Re: Sibelius Software UK office shuts down

[George_]

Han-Wen Nienhuys-5 wrote:
> 
> On Mon, Aug 6, 2012 at 6:53 PM, Joseph Rushton Wakeling
> <address@hidden> wrote:
>> On 06/08/12 20:26, Han-Wen Nienhuys wrote:
>>>
>>> Also, going MT will give you a max 8x speedup (assuming perfect
>>> parallelization on an 8 core machine). That is not going to bring down
>>> processing costs to interactive rates.
>>
>>
>> I think you're focusing on the wrong kind of architecture.
> 
> I'm talking about the architecture of computers that people can buy in
> the shops today. While cute, a 192-way ARM server is useless in
> realistic scenarios. See eg.
> http://static.googleusercontent.com/external_content/untrusted_dlcp/research.google.com/pt-BR/us/pubs/archive/36448.pdf
> - aka. "Let's use 9 pregnant women, we'd have a baby within the
> month."
> 
> Unless you have a embarrassingly parallel problem to begin with (which
> music typesetting is not), lots of parallelism only buys you
> synchronization overhead, both lock contention at run-time, and the
> overhead of having to write race-condition-free parallel code.
> 
> Note that lilypond is embarassingly parallel at the file level, so for
> the regression test, we already distribute the files on as many CPUs
> as we have available.
> 
>> _This_ is the kind of setup that you should be aiming to exploit the
>> multithreaded possibilities of:
>> http://www.zdnet.com/boston-virdis-192-core-server-consumes-only-300-watts-of-datacenter-power-7000001654/
> 
> -- 
> Han-Wen Nienhuys - address@hidden - http://www.xs4all.nl/~hanwen
> 
> _______________________________________________
> lilypond-user mailing list
> address@hidden
> https://lists.gnu.org/mailman/listinfo/lilypond-user
> 
> 

The reason this is important is because while IPC goes up incrementally and
relatively slowly (IPC has done little more than double between 2005 [P4
660] and now [i7 3930X]) and clock speed is relatively stagnant (it's
unlikely we'll ever get 8GHz stock x86 CPUs the way Intel predicted), core
count is the only real way to dramatically improve performance - over a
similar period, core count has gone up six-fold (in high-end parts), and
it's set to continue. I agree, talking about a typesetting program running
on a 192-core ARM server is a bit silly, but then, so is saying that an
8-fold increase in speed won't make the process instantaneous, then implying
that for this reason we shouldn't look for ways to make it work.
-- 
View this message in context: 
http://old.nabble.com/Sibelius-Software-UK-office-shuts-down-tp34245636p34264057.html
Sent from the Gnu - Lilypond - User mailing list archive at Nabble.com.