Radiance on Raspberry Pi

For those interested, I built Radiance on a Raspberry Pi yesterday
(700 MHz single-core ARMv6, 3.5W, 512 MB RAM) and ran the speed test
from http://markjstock.org/pages/rad_bench.html.

The results are not encouraging, but you don't buy a Raspberry Pi for
speed. It took 17 hours to finish the rpict portion of the test. The
best serial time ever recorded was just under 18 minutes on an Intel
i7-2600K from 1.5 years ago (57x faster). The peak SMP time is 3.5
minutes, recorded earlier this year (296x faster). The Raspberry Pi's
time was similar to what we would expect from computers from 2001-2: a
Pentium II at about 250MHz, or a 400HMz Sun UltraSparc, both which
regularly sported about 512MB of RAM, but obviously cost about 100x
more.

100 Raspberry Pi machines, networked and working in parallel, would
consume 350W and cost $3500 (not counting networking hardware), both
numbers in the ballpark of the best current CPU-intensive
single-workstation systems. Assuming perfect parallelization, it could
finish in a little over 10 minutes, but a more realistic guess would
be 15-20 minutes. I'm glad NREL decided to go with the Xeon Phi
instead of a cluster of RPis.

Maybe I should start a Radiance "speed" test that measures the results
in Joules instead of seconds.

Mark

1 Like

Thanks for sharing, this is perhaps the geekiest thing I've ever read. Is this what you do when your city is holed up inside?

- Rob

···

On Apr 20, 2013, at 1:03 PM, Mark Stock <[email protected]> wrote:

For those interested, I built Radiance on a Raspberry Pi yesterday
(700 MHz single-core ARMv6, 3.5W, 512 MB RAM) and ran the speed test
from http://markjstock.org/pages/rad_bench.html.

The results are not encouraging, but you don't buy a Raspberry Pi for
speed. It took 17 hours to finish the rpict portion of the test. The
best serial time ever recorded was just under 18 minutes on an Intel
i7-2600K from 1.5 years ago (57x faster). The peak SMP time is 3.5
minutes, recorded earlier this year (296x faster). The Raspberry Pi's
time was similar to what we would expect from computers from 2001-2: a
Pentium II at about 250MHz, or a 400HMz Sun UltraSparc, both which
regularly sported about 512MB of RAM, but obviously cost about 100x
more.

100 Raspberry Pi machines, networked and working in parallel, would
consume 350W and cost $3500 (not counting networking hardware), both
numbers in the ballpark of the best current CPU-intensive
single-workstation systems. Assuming perfect parallelization, it could
finish in a little over 10 minutes, but a more realistic guess would
be 15-20 minutes. I'm glad NREL decided to go with the Xeon Phi
instead of a cluster of RPis.

Maybe I should start a Radiance "speed" test that measures the results
in Joules instead of seconds.

Mark

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

Hey Mark,

Awesome!

-Jack

···

On 4/20/2013 3:03 PM, Mark Stock wrote:

For those interested, I built Radiance on a Raspberry Pi yesterday
(700 MHz single-core ARMv6, 3.5W, 512 MB RAM) and ran the speed test
from http://markjstock.org/pages/rad_bench.html.

The results are not encouraging, but you don't buy a Raspberry Pi for
speed. It took 17 hours to finish the rpict portion of the test. The
best serial time ever recorded was just under 18 minutes on an Intel
i7-2600K from 1.5 years ago (57x faster). The peak SMP time is 3.5
minutes, recorded earlier this year (296x faster). The Raspberry Pi's
time was similar to what we would expect from computers from 2001-2: a
Pentium II at about 250MHz, or a 400HMz Sun UltraSparc, both which
regularly sported about 512MB of RAM, but obviously cost about 100x
more.

