[Pw_forum] Seeking Advice on Small Hardware Platforms for PWscf Implementation

[Nicola Marzari]

Hi Paul,

let us know if you get PWSCF on a PS3 !

I generally echo the other suggestions - the speed of the memory bus
is critical, having fast memory (matched with the bus) is good,
having the fastest clock on current Intel chip is good, having
two cores is a 25-50% improvement, and having 4 cores is
a negligible improvement on 2.

If you plan to link blades in a cluster, it's vital to have fast,
low latency communication - if you use gigabit, you might get good
parallelism with as many as 8 blades, or as little as 2, depending
on the quality of your switch, using a very recent mpi implementation,
and the quality of your ethernet controllers on board.

We have some tests on different platforms on
http://quasiamore.mit.edu/pmwiki/index.php?n=Main.CP90Timings


			nicola


Paul M. Grant wrote:
> To All Forum Members:
> 
> I’m planning on building a new Linux box (or boxes) to explore highly 
> correlated systems (e-p coupling plus LDA+U), and am seeking the 
> collective experience and advice of the PWscf community on a suitable, 
> inexpensive (< 2000 USD, MB+CPUs+RAM, exclusive of power supplies, 
> enclosures, and accessories) hardware platform.  I emphasize that the 
> principal purpose of this new box would be exploratory, or for 
> development, not production. 
> 
> I’ve built several past platforms, both Windows and Linux based, using 
> server boards manufactured by Supermicro and have had generally good 
> experience and service (the factory is only 15 miles from where I 
> live).  Currently, I use two machines with dual Xeon processors (single 
> core, 32 bits), one with 1 GHz cpus, 1 GB RAM, the other 2.4 GHz and 
> 3.25 GB RAM, both with bus speeds of 133 MHz, the newest 3 years old.  
> However, occasionally I run PWscf exercises on my little Thinkpad X41 
> tablet (single processor, 1.5 GHz, 1.5 GB), and the scf computation will 
> run 3-5 times faster than on the other machines!  I suspect this rather 
> surprising result is because the Thinkpad has a 400 MHz bus clock speed.
> 
> One option I’m considering is using a “gaming” or server class 
> motherboard with dual 2.33 GHz quad-core 64-bit processors, a 1333 MHz 
> FSB, and 16 GB RAM.  Having said this, I’m not sure PWscf (and the 
> Fortran compilers available) can handle all this parallelism efficiently 
> on a single motherboard.  I’ve noticed when running pw.x, the CPU 
> activity “flips” between processors every several seconds, instead of 
> sharing each at 90-100% full time.
> 
> On the other hand, one could consider building a small MPI-connected 
> cluster for about the same amount of money.
> 
> When IBM announced a couple of years ago the incredible performance 
> details about the Cell processor that would go into Playstation 3, I 
> thought, “Wow, maybe the future of computational physics rests with 
> gamers.”  I’m sure most of you know this is actually beginning to 
> happen, spurred on by the fact that the PS3 is “open architecture” and 
> can run a Linux distro.  Moreover, there apparently are “open software” 
> numerical analysis tools available from IBM.  At least four US 
> universities are experimenting with off-the-shelf PS3 clusters, perhaps 
> one of the more interesting is at UMass, 
> _http://gravity.phy.umassd.edu/ps3.html_.   In the last week or so, Sony 
> lowered the entry level price of the PS3 to 400 USD.  So, a cluster of 
> four with a cheap switch could be purchased for about the same price at 
> the single motherboard configuration I mentioned above.
> 
> My teenager, a gamer, tells me the PS3 has problems.  He says it’s 
> unreliable and overheats and only has 256 MB RAM on board (he owns a 
> Wii, which outsells the PS3 in the US by a factor of three).
> 
> Has anybody tried porting PWscf to a PS3?
> 
> Any and all advice is welcome.
> 

-- 
---------------------------------------------------------------------
Prof Nicola Marzari   Department of Materials Science and Engineering
13-5066   MIT   77 Massachusetts Avenue   Cambridge MA 02139-4307 USA
tel 617.4522758 fax 2586534 marzari at mit.edu http://quasiamore.mit.edu