[Beowulf] Roadrunner shutdown

[Lux, Jim (337C)]

On 4/4/13 8:39 AM, "Max R. Dechantsreiter" <max at performancejones.com>
wrote:

>Vincent,
>
>> It doesn't matter whether you code for blue gene, cuda or phi - from
>> a software viewpoint it's all vector type coding you've got to do.
>> the price of 1 coder is total peanuts compared to
>> the price of those supercomputers. So specialistic written software
>> is what you need anyway.
>
>Simply porting an application shouldn't take much effort,
>as long as vendor-specific libraries aren't involved.
>However, special and sometimes intensive efforts are often
>required to achieve good performance (high utilization).
>
>The unfortunate fact is that funds are more easily spent
>on hardware than on the human resources needed to utilize
>them effectively.
>


I would think that the problem is more that you can easily stamp out
another 1000 processors than another 10 software developers.  HPC
developers are the scarce commodity, and just throwing money at it doesn't
solve the problem.

That is why the real challenge for HPC is in developing smart compilers
and tools that make it easier to do HPC, even at the cost of needing more
computational resources.

I can buy a lot of computers and kilowatt hours of electricity for the
cost of one developer for a year.  Sometimes, I think that the frenzy of
"quest for efficiency" is sort of misdirected.  I'm not sure that it's
really a problem (yes, in the abstract, high efficiency is good, but in
the big picture, I'm not sure it is).

Just back of the enveloping here..
Let's say a "node" costs about $3k plus about $1500 in operating costs
over 3 years. Make it a round $5000 all told. A developer costs about
$250-300k, fully burdened for a year. So I can make the choice.. Buy
another 150 nodes or pay for the developer to make the processing more
efficient.  Of course, if I buy the nodes, I get the faster computation
today. If I buy the developer, I have to wait some time for my faster
solution.



>