100 Raspberry Pi machines, networked and working in parallel, would
consume 350W and cost $3500 (not counting networking hardware), both
numbers in the ballpark of the best current CPU-intensive
single-workstation systems. Assuming perfect parallelization, it could
finish in a little over 10 minutes, but a more realistic guess would
be 15-20 minutes. I'm glad NREL decided to go with the Xeon Phi
instead of a cluster of RPis.

Maybe I should start a Radiance "speed" test that measures the results
in Joules instead of seconds.

Mark

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

Hi Mark,

This is indeed very interesting. (And very cool, in my opinion!)

I have been looking for a low-cost platform to do HDR image generation (in
the field) with the program HDRGEN, the command-line HDR image builder.

http://www.anyhere.com/

Any idea how long it would take the Pi to composite 7 .JPG images (say
about 1000px by 1000px in dimension) into an HDR image?.

(Then i would do a bit of analysis with Radiance and send some commands to
control physical devices)

Best,

-Kyle

···

-----------------------------------------------
Kyle Konis, AIA, Ph.D
Assistant Professor
School of Architecture, WAH 204
University of Southern California
Los Angeles, CA 90089-0291
http://arch.usc.edu/faculty/kkonis
-----------------------------------------------

On Sat, Apr 20, 2013 at 12:03 PM, Mark Stock <[email protected]> wrote:

For those interested, I built Radiance on a Raspberry Pi yesterday
(700 MHz single-core ARMv6, 3.5W, 512 MB RAM) and ran the speed test
from http://markjstock.org/pages/rad_bench.html.

The results are not encouraging, but you don't buy a Raspberry Pi for
speed. It took 17 hours to finish the rpict portion of the test. The
best serial time ever recorded was just under 18 minutes on an Intel
i7-2600K from 1.5 years ago (57x faster). The peak SMP time is 3.5
minutes, recorded earlier this year (296x faster). The Raspberry Pi's
time was similar to what we would expect from computers from 2001-2: a
Pentium II at about 250MHz, or a 400HMz Sun UltraSparc, both which
regularly sported about 512MB of RAM, but obviously cost about 100x
more.

100 Raspberry Pi machines, networked and working in parallel, would
consume 350W and cost $3500 (not counting networking hardware), both
numbers in the ballpark of the best current CPU-intensive
single-workstation systems. Assuming perfect parallelization, it could
finish in a little over 10 minutes, but a more realistic guess would
be 15-20 minutes. I'm glad NREL decided to go with the Xeon Phi
instead of a cluster of RPis.

Maybe I should start a Radiance "speed" test that measures the results
in Joules instead of seconds.

Mark

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

Oops, i've done it again,

(...sent a personalized message to the whole list by a hasty use of the
"reply")

Apologies!

-Kyle

···

On Mon, Apr 22, 2013 at 2:49 PM, Kyle Konis <[email protected]> wrote:

Hi Mark,

This is indeed very interesting. (And very cool, in my opinion!)

I have been looking for a low-cost platform to do HDR image generation (in
the field) with the program HDRGEN, the command-line HDR image builder.

http://www.anyhere.com/

Any idea how long it would take the Pi to composite 7 .JPG images (say
about 1000px by 1000px in dimension) into an HDR image?.

(Then i would do a bit of analysis with Radiance and send some commands to
control physical devices)

Best,

-Kyle

-----------------------------------------------
Kyle Konis, AIA, Ph.D
Assistant Professor
School of Architecture, WAH 204
University of Southern California
Los Angeles, CA 90089-0291
http://arch.usc.edu/faculty/kkonis
-----------------------------------------------

On Sat, Apr 20, 2013 at 12:03 PM, Mark Stock <[email protected]> wrote:

For those interested, I built Radiance on a Raspberry Pi yesterday
(700 MHz single-core ARMv6, 3.5W, 512 MB RAM) and ran the speed test
from http://markjstock.org/pages/rad_bench.html.

