by Devang Sachdev

The CUDA on ARM development kit gets official today, with a final spec-sheet and its crowdsourced codename.

Last week, we asked you, dear reader, for a few good codename suggestions. Within a matter of days, we were inundated with hundreds of thoughtful codename suggestions. Thanks to the help of our enthusiastic and creative NVIDIA community, we now have an official codename for our CUDA on ARM dev kit.

So, without further ado, meet CARMA.

Powered by a Tegra 3 quad-core ARM-based processor and an NVIDIA CUDA-enabled GPU, the CARMA DevKit is being developed to support energy-efficient HPC projects using ARM-based GPU computing. In fact, this technology will power the Barcelona Supercomputing Center’s ARM-based GPU supercomputer.

CARMA specs:

  • Energy-Efficient Tegra3 Quad-Core ARM A9 processor with 2GB Processor Memory
  • NVIDIA Quadro 1000M GPU with 96 CUDA Cores and 2GB GPU Memory
  • 270 Single Precision GFlops Performance
  • 4x PCIe Gen1 link connecting processor to GPU
  • 1 Gigabit Ethernet Network support
  • Direct attach storage support with SATA port
  • HDMI and DisplayPort
  • USB port

Congratulations to Daniel Holt for submitting the winning codename. Daniel will receive a free CARMA DevKit, fresh from our first-run production batch, when it becomes available in Q2 2012.

Interested in more details on the CARMA DevKit? Find more comprehensive specs and pre-order information by clicking here.

Will you be getting a CARMA DevKit? Tell us how you plan to use it in the comments below!

Similar Stories

  • Panos

    Congratulations to Daniel! Great news! I would like to order one when they become available, in order to port and develop software to be used in the next generation of radio astronomical arrays. Power efficiency is going to be a key design factor.

  • Nikolai Kondrashov

    So, is it “Colorado”, or “Carma” :)? And I guess you forgot to put global hostname into the “details” link.

    Anyway, a great name :)!

  • Doug

    I would like to use it, in part, for the generation and manipulation of arrays of on-the-fly random numbers to be used as the foundation for a new system that rapidly prototypes clever, new Twitter-related responses/suggestions so that I can win one of those fancy new CARMA devkits.

  • Doug

    Wait, now I think I’m stuck in a loop…. D’oh!

  • Anonymous

    hi Nikolai, we just fixed the typo. It is CARMA. For simpler access to specs link you can also try

  • Anonymous


  • Anonymous

    Radio astronomical arrays have benefited with use of CUDA GPUs, porting to ARM will enable further efficiency. Good luck on the future project Panos.

  • Pablo Odorico

    I would use one for prototyping client-side decompression algorithms, as part of my R&D on next-generation DTV set-top-boxes for the Argentinian Goverment.

  • Pablo Odorico

    Does it have HDMI 3D and Quad Buffer Stereo support?

  • Anonymous

    hi Pablo, at launch it will not support HDMI 3D and Quad Buffer Stereo. However, we will look into these as future enhancements.

  • Anders Salvesen

    What kind of machine and purpose will this be used for? 
    Personal nvidia based computer? 
    game developing?

  • Cybers Androidz

    i would use it for replacing my old computer…:(
    and of course for developing some new application and system (mainly about automation and image processing) that use ARM as its main platform, because of its efficiency on energy use 🙂

  • Anonymous

    hi Anders, the CARMA Devkit is aimed at enabling energy efficient computing research and development through use of ARM processors and CUDA GPUs. I expect wide range of scientific and embedded applications to be developed on the Devkit.

  • Harsha

    image processing and real time use of machine vision for Augmented Reality Applications. Hopefully we will have a first OpenCL implementation on TEGRA along with a CARMA implementation.  It exciting days of super computing on low power mobile devices.

