[erlang-questions] Erlang and Larrabee CPU

Zvi <>
Sun Apr 5 10:41:48 CEST 2009


buzzwords, buzzwords...
I think with all the multicore buzz (partially attributed to Yariv's blog
and Joe's book), people mixing different kind of parallelism:

1. Task-level parallelism
     Irregular concurrency - applications that need to derive parallelism
from disparate concurrent tasks.
2. Data-parallel 
     Concurrency more apporpriate for synchronized parallel execution (like
in GPU/CUDA).

Erlang's actor-based model is more apporpiate for #1. 
While bad scalar floating point performance and absence of built-in
homogenous non-sequential data structures (vectors, matrices) doesn't help
for #2.
Erlang still can be used for controlling various accelerator devices, like
Larabee and GPUs.

Your are right, trhat LRB cores are essentialy general purpouse Pentium
cores with extended x86-64 IA with 512-bit wide SIMD and hyperthreading.
But initially it will be used as a GPU. With DirectX/OpenGL/CUDA/whatever
drivers emulating GPU.
I don't know in how many years we'll be able to run regular C/C++ code
natively on it, yet Erlang ;-)

Zvi




G.S.-2 wrote:
> 
> Those are good points Tim, I've also came over the following while reading
> up on Larrabee:
> 
> "A different version of Larrabee might sit in motherboard CPU sockets
> using
> QuickPath
> <http://en.wikipedia.org/wiki/Intel_QuickPath_Interconnect><http://en.wikipedia.org/wiki/Larrabee_%28GPU%29#cite_note-9>,
> but Intel has not yet announced plans for this"
> 
> The following article
> http://arstechnica.com/hardware/news/2007/06/clearing-up-the-confusion-over-intels-larrabee-part-ii.ars
> also aludes to the possible future works of something like larrabee
> running
> directly in a CPU socket. By the way, at the end the mentioned "Gesher"
> cpu
> is currently known as "Sandy Bridges" (6-8 core, should be released this
> or
> next year).
> 
> In any case, I just think that Erlang has a lot of potential for some
> hardcore computing (I primarily use it for computationally intensive
> projects on quad cores, and it's been working very well), and that we have
> not yet utilized Tesla and similar potential "processing force
> multipliers"
> is sad since Erlang can really scale on those easier than other languages.
> 
> On Sat, Apr 4, 2009 at 3:23 PM, Timothy Baldridge
> <>wrote:
> 
>> Right, although at launch, the Larrabee will be unable to run Erlang.
>> IIRC, at launch Larrabee will require a program running in the CPU to
>> hand it the instructions to run. This basically means that the Erlang
>> VM would have to be rewritten to support Larrabee. As Larrabee does
>> not support interrupts, and other hardware communication instructions,
>> Erlang would have to make sure that all OS processes are kept in the
>> conventional CPU cores, while the Larrabee is only handed the
>> processes that only manipulate memory. On top of that, Larabee
>> implementions will most likely not have direct access to the main CPU
>> memory, so any process data would need to be transported to the GPU
>> memory.
>>
>> Not impossible, but it still won't work out of the box.
>>
>> Timothy
>>
>>
>> On Sat, Apr 4, 2009 at 4:04 PM, G.S. <> wrote:
>> > The fact that Larrabee has 64 general purpose (relatively speaking)
>> Cores,
>> > and one can utilize them for processing is a benefit in itself. It's
>> much
>> > more general than the Nvidia's Tesla, and unlike cell architecture, all
>> > Cores are the same.
>> > After all, this is what Erlang is all about, concurrent high throughput
>> > computing.
>> >
>> >
>> > On Sat, Apr 4, 2009 at 1:53 PM, Kenneth Lundin
>> <
>> >
>> > wrote:
>> >>
>> >> The Erlang VM needs an operating system to run on.
>> >> Is there any OS that runs on the Larrabee?
>> >>
>> >> I have never heard of anyone running Erlang on the Larrabee and we
>> >> have for sure never tried it and I don't really
>> >> understand why that would be very interesting.
>> >>
>> >> Erlang can utilize a CPU with many general purpose cores or maybe act
>> >> as a controller running still running on geneal purpose cores but
>> >> administering jobs to be run on other special purpose cores.
>> >> The extra instructions available on Larrabee is nothing the current
>> >> Erlang VM can make benefit of.
>> >>
>> >> /Kenneth Erlang/OTP, Ericsson
>> >>
>> >> On Sat, Apr 4, 2009 at 6:32 PM, G.S. <>
>> wrote:
>> >> > Hello everyone,
>> >> >
>> >> > Does the Erlang community know by any chance whether Erlang will run
>> on
>> >> > the
>> >> > Larrabbee cpu, and will be able to utilize all the cores properly,
>> >> > compile...?
>> >> > Larrabee is MIMD as you guys know, and so would be perfect for
>> Erlang.
>> >> >
>> >> > Regards,
>> >> > -Gene
>> >> >
>> >> > _______________________________________________
>> >> > erlang-questions mailing list
>> >> > 
>> >> > http://www.erlang.org/mailman/listinfo/erlang-questions
>> >> >
>> >
>> >
>> > _______________________________________________
>> > erlang-questions mailing list
>> > 
>> > http://www.erlang.org/mailman/listinfo/erlang-questions
>> >
>>
>>
>>
>> --
>> “One of the main causes of the fall of the Roman Empire was
>> that–lacking zero–they had no way to indicate successful termination
>> of their C programs.”
>> (Robert Firth)
>>
> 
> _______________________________________________
> erlang-questions mailing list
> 
> http://www.erlang.org/mailman/listinfo/erlang-questions
> 

-- 
View this message in context: http://www.nabble.com/Erlang-and-Larrabee-CPU-tp22885451p22891627.html
Sent from the Erlang Questions mailing list archive at Nabble.com.




More information about the erlang-questions mailing list