The results are not encouraging, but you don't buy a Raspberry Pi for
speed. It took 17 hours to finish the rpict portion of the test. The
best serial time ever recorded was just under 18 minutes on an Intel
i7-2600K from 1.5 years ago (57x faster). The peak SMP time is 3.5
minutes, recorded earlier this year (296x faster). The Raspberry Pi's
time was similar to what we would expect from computers from 2001-2: a
Pentium II at about 250MHz, or a 400HMz Sun UltraSparc, both which
regularly sported about 512MB of RAM, but obviously cost about 100x
more.

100 Raspberry Pi machines, networked and working in parallel, would
consume 350W and cost $3500 (not counting networking hardware), both
numbers in the ballpark of the best current CPU-intensive
single-workstation systems. Assuming perfect parallelization, it could
finish in a little over 10 minutes, but a more realistic guess would
be 15-20 minutes. I'm glad NREL decided to go with the Xeon Phi
instead of a cluster of RPis.

Maybe I should start a Radiance "speed" test that measures the results
in Joules instead of seconds.

Mark

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

That's awesome. Now if only someone would do something like a Raspberry Pi with a multicore ARM.

Ummmmm. You know, Archos 101 G9s are multicore ARM tablets. They run Android, which runs on Linux. It *could* be done…

Randolph

It's already happening. Tilera has a 72-core chip, and Parallella is
hoping to sell a 16-core chip for $100.

How well these chips will perform Radiance simulations is still
unknown, though some knowledge will definitely be gained when NREL
begins exercising their new Xeon Phi-powered supercomputer.

···

On 4/22/13, Randolph M. Fritz (i) <[email protected]> wrote:

That's awesome. Now if only someone would do something like a
Raspberry Pi with a multicore ARM.

Ummmmm. You know, Archos 101 G9s are multicore ARM tablets. They run
Android, which runs on Linux. It *could* be done…

Randolph

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

Greg,

Inspired by Mark Stock's semi-recent post, i am curious if HDRgen could be
compiled to work on the Beagle Bone?

···

#####
http://beagleboard.org/Products/BeagleBone
BeagleBone is an $89 MSRP, credit-card-sized Linux computer that connects
to the Internet and runs software such as Android 4.0 and Ubuntu. With
plenty of I/O and processing power for real-time analysis provided by an
AM335x 720MHz ARM® processor, BeagleBone can be complemented with cape
plug-in boards to augment functionality.
#####

From a search on the HDRI list archive, i found that:

##################

I managed to compile hdrgen under Lubuntu using VMWare, so if anyone else
wants a copy, let me know. I'll send one to Axel since it's a little more
recent version of the code (I think).

-Greg

##################

In summary, i plan to procure a Beagle Bone, install Ubuntu and Radiance,
and include hdrgen to complete the setup if you can supply me with a copy.

My interest here is in using a CCD camera to capture bracketed sets of
images, composite and analyze them on the Beagle Bone (via Radiance and
HDRgen), then use the information to thoughtfully control lighting and
shading systems in a physical space.

This plea may have more appropriately been made on the HDRI list, but i
wanted to add it to the current topic of Radiance paired with tiny low-cost
computers.

If others have comments or suggestions on the viability of this effort i
would be grateful to hear them!

-Kyle

-----------------------------------------------
Kyle Konis, AIA, Ph.D
Assistant Professor
School of Architecture, WAH 204
University of Southern California
Los Angeles, CA 90089-0291
http://arch.usc.edu/faculty/kkonis
-----------------------------------------------

There's the hitch that I don't distribute the code for hdrgen....

-G

···

From: Kyle Konis <[email protected]>
Date: June 19, 2013 5:23:18 PM PDT

Greg,

Inspired by Mark Stock's semi-recent post, i am curious if HDRgen could be compiled to work on the Beagle Bone?

