you, the Top 500 list is interesting reading and changes monthly. As of November, Japan has the fastest machine at ~500 PFlops, >7 million cores and 30 megawatts of power. That will change late this year when the Argonne, Oak Ridge and LLNL Exaflop machines come on line. https://www.top500.org/lists/top500/2020/11/

Doc Barry Loc: Huntsville, Alabama USA

TriX wrote:

If high performance computing interests you, the Top 500 list is interesting reading and changes monthly. As of November, Japan has the fastest machine at ~500 PFlops, >7 million cores and 30 megawatts of power. That will change late this year when the Argonne, Oak Ridge and LLNL Exaflop machines come on line. https://www.top500.org/lists/top500/2020/11/

Thanks. Interesting reading.

waldron7 Loc: State of Confusion

All from a crashed saucer at Roswell

TriX Loc: Raleigh, NC

Here’s the storage for the 1 PFlop Machine at Argonne (Mira). 16,600 drives in 18 racks (all inserted by us) and 350 GBytes/sec sustained I/O performance (and that’s now 10 year old technology)

Doc Barry Loc: Huntsville, Alabama USA

TriX wrote:

Here’s the storage for the 1 PFlop Machine at Argonne (Mira). 16,600 drives in 18 racks (all inserted by us) and 350 GBytes/sec sustained I/O performance (and that’s now 10 year old technology)

Beautiful equipment TriX. Thanks for sharing this. When I was a teenager in the late 50s, the Regional Engineering Director for IBM lived across the street from us. He had an extra detached garage he used for storing "junk" parts from the IBM computers. He of course gave me whatever I wanted to play with and he taught me a lot. Computers then were tube-based and the core memory was exactly that ... tiny cores with read, write, sense coils wrapped around then. He said that young women made these by hand because machines could not do it as well. Never understood why, but each core was large enough to see the windings without using a magnifying glass. Advances came quickly after that. A decade later or so, I was interviewing with Dr. Cray and marveling at the incredible achievements happening at CDC. For several reasons, I took a job at TI rather than the one he offered me at CDC. He was a brilliant man! You are a fortunate person TriX to have (and I guess continue to be) been involved with these hyper-super machines!

TriX Loc: Raleigh, NC

Doc Barry wrote:

Beautiful equipment TriX. Thanks for sharing this.... (show quote)

I’m retired now but was fortunate enough to see the industry from the beginning (I was building flip flops from 12AT7 vacuum tubes when I was 14) and still keep up with old friends in the industry. You were lucky to have met Dr. Cray. For awhile, it looked like IBM’s Blue Genes would be the cluster of choice for HPC (that is a Blue Gene in the Argonne photo and the DataDirect storage), but Cray has come roaring back and is the prime for two of the ExaFlop machines I mentioned. I’ve worked in all types of computing, but the 8 years I spent in HPC were, by far the most interesting with the most talented and nicest people to work with.

RS Loc: W Columbia, SC

All of this sure brought back a few memories to me, too. Sorry, but my 'rememberizer' doesn't work too well these days; but back in my early days in the Army, I repaired IBM and UNIVAC computers(?) - 'card processors'. I did so during my trip to Vietnam in '67 - '68, then at Ft Hood for a few years. Being in the repair field, we also had to know the jobs of 'programmers' and 'operators', so we could point out that they often were doing something wrong as opposed to there being a mechanical problem with the machines. I left that field of work when the chance came along, and I didn't regret it.

Doc Barry Loc: Huntsville, Alabama USA

RS wrote:

All of this sure brought back a few memories to me... (show quote)

I still have a box of unused punch cards. Sometimes I give a few away to folks who have never seen them. 😃

burkphoto Loc: High Point, NC

Doc Barry wrote:

I still have a box of unused punch cards. Sometimes I give a few away to folks who have never seen them. 😃

Thankfully, we have evolved from those evil little annoyances.

When I was in college in the mid-1970s, Davidson had an NCR Century 100 with 64K of memory. Input was always from punch cards. Output was always to line printer. Storage? What storage? We kept our stacks of cards in order!

I thought I'd reached nirvana when I saw my first Apple II!