[Beowulf] Is there really a need for Exascale?

[Eugen Leitl]

On Fri, Nov 30, 2012 at 01:24:20AM -0500, Mark Hahn wrote:

>> The classical CPU core seems to be regressing to the role of a co-processor,
>> and real heavy lifting is being done with OpenCL, a la APU.
>
> my colleagues who attended SC12 say that OpenCL was almost entirely

That is unfortunate. Apple seems to like OpenCL, is there any
actual evidence Apple developers use it much? Ditto for mobile
game apps which use the GPU. Does anyone know?

> absent (along with AMD :( ).  it's not clear how generally applicable
> Cuda's SIMT programming model is.  and having it as a separate ISA
> (versus traditional cores) is a problem, complexity-wise.

In general CUDA seems to hide the hardware poorly, so
it probably doomed long-term.

>> GPU into every mobile device CPU. As it's memory starved, there's already
>> memory stacking http://www.ifixit.com/Teardown/Apple+A6+Teardown/10528/1?singlePage
>> I presume Exynos 5 teardown would not look much different.
>
> samsung has been producing stacked chips for quite a while - back to at 
> least the a5 generation, probably before that
> (stacked dram, stacked flash too I think.)

Yes, it's not new, but it makes things cheaper, and also improves
performance somewhat.

>> The interesting part about forthcoming real (TSV) stacking
>> that you can mix hybrid memories, whether SRAM, MRAM, DRAM, flash,
>> as you don't have to be process compatible as with true embedded
>> memory (though MRAM might well be process compatible with CPUs).
>
> stacking is great, but not that much different from MCMs, is it?

Real memory stacking a la TSV has smaller geometries, way more
wire density, lower power burn, and seems to boost memory bandwidth
by one order of magnitude

http://nepp.nasa.gov/workshops/etw2012/talks/Tuesday/T08_Dillon_Through_Silicon_Via.pdf