#####
http://beagleboard.org/Products/BeagleBone
BeagleBone is an $89 MSRP, credit-card-sized Linux computer that connects to the Internet and runs software such as Android 4.0 and Ubuntu. With plenty of I/O and processing power for real-time analysis provided by an AM335x 720MHz ARM® processor, BeagleBone can be complemented with cape plug-in boards to augment functionality.
#####

From a search on the HDRI list archive, i found that:

##################

I managed to compile hdrgen under Lubuntu using VMWare, so if anyone else wants a copy, let me know. I'll send one to Axel since it's a little more recent version of the code (I think).

-Greg

##################

In summary, i plan to procure a Beagle Bone, install Ubuntu and Radiance, and include hdrgen to complete the setup if you can supply me with a copy.

My interest here is in using a CCD camera to capture bracketed sets of images, composite and analyze them on the Beagle Bone (via Radiance and HDRgen), then use the information to thoughtfully control lighting and shading systems in a physical space.

This plea may have more appropriately been made on the HDRI list, but i wanted to add it to the current topic of Radiance paired with tiny low-cost computers.

If others have comments or suggestions on the viability of this effort i would be grateful to hear them!

-Kyle

-----------------------------------------------
Kyle Konis, AIA, Ph.D
Assistant Professor
School of Architecture, WAH 204
University of Southern California
Los Angeles, CA 90089-0291
http://arch.usc.edu/faculty/kkonis
-----------------------------------------------

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

Would it make any difference if I loaned you a Raspberry Pi kit to build on?

Mark

···

On 6/19/13, Greg Ward <[email protected]> wrote:

There's the hitch that I don't distribute the code for hdrgen....

-G

From: Kyle Konis <[email protected]>
Date: June 19, 2013 5:23:18 PM PDT

Greg,

Inspired by Mark Stock's semi-recent post, i am curious if HDRgen could be
compiled to work on the Beagle Bone?

#####
http://beagleboard.org/Products/BeagleBone
BeagleBone is an $89 MSRP, credit-card-sized Linux computer that connects
to the Internet and runs software such as Android 4.0 and Ubuntu. With
plenty of I/O and processing power for real-time analysis provided by an
AM335x 720MHz ARM® processor, BeagleBone can be complemented with cape
plug-in boards to augment functionality.
#####

From a search on the HDRI list archive, i found that:

##################

I managed to compile hdrgen under Lubuntu using VMWare, so if anyone else
wants a copy, let me know. I'll send one to Axel since it's a little more
recent version of the code (I think).

-Greg

##################

In summary, i plan to procure a Beagle Bone, install Ubuntu and Radiance,
and include hdrgen to complete the setup if you can supply me with a
copy.

My interest here is in using a CCD camera to capture bracketed sets of
images, composite and analyze them on the Beagle Bone (via Radiance and
HDRgen), then use the information to thoughtfully control lighting and
shading systems in a physical space.

This plea may have more appropriately been made on the HDRI list, but i
wanted to add it to the current topic of Radiance paired with tiny
low-cost computers.

If others have comments or suggestions on the viability of this effort i
would be grateful to hear them!

-Kyle

-----------------------------------------------
Kyle Konis, AIA, Ph.D
Assistant Professor
School of Architecture, WAH 204
University of Southern California
Los Angeles, CA 90089-0291
http://arch.usc.edu/faculty/kkonis
-----------------------------------------------

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

Well, if it's really important to you guys, you or Kyle could "borrow" my code to compile on your little credit-card computers. I don't have much time right now to mess around with it.

-Greg

···

From: Mark Stock <[email protected]>
Date: June 19, 2013 8:45:41 PM PDT

Would it make any difference if I loaned you a Raspberry Pi kit to build on?

Mark

On 6/19/13, Greg Ward <[email protected]> wrote:

There's the hitch that I don't distribute the code for hdrgen....

-G

From: Kyle Konis <[email protected]>
Date: June 19, 2013 5:23:18 PM PDT

Greg,

Inspired by Mark Stock's semi-recent post, i am curious if HDRgen could be
compiled to work on the Beagle Bone?

