helpGCC Compile Farm - Support: sr #2689, Google Summer of Code

Show feedback again

sr #2689: Google Summer of Code

Submitted by:  Laurent GUERBY <guerby>
Submitted on:  Tue Jan 11 20:44:25 2011  
Category: NonePriority: 5 - Normal
Severity: 3 - NormalStatus: Done
Privacy: PublicAssigned to: Laurent GUERBY <guerby>
Open/Closed: ClosedOperating System: None

Add a New Comment (Rich MarkupRich Markup):

You are not logged in

Please log in, so followups can be emailed to you.


(Jump to the original submission Jump to the original submission)

Thu Oct 8 07:59:19 2015, comment #22:


nvidia GPU coming soon :)

Laurent GUERBY <guerby>
Project AdministratorIn charge of this item.
Thu Mar 31 12:31:38 2011, comment #21:

I'd like to use it to just learn OpenCL in general. ~nightstrike

Wed Mar 30 22:00:47 2011, comment #20:

I'd like an Nvidia GPU to create a regression testing and benchmarking system for KVM PCI passthrough. Right now there isn't a public tool for the community to determine whether choosing a different combination of Nvidia and KVM drivers will have a negative impact or not.

I already have a machine to host the GPU with and the management infrastructure to slave it to a regression testing server. It would make my life a little easier to have a networkable PDU though.

A newer Dell Poweredge or Supermicro IPMI card with a WPCM450 management chip on it would also be nice to use for creating a free release of the Linux distribution that runs on the management chip by default.

I'd also like to shore up the coreboot support for running the native Nvidia BIOS.

Jason Spence <jrspence>
Tue Mar 29 18:59:40 2011, comment #19:


I would be interested in getting access to GPUs of any kind to work on OpenCL bindings for Parrot Virtual Machine and Perl 6.

Jonathan "Duke" Leto

Duke Leto <dukeleto>
Tue Mar 29 13:48:59 2011, comment #18:


I'd be interested in getting access to NVIDIA GPUs (if possible, Fermi with strong double-precision float performance) for two related projects :
- JavaCL (OpenCL bindings for Java :
- ScalaCL (Run Scala on your graphic card :

The goals are to ensure JavaCL behaves well on high-end cards and to benchmark / fine-tune JavaCL and ScalaCL high-level utilities.

Thanks for your work !

Olivier Chafik <ochafik>
Tue Mar 29 11:48:01 2011, comment #17:

Thanks to Dug Song we've been talking to Nvidia to be part of their partnership program:

It should be okay to get a few GPU cards for the farm as long
as we can identify any projects that people have expressed interest in doing on GPU hardware.

Please fill in.

Laurent GUERBY <guerby>
Project AdministratorIn charge of this item.
Thu Feb 3 17:50:19 2011, comment #16:

Oh, thanks. Speaking of the hardware, I cannot suggest anything but a dual/quad core PowerMac G5. Of course, if someone could donate some IBM POWER server then it would be perfect. :)


Thu Feb 3 04:00:57 2011, comment #15:

mik: I cleaned up some stuff on gcc40 which has now 15 GB free.

Laurent GUERBY <guerby>
Project AdministratorIn charge of this item.
Wed Feb 2 17:23:10 2011, comment #14:

What about a ppc64? From my observations gcc38 crashes too easily with no good reason, gcc40 is quite loaded (even / is full).


Wed Feb 2 15:43:06 2011, comment #13:

I vote for both GPU platforms as well as a Windows platform for working on the Mingw-w64 toolchain.

nightstrike <nightstrike>
Wed Jan 26 15:28:03 2011, comment #12:

I will try to schedule a return of gcc30, stay tuned.

Laurent GUERBY <guerby>
Project AdministratorIn charge of this item.
Wed Jan 19 17:18:24 2011, comment #11:

It is possible to return Alpha (gcc30) machine back to on-line?

I make patches for hookineze varios target macros in target hooks in the GCC and want to test them on the Alpha target also.


Anatoly Sokolov <aesok>
Sat Jan 15 22:37:57 2011, comment #10:

PandaBoard definitely has no really fast interface to attach the storage, but using NFS as it works for gcc3x will kill the overall performance. I did
dd if=/dev/zero of=$HOME/zero bs=1M count=100
on gcc33 and got 890K/s (LA was equal to 0 before the test). Please, take that into account.

Thu Jan 13 19:47:17 2011, comment #9:

A GPGPU-enabled architecture with as free as possible drivers and toolchain supporting OpenCL would be great. I don't know the release date of AMD Fusion and Intel SandyBridge chips meeting those requirements, but I think such next-gen architecture would be nice to experiment with.

