Hello out there!
I just asked myself, just like that: Why shouldn't it be possible to create
a php-work flow which allows the immediate parallelization over GPU to make,
for instance, the execution a hundredth times higher?
I mean, well - yes it sounds like a stupid idea, we've got 6 GPU-Horses (
servers with Radeon 5970 x2 ) and they could deliver blazing performance.
We could be a gazillion times be better in serving or processing content.
Right? even thought it's meant to be parallelization, we could rewrite the
code..
So, I'd thank you for constructive answers.
Cheers,
(c) Kenan Sulayman
Freelance Designer and Programmer
Life's Live Poetry
Do you have a patch for this? The only thing stopping it is no one had written it.
- Scott
Hello out there!
I just asked myself, just like that: Why shouldn't it be possible to create
a php-work flow which allows the immediate parallelization over GPU to make,
for instance, the execution a hundredth times higher?I mean, well - yes it sounds like a stupid idea, we've got 6 GPU-Horses (
servers with Radeon 5970 x2 ) and they could deliver blazing performance.
We could be a gazillion times be better in serving or processing content.
Right? even thought it's meant to be parallelization, we could rewrite the
code..So, I'd thank you for constructive answers.
Cheers,
(c) Kenan Sulayman
Freelance Designer and ProgrammerLife's Live Poetry
Hi:
Do you have a patch for this? The only thing stopping it is no one had written it.
So, I'd thank you for constructive answers.
sigh He was asking for constructive answers.
Kenan, I guess the most viable approach to go in such a direction is to look at something like HipHop[1] or any other PHP compilation approach.
The problem is to find an appropriate workload for the GPU.
There is high overhead involved in starting up code on a GPU and typical PHP workloads aren't a good match at all. At least if you stay in the Web app/DB-bound application area.
Best regards
Stefan
[1] https://github.com/facebook/hiphop-php
[2] http://www.phpcompiler.org/
--
Stefan Marr
Software Languages Lab
Vrije Universiteit Brussel
Pleinlaan 2 / B-1050 Brussels / Belgium
http://soft.vub.ac.be/~smarr
Phone: +32 2 629 2974
Fax: +32 2 629 3525
Hi:
Do you have a patch for this? The only thing stopping it is no one had written it.
So, I'd thank you for constructive answers.
sigh He was asking for constructive answers.
What's not constructive about it? Seems pretty constructive to me. The original poster has done research into the issue already.
The reason most things don't exist is that no one has a patch for it. If it doesn't core in core then it heads into a PECL module or lives as a patch on some site.
- Scott
Okay, I got it :)
I know Hip-Hop, but I mean in general; Most of the stuff here is done the
CLI-way.. and okay, I accept Scott's answer ;)
Thanks,
(c) Kenan Sulayman
Freelance Designer and Programmer
Life's Live Poetry
Hi:
Do you have a patch for this? The only thing stopping it is no one had
written it.So, I'd thank you for constructive answers.
sigh He was asking for constructive answers.
What's not constructive about it? Seems pretty constructive to me. The
original poster has done research into the issue already.The reason most things don't exist is that no one has a patch for it. If it
doesn't core in core then it heads into a PECL module or lives as a patch on
some site.
- Scott
-----Original Message-----
From: Scott MacVicar [mailto:scott@macvicar.net]
Sent: Friday, November 12, 2010 8:22 AM
To: Stefan Marr
Cc: Kenan Sulayman; PHP internals
Subject: Re: [PHP-DEV] GPU AccelerationWhat's not constructive about it? Seems pretty constructive to me. The original
poster has done research into the issue already.
What research are you referring to? I must have missed that.
There's no magic with GPUs. There are good reasons why few applications/infrastructure components use them. GPUs can be useful for very specific workloads but not necessarily general purpose app server.
Btw there are also some workloads that could benefit from leveraging some of the x86 extensions around encryption, vector computation, ...
So it boils down to being more specific re: where you think it could benefit and why.
Andi
The reason most things don't exist is that no one has a patch for it. If it doesn't
core in core then it heads into a PECL module or lives as a patch on some site.