#####
http://beagleboard.org/Products/BeagleBone
BeagleBone is an $89 MSRP, credit-card-sized Linux computer that connects
to the Internet and runs software such as Android 4.0 and Ubuntu. With
plenty of I/O and processing power for real-time analysis provided by an
AM335x 720MHz ARM® processor, BeagleBone can be complemented with cape
plug-in boards to augment functionality.
#####

From a search on the HDRI list archive, i found that:

##################

I managed to compile hdrgen under Lubuntu using VMWare, so if anyone else
wants a copy, let me know. I'll send one to Axel since it's a little more
recent version of the code (I think).

-Greg

##################

In summary, i plan to procure a Beagle Bone, install Ubuntu and Radiance,
and include hdrgen to complete the setup if you can supply me with a
copy.

My interest here is in using a CCD camera to capture bracketed sets of
images, composite and analyze them on the Beagle Bone (via Radiance and
HDRgen), then use the information to thoughtfully control lighting and
shading systems in a physical space.

This plea may have more appropriately been made on the HDRI list, but i
wanted to add it to the current topic of Radiance paired with tiny
low-cost computers.

If others have comments or suggestions on the viability of this effort i
would be grateful to hear them!

-Kyle

That would be excellent. I will discuss with Mark offline and get back to
you.

-Kyle

···

On Wed, Jun 19, 2013 at 9:26 PM, Greg Ward <[email protected]> wrote:

Well, if it's really important to you guys, you or Kyle could "borrow" my
code to compile on your little credit-card computers. I don't have much
time right now to mess around with it.

-Greg

> From: Mark Stock <[email protected]>
> Date: June 19, 2013 8:45:41 PM PDT
>
> Would it make any difference if I loaned you a Raspberry Pi kit to build
on?
>
> Mark
>
> On 6/19/13, Greg Ward <[email protected]> wrote:
>> There's the hitch that I don't distribute the code for hdrgen....
>>
>> -G
>>
>>> From: Kyle Konis <[email protected]>
>>> Date: June 19, 2013 5:23:18 PM PDT
>>>
>>> Greg,
>>>
>>> Inspired by Mark Stock's semi-recent post, i am curious if HDRgen
could be
>>> compiled to work on the Beagle Bone?
>>>
>>> #####
>>> http://beagleboard.org/Products/BeagleBone
>>> BeagleBone is an $89 MSRP, credit-card-sized Linux computer that
connects
>>> to the Internet and runs software such as Android 4.0 and Ubuntu. With
>>> plenty of I/O and processing power for real-time analysis provided by
an
>>> AM335x 720MHz ARM® processor, BeagleBone can be complemented with cape
>>> plug-in boards to augment functionality.
>>> #####
>>>
>>>
>>> From a search on the HDRI list archive, i found that:
>>>
>>> ##################
>>>
>>> I managed to compile hdrgen under Lubuntu using VMWare, so if anyone
else
>>> wants a copy, let me know. I'll send one to Axel since it's a little
more
>>> recent version of the code (I think).
>>>
>>> -Greg
>>>
>>> ##################
>>>
>>> In summary, i plan to procure a Beagle Bone, install Ubuntu and
Radiance,
>>> and include hdrgen to complete the setup if you can supply me with a
>>> copy.
>>>
>>>
>>> My interest here is in using a CCD camera to capture bracketed sets of
>>> images, composite and analyze them on the Beagle Bone (via Radiance and
>>> HDRgen), then use the information to thoughtfully control lighting and
>>> shading systems in a physical space.
>>>
>>> This plea may have more appropriately been made on the HDRI list, but i
>>> wanted to add it to the current topic of Radiance paired with tiny
>>> low-cost computers.
>>>
>>> If others have comments or suggestions on the viability of this effort
i
>>> would be grateful to hear them!
>>>
>>> -Kyle

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

Dear Greg

I hope you don't mind me contacting you - i'm a lecturer and researcher at De Montfort University, working on an exciting HDR project with Birgit Painter. Birgit has told me a lot about the work that she did with Denis Fan and John Mardaljevic.