Nicolas BENOIT <nbenoit>
Thu Jan 13 12:16:59 2011, comment #8:

Regarding ARM the compile farm already received a Pandaboard donated by Texas Instrument (gcc44 on gcc-testresults@) but it hasn't been made public yet because under the first setup I installed the board freezes from time to time, Canonical and Texas Instrument will be looking into it:

For loongson 3A I contacted the people working on the GNU toolchain (one even has an account on the farm :) and they told me they'll ping us as soon as the hardware is more available.

Alex, thanks for sharing your insight on GPU and volunteering, we'll have to decide what to do when more people have contributed.

Laurent GUERBY <guerby>
Project AdministratorIn charge of this item.
Thu Jan 13 10:00:07 2011, comment #7:

I agree with Alex, Loongson 3 looks promising. Regarding the ARM hardware, there is the PandaBoard: dual-core Cortex-A9 1 GHz / 1 GB DDR2, that is awesome IMO.


Wed Jan 12 22:46:58 2011, comment #6:

(My comment was #1 below, regarding GPUs:) I'll be glad to help regarding further researching the options, and also installing drivers and toolchain in case you eventually decide for spending some of the money this way. As mentioned in my previous comment, I'm convinced NVIDIA hardware is all that would be needed: Intel GPUs are really not yet for any kind of GPGPU work, and in my experience AMD drivers and toolchain implementation is plain crappy (and CUDA is supported only by NVIDIA anyway). So I guess some kind of dual-card, compute-only setup with some mid-range NVIDIA offer could do the job fine. The cons of the whole thing are: am not sure how many developers would be actually interested in the whole thing (my own work in past couple years was in GPGPU, but unfortunately none of this in open-source world), and this type of hardware becomes obsolete very fast; also, drivers and everything else is proprietary in this game, so again am not sure is it appropriate to spend the money this way. Also, the biggest issue regarding the installation is actually providing for multiple people to use resources simultaneously - was using GPU on-demand clusters, like offer from Penguin Computing - but to be honest I have no clue how are they providing for sharing GPU resources...

As an alternative proposal: in the light of recent events (I mean on all of these recent CES 2011 announcements), maybe investing in some ARM hardware would make sense. Also, I'm big fan of Lemote developments (I have to admit that out of all machines on compile farm I'm using gcc51 most) so maybe saving some money for some point in the future when some kind of Loongson 3 machine becomes available would make sense too.


Wed Jan 12 15:48:22 2011, comment #5:

A system with some X86 class but non-Intel/AMD CPU would be good.
For example VIA C3.


Wed Jan 12 11:43:11 2011, comment #4:

#3 => We should have an x86_64-netbsd system online next week (without using the google money :).

Laurent GUERBY <guerby>
Project AdministratorIn charge of this item.
Wed Jan 12 11:37:58 2011, comment #3:

GCC currently doesn't bootstrap on NetBSD, so a NetBSD machine, even a low-power VM, would be helpful

Tue Jan 11 21:38:02 2011, comment #2:

Do you know of someone who knows well GPU and would be willing to help setup GPUs toolchains and recommand specific hardware?

We could get NVIDA, AMD and Intel GPU in the spirit of hardware coverage of the compile farm but we really need knowledgeable volunteers here.

PS: if not logged in please do not forget to add your name somewhere :)

Laurent GUERBY <guerby>
Project AdministratorIn charge of this item.
Tue Jan 11 21:33:41 2011, comment #1:

As this is larger chunk of money available: Maybe one idea could be adding some kind of NVIDIA GPU to one of existing Linux x86 machines? Don't know would it be that much developers interested but if so, then putting in some Fermi generation card could make it possible to test both CUDA and OpenCL codes, and this GPGPU stuff is becoming more and more important...

Tue Jan 11 20:44:25 2011, original submission:

Open to suggestions :).

Laurent GUERBY <guerby>
Project AdministratorIn charge of this item.


(Note: upload size limit is set to 1024 kB, after insertion of the required escape characters.)

Attach File(s):

No files currently attached


Depends on the following items: None found

Items that depend on this one: None found


Carbon-Copy List
  • -unavailable- added by jrspence (Posted a comment)
  • -unavailable- added by dukeleto (Posted a comment)
  • -unavailable- added by ochafik (Posted a comment)
  • -unavailable- added by nightstrike (Posted a comment)
  • -unavailable- added by aesok (Posted a comment)
  • -unavailable- added by nbenoit (Posted a comment)
  • -unavailable- added by guerby (Submitted the item)

    Do you think this task is very important?
    If so, you can click here to add your encouragement to it.
    This task has 0 encouragements so far.

    Only logged-in users can vote.


    Error: not logged in



    Follow 2 latest changes.

    Date Changed By Updated Field Previous Value => Replaced By
    Thu Oct 8 07:59:19 2015guerbyStatusIn Progress=>Done
    Show feedback again

    Back to the top

    Powered by Savane 3.1-cleanup