Re: GPU-based Compute Cloud

Here is another step to GPU cloud:

Acceleware released the world's first commercially available GPU-based cluster solution, the C30-16 (16 GPUs expandable to 64 NVidia Tesla GPUs).

quote:

The GPU cluster solution combined with Acceleware's computational algorithms was benchmarked with data provided by end users, confirming processing power capable of solving problems billions of cells in size with speeds approaching 14 Gigacells per second for electronic design customers. At these processing speeds, Acceleware's C30-16 is competitive against traditional, more expensive clusters of around 1000 cores. For seismic customers using techniques like Reverse Time Migration, problem sets in the range of billions of cells over thousands of square kilometers found in large marine surveys will now be commercially possible with the Acceleware solution.

source:
http://www.acceleware.com/newsEvents/newsreleasearchive/20080617clustersolution.cfm

salut,
Khaz Sapenov

Re: GPU-based Compute Cloud

I believe there will be clouds that offer specialized services that
take advantage of a certain processor or IO architecture and charge a
premium for it. My guess is that CDN vendors will look at pushing the
application to the "edge" along with the data. Whatever standard
shape the cloud takes (think EC2 vs. GAE) being able to push it around
the globe will be a value added feature of a particular cloud.

On Jun 14, 7:37 pm, Paul Moen <paulm...@gmail.com> wrote:
> Don't underestimate the power of specialization in driving compute
> clouds for specific compute tasks.
>
> It is not just GPUs, Kickfire (www.kickfire.com) have released a
> so-called SQL chip (DPU or SQLPU)?
>
> I see some reasons for software vendors looking at hardware, one is, if
> your software is becoming a commodity you need to move up or down the
> stack. So why not use FPGAs from Altera or Xilinx to test run burning
> your software into hardware.
>
> The other reason is the recent super computing record using the IBM cell
> chip which is used in current game consoles. Essentially people are
> rediscovering the power of parallelization and also specialization that
> game console makers have known for ages. The current PC and Server CPUs
> are made to be good at many tasks, the future may involve increased
> specialization of hardware for specific tasks.
>
> If you were are Database as a service provider and had a stack of
> kickfire applicances, suddenly given the performance gains of running
> part of the database in hardware means you can host more databases per
> machine per cubic feet (or cubic metre) having a real impact your bottom
> line.
>
> Have Fun
>
> Paul
>
> http://blog.dbadojo.comhttp://blog.vmdatamine.com

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "Cloud Computing" group.
To post to this group, send email to cloud-computing@googlegroups.com
To unsubscribe from this group, send email to cloud-computing-unsubscribe@googlegroups.com
For more options, visit this group at http://groups.google.ca/group/cloud-computing?hl=en
-~----------~----~----~----~------~----~------~--~---

Re: GPU-based Compute Cloud

Don't underestimate the power of specialization in driving compute
clouds for specific compute tasks.

It is not just GPUs, Kickfire (www.kickfire.com) have released a
so-called SQL chip (DPU or SQLPU)?

I see some reasons for software vendors looking at hardware, one is, if
your software is becoming a commodity you need to move up or down the
stack. So why not use FPGAs from Altera or Xilinx to test run burning
your software into hardware.

The other reason is the recent super computing record using the IBM cell
chip which is used in current game consoles. Essentially people are
rediscovering the power of parallelization and also specialization that
game console makers have known for ages. The current PC and Server CPUs
are made to be good at many tasks, the future may involve increased
specialization of hardware for specific tasks.

If you were are Database as a service provider and had a stack of
kickfire applicances, suddenly given the performance gains of running
part of the database in hardware means you can host more databases per
machine per cubic feet (or cubic metre) having a real impact your bottom
line.

Have Fun

Paul

http://blog.dbadojo.com
http://blog.vmdatamine.com

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "Cloud Computing" group.
To post to this group, send email to cloud-computing@googlegroups.com
To unsubscribe from this group, send email to cloud-computing-unsubscribe@googlegroups.com
For more options, visit this group at http://groups.google.ca/group/cloud-computing?hl=en
-~----------~----~----~----~------~----~------~--~---

Re: GPU-based Compute Cloud

Good article in HPCWire this week on when and when not to use
accelerators:
http://www.hpcwire.com/features/In-Socket_Accelerators__When_to_Use_Them.html

One illustrative quote:
**********************************
For 2008, the industry buzzwords are "hardware acceleration." CPU
vendors are integrating custom integrated IP into their chips. AMD and
Intel are creating an ecosystem for third party accelerators, named
Torenzza and QuickAssist. GPU vendors are setting their sights on
general purpose functionality. Meanwhile, a host of other chip
companies, too many to mention, are developing new products that
target this high performance computing (HPC) market.

A little known fact in the HPC community is that the embedded
computing market has always solved their problems with a combination
of CPUs and accelerators. Due to different space, weight, power, and
environmental requirements for high performance embedded computing
(HPEC), a large portion of those accelerators are implemented with
field programmable gate arrays (FPGAs) from companies like Altera,
Xilinx, and others.

As these two markets collide,....
***********************************

--
Greg Pfister

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "Cloud Computing" group.
To post to this group, send email to cloud-computing@googlegroups.com
To unsubscribe from this group, send email to cloud-computing-unsubscribe@googlegroups.com
For more options, visit this group at http://groups.google.ca/group/cloud-computing?hl=en
-~----------~----~----~----~------~----~------~--~---

Re: GPU-based Compute Cloud

On Thu, Jun 12, 2008 at 5:33 PM, Timur Yusupov <y...@bwl.uni-kiel.de> wrote:

Hi All,

I was following your discussion on GPU and PS3 based computing.
The key issue is easy transferability of a code. Do you have in mind SPU and
GPU implementation of BLAS routines? How far are we in implementing BLAS on
SPU and GPU (double precision)?