The current project involves using a Raspberry Pi to capture bracketed shots and I would love to be able to code Radiance onto it. I saw a post on radiance-online (regarding Beagle Bone), and therefore wondered if a compatible version of the code was now available for a Raspberry Pi please? I have trawled the internet and not as yet found a suitable version.

I look forward to hearing from you.

Kind regards

Marie

···

____________________________________________

Dr. Marie Bassford MPhys PhD MInstP
BSc/MPhys Physics Course Leader
Senior Lecturer, Faculty of Technology
Member of the Centre for Computational Intelligence
STEM Ambassador

De Montfort University
T: +44 (0) 116 2577055
E: [email protected]

Hello Marie,

You actually wrote to the Radiance mailing list, not me directly. It's just as well, though, since I'm not aware if anyone has ported Radiance to the Raspberry Pi. If anyone on the mailing list has worked on this or has some helpful information, I hope they will respond.

There were some folks working on a RP port of hdrgen, which I think they got working. Perhaps this would be useful to your project as well?

Best,
-Greg

···

From: Marie Bassford <[email protected]>
Date: January 28, 2014 1:38:08 AM PST

Dear Greg

I hope you don't mind me contacting you - i'm a lecturer and researcher at De Montfort University, working on an exciting HDR project with Birgit Painter. Birgit has told me a lot about the work that she did with Denis Fan and John Mardaljevic.

The current project involves using a Raspberry Pi to capture bracketed shots and I would love to be able to code Radiance onto it. I saw a post on radiance-online (regarding Beagle Bone), and therefore wondered if a compatible version of the code was now available for a Raspberry Pi please? I have trawled the internet and not as yet found a suitable version.

I look forward to hearing from you.

Kind regards

Marie
____________________________________________

Dr. Marie Bassford MPhys PhD MInstP

Dear Marie,

the raspberry pi is just a Linux system with limited resources. If you have a Debian system installed,

sudo apt-get install radiance

should be all you need to install Radiance, given a working Internet connection or installation media.

There used to be a define to reduce memory requirements by using single precision floats when compiling Radiance, but probably no one used that for decades.

All in all, using the image processing tools may make sense, but I would not spend my time trying to get complete renderings from such hardware :slight_smile:

Cheers, Lars.

The Raspberry Pi build of hdrgen worked. And Radiance on the Pi works
as well. Note that rpict on the Pi runs about as fast as you'd expect
from a $30 computer---it finishes my benchmark test in 1021 minutes,
while a 4-core i7 can do it in parallel in 4 minutes.

Mark

···

On 1/28/14, Greg Ward <[email protected]> wrote:

Hello Marie,

You actually wrote to the Radiance mailing list, not me directly. It's just
as well, though, since I'm not aware if anyone has ported Radiance to the
Raspberry Pi. If anyone on the mailing list has worked on this or has some
helpful information, I hope they will respond.

There were some folks working on a RP port of hdrgen, which I think they got
working. Perhaps this would be useful to your project as well?

Best,
-Greg

From: Marie Bassford <[email protected]>
Date: January 28, 2014 1:38:08 AM PST

Dear Greg

I hope you don't mind me contacting you - i'm a lecturer and researcher at
De Montfort University, working on an exciting HDR project with Birgit
Painter. Birgit has told me a lot about the work that she did with Denis
Fan and John Mardaljevic.

The current project involves using a Raspberry Pi to capture bracketed
shots and I would love to be able to code Radiance onto it. I saw a post
on radiance-online (regarding Beagle Bone), and therefore wondered if a
compatible version of the code was now available for a Raspberry Pi
please? I have trawled the internet and not as yet found a suitable
version.

I look forward to hearing from you.

Kind regards

Marie
____________________________________________

Dr. Marie Bassford MPhys PhD MInstP