Thought this was interesting. Not so much for all the cost of it but the NVLink they talk about would be a great advance for all of us.


http://www.digitaltrends.com/computing/obama-signs-executive-order-paving-the-way-for-an-exaflop-supercomputer/

---

Not to get too political, but the government is already way too huge. Has its fingers in way too much. What does this require an executive order? Executive orders are a way to bypass the other branches of government and I know all presidents do them.

Not of a fan of big government, so I guess I will stop here and just let the machine roll on.

Bob

---

Sometimes I wonder, what happened to all the people I gave directions to?
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
SETI@home classic workunits 4,321
SETI@home classic CPU time 22,169 hours

This is a political issue. In the top500 list, updated twice a year. China has taken the lead. Now they are installing aluminium panels on the 500 meters FAST radiotelescope, bigger than Arecibo, and it will have a new supercomputer too.So the US Gov is trying to catch up with the Joneses...
Tullio

---

I guess Moore's law really doesn't matter for these supercomputers if you have a big enough room. ;~)

---

The mind is a weird and mysterious place

What you need is cooling system for thousands of CPU and GPU. Of course nVidia will provide the GPUs and Intel the CPUs.
Tullio

---

The large super computers in the world can compute perhaps 30 times faster than all of the seti@home computers. Currently a GPU can process on the order of 1 tera FLOPS; seti@home can command about 700 teraflops but might be working on 100,000 or more separate work units.

This suggests that prescreening of work units as they roll in off of the world's largest radio telescope antennas could be accomplished by a low cost applique to the antenna---reserving only the most promising scans for an extended computational analysis for purposive content.

Perhaps the new-found money could be directed towards revamping current SETI efforts under a savvy project manager who would dole out the funds wisely.

As I said before: " who will call the meeting" ?

ON topic:
an EXO-FLOP is 1000 Peta FLOPS (quadrillion floating point instructions per second)
or if you like. 1,000,000,000,000,000,000 floating point operations per second.

Whether you like it or don't like it, I posted this to show the promise of new technology that could affect how we process the data.

Any development that accelerates communication between the CPU and GPU would be a boom for all of us.

And yes, with all the new data that will come with this new initiative, they are going to looking at new ways to analyze all that data.

---

More computers and especially more large computers means more garbage in and more unncessary results back.

Even the most powerful supercomputer is limited to the most powerful motherboad currently available. Without checking any further, this could be something like a dual-CPU motherboard with two Intel 5960-X processors or the similar being installed.

Next comes the power supply, cooling fans, graphics for viewing and other external components like disc drives or drive arrays, making such things even more complex and expensive.

Because of this the whole concept behind the whole idea of such a computer system, meaning the architecture behind it needs to be considered. You probably will always be left with one empty box not being used in even such a system.

Of course I do not reject the whole idea or concept, but when dealing with supercomputers some clever thinking is needed in order to get the best results.

In the same way I have always been wondering why there are not any, or at least many supercomputers contributing to distributed projects like BOINC for scientific computing.

Possibly the owners of such mainframe computers have found it best to leave such projects to the users of personal computers instead, rather than becoming directly involved in such projects.

China's TianHe2 has reached 54 petaflops. But reaching exaflops won't b easy, because of cooling problems. Quantum computers anyone?
Tullio

---

What you need is cooling system for thousands of CPU and GPU. Of course nVidia will provide the GPUs and Intel the CPUs.
Tullio

Actually an APU with integrated CPU/GPU cores makes a lot of sense for a machine this big. The HSA unified address space also makes a lot of sense, an AMD technology. And AMD are already using HBM. AMD's position wrt exascale is interesting.

http://www.hpcwire.com/2015/07/29/amds-exascale-strategy-hinges-on-heterogeneity/
http://wccftech.com/amd-exascale-heterogeneous-processor-ehp-apu-32-zen-cores-hbm2/

I am an AMD user myself, with a Opteron 1210, E-450 and A10-6700, but with exception of IBM with its Power CPUs all main entrants in the top500 list seem to use Intel CPUs. There are also a few Sparcs from Oracle (former SUN).
Tullio

---

I think it's a mater of prestige to be a exa-flop producer...so nVidia might make a cent on a deal, but on commercial scale it SHALL BOOM!

now it's time on AMD/ATi... ;)

---

non-profit org. Play4Life in Zagreb, Croatia, EU