Thank you,
Timur

Timur,

We have local company called RapidMind, who was able to run BLAS routine SGEMM on GPU

quote:

While using the RapidMind Multi-core Development Platform, we can compare the performance of BLAS dense linear algebra operations, the FFT, and European option pricing on the GPU against highly tuned CPU implementations on the fastest available CPUs.

source:

http://www.rapidmind.net/case-gpu.php

You can download RapidMind Developer Edition for free at their site.

As for transferability, I guess, initially it (GPGPU Cloud) should serve only specific class of applications, suitable/tailored to underlying HW/SW.

cheers,

Khaz Sapenov

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "Cloud Computing" group.
To post to this group, send email to cloud-computing@googlegroups.com
To unsubscribe from this group, send email to cloud-computing-unsubscribe@googlegroups.com
For more options, visit this group at http://groups.google.ca/group/cloud-computing?hl=en
-~----------~----~----~----~------~----~------~--~---

Re: GPU-based Compute Cloud

Hi All,

I was following your discussion on GPU and PS3 based computing.
The key issue is easy transferability of a code. Do you have in mind SPU and
GPU implementation of BLAS routines? How far are we in implementing BLAS on
SPU and GPU (double precision)?

Thank you,
Timur

----- Original Message -----
From: "randall" <r....@qrimp.com>
To: "Cloud Computing" <cloud-computing@googlegroups.com>
Sent: Thursday, June 12, 2008 5:05 PM
Subject: Re: GPU-based Compute Cloud

>
> I just read this story about Roadrunner. It's a super computer, but
> why not the same model for a cloud? It's build on almost 300 Cell
> processors and has the highest computes per watt of any super computer
> and broke the petaflop record.
>
> http://www.beyond3d.com/content/news/651
>
> What is the difference between a cloud and a super computer like this?
> Could something like this be used as a cloud?
>
> - randall
>
>
>
> On Jun 11, 11:21 pm, Alan Ho <karlu...@yahoo.ca> wrote:
>> Disclaimer : rumours below
>>
>> I've heard some rumour of a startup company in stealth mode in boulder
>> doing exactly that. I'd expect that IBM may combine its cell processor
>> and upcoming Blue Cloud initiaves at some point too.
>>
>> Why stop at graphics processors ? Another rumour is that a quantum
>> computing company may offer its services through a cloud initiative too -
>> makes sense since it takes near kelvin temperatures to operate quantum
>> computer :)
>>
>> -----Original Message-----
>> From: Khazret Sapenov <sape...@gmail.com>
>> Sent: June 11, 2008 7:46 PM
>> To: cloud-computing@googlegroups.com
>> Subject: GPU-based Compute Cloud
>>
>> What do you think about compute cloud, powered by general purpose GPUs
>> (graphics processing units), dedicated entirely to stream processing?
>> It'll provide good(cheap?) facilities for evolutionary computation or
>> media processing (e.g. transcoding video files).
>> I've read that SUN is working in this direction (apparently due to
>> inoptimal price/performance ratio), and Apple just makes alternative to
>> CUDA, which eventually is one step further to commodity GPGPUs, suitable
>> for cloud.
>>
>> cheers,
>> Khaz Sapenov
>
> >
>

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "Cloud Computing" group.
To post to this group, send email to cloud-computing@googlegroups.com
To unsubscribe from this group, send email to cloud-computing-unsubscribe@googlegroups.com
For more options, visit this group at http://groups.google.ca/group/cloud-computing?hl=en
-~----------~----~----~----~------~----~------~--~---

Re: GPU-based Compute Cloud

I just read this story about Roadrunner. It's a super computer, but
why not the same model for a cloud? It's build on almost 300 Cell
processors and has the highest computes per watt of any super computer
and broke the petaflop record.

http://www.beyond3d.com/content/news/651

What is the difference between a cloud and a super computer like this?
Could something like this be used as a cloud?

- randall

On Jun 11, 11:21 pm, Alan Ho <karlu...@yahoo.ca> wrote:
> Disclaimer : rumours below
>
> I've heard some rumour of a startup company in stealth mode in boulder doing exactly that. I'd expect that IBM may combine its cell processor and upcoming Blue Cloud initiaves at some point too.
>
> Why stop at graphics processors ? Another rumour is that a quantum computing company may offer its services through a cloud initiative too - makes sense since it takes near kelvin temperatures to operate quantum computer :)
>
> -----Original Message-----
> From: Khazret Sapenov <sape...@gmail.com>
> Sent: June 11, 2008 7:46 PM
> To: cloud-computing@googlegroups.com
> Subject: GPU-based Compute Cloud
>
> What do you think about compute cloud, powered by general purpose GPUs (graphics processing units), dedicated entirely to stream processing?
> It'll provide good(cheap?) facilities for evolutionary computation or media processing (e.g. transcoding video files).
> I've read that SUN is working in this direction (apparently due to inoptimal price/performance ratio), and Apple just makes alternative to CUDA, which eventually is one step further to commodity GPGPUs, suitable for cloud.
>
> cheers,
> Khaz Sapenov

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "Cloud Computing" group.
To post to this group, send email to cloud-computing@googlegroups.com
To unsubscribe from this group, send email to cloud-computing-unsubscribe@googlegroups.com
For more options, visit this group at http://groups.google.ca/group/cloud-computing?hl=en
-~----------~----~----~----~------~----~------~--~---