  • Erkan Tairi

    First of all, congratulations to Daniel Holt for suggesting this beautiful and meaningful codename. About CARMA, we are really looking forward to have at least one of them, which we, as a company are going to use in our researches. We are doing some research about developing Exascale systems and software, which will be both energy efficient and powerful. And, I think that for advancing in our research, CARMA is a must!

  • Paul Normand

    Could it support an OCZ revo drive?

  • Yipley McKwok

    Our research group in the U. of TX system is planning to pre-register to receive the devkit. How much does each unit cost? Thank you very much.

  • Anonymous

    Hi Yipley, thanks for pre-registering. 
    The price for the devkit is not available yet. We will list it on the webpage once available. Also you could follow @GPUComputing:twitter  with #NVIDIACARMA tag to find out as soon as we make it available.

  • Anonymous

    Hi Paul
    At launch the CARMA DevKit will not support PCIe based SSD. However SATA port is available for storage. We may consider it as a future enhancement.

  • Garrett Rhodes

    I would love to try Linux Media Center + file server on a single low voltage/size platform!

  • rtfss none

    lot of interesting questions:
    I want to know if also a ARM OpenCL driver for Quadro GPU will be supplied..
    Lastly is Nvidia investing also in porting its great OpenGL drivers (4.2 at time) to ARM architecture
    so we can use efficient CUDA/OpenGL interop
    (would be better if quadro card has video output but if not Nvidia could use similar to Optimus tech we framebuffer gets passed via PCIe bus to Tegra GPU for display)
    *will sdk avaiable to everyone without dev cards (at least registered developers as I am 🙂 ) so for example be avaiable a Windows 8 ARM SDK so we can compile apps for this board..
    great christmas..

  • Anonymous

    Hi Rtffs

    Thanks for reading the blog. At launch we plan to support CUDA on the GPU. We may look into OpenGL and OpenCL support as future enhancements. It would certainly be great to have Optimus supported in the future as well. It is a good suggestion to have the SDK available to registered developers (w/ or w/o the CARMA Devkit). We will look into it. Have a great christmas too.

  • Christopher Bergström

    This is awesome!  I’d love for us to get one so we could port ENZO our GPGPU compiler/solution.  What level of documentation will be available to early developers?

  • Anonymous

    Hi Christopher

    We don’t have details on documentation yet but we will try our best to support developers with docs and libs.

  • Baybal Ni

    I want it!

  • Baybal Ni

    Hmm, hardware specs says that it have 1 gbit Ethernet iface, but i see two on the picture. Is one inactive?

  • Monish Tandale

    Are any commercial products (laptops,mobile devices) currently available or planned with the same specs as that of CARMA?

  • Anonymous

    hi Monish
    I am not aware of commercially shipping products with CARMA type configuration yet. 

  • Pete Eby

    The specs say the OS is an Ubuntu derivative. Will this be an image that is intended to be installed on the SATA storage or as a PXE boot image which has the CUDA toolkit already installed? I rather like a PXE bootable solution as then you can just skip any local storage if not needed.

  • Donald Becker

    Internally we are developing using a PXE-like network boot with a NFS mounted root file system.

    The release boards will likely be configured to boot and run from a local flash file system by default.

    I say “PXE-like” because PXE isn’t really defined for ARM.  But the combination of DHCP and TFTP servers that most people use for x86 PXE clients will work fine.  You just pass a uImage format kernel instead of going through an intermediate loader such as pxelinux.

    The initial booting process is based on Uboot (thus the ‘uImage’ format kernel), and thus is fairly quick.  It takes only a few seconds from applying power until the first DHCP request. 

  • Anonymous

    Thanks Don!

  • don

    I would like to use it in simulation ,in a industrial setting for simulating high risk tasks, prior to actually doing them in realtime to help forsee risk and hazzards and to minimise actual time to project completion…should be great for light CAD work,trouble shooting..

  • Anonymous

    We are planning to use Tesla GPU along with ARM CPU from Marvell. Will the CARMA work in the described configuration? Thanks.