64-bit CUDA 7.5 on the TX1

In this developer blog post (https://devblogs.nvidia.com/parallelforall/nvidia-jetson-tx1-supercomputer-on-module-drives-next-wave-of-autonomous-machines/) it mentions that CUDA 7.5 will be available in a future release. Is there a timeline on this release?

We need this as we are trying to build a 64 bit application, but the CUDA libraries installed by JetPack are 32 bit.

Our application requires CUDA 7.5 features, but on this forum post (https://devtalk.nvidia.com/default/topic/904664/jetson-tx1/cuda-7-5-for-tx1/) dusty_nv mentions that some features from CUDA 7.5 have been backported to the TX1’s version of CUDA 7.0. Is there a list of the features that have been backported in this way?

Instead of releasing a separate CUDA toolkit 7.5 for Jetson TX1, features and maintenance fixes relevant to TX1 were backported and released as v7.0

Currently the sample root filesystem for the Jetson TX1 Developer Kit is armhf build (32-bit). The 64-bit userspace will be released in the next update.

Thank you for your quick response. We have another question to ensure we are not misreading anything.

Will the 64-bit userspace come with a 64-bit version of CUDA 7.0 for the TX1?

If so, is there an expected date for this release?

Hi Dusty!

do you have any info about the estimated release date for the update?

I got my TX1 yesterday and I am currently absolutely disappointed about the missing ability to compile to 64 Bit. That was the only reason to spend 600 Euros: To get 64 Bit… (and every NVIDIA commercial / info says TX1 = 64 Bit)

Thanks,
Andreas

You can compile 64-bit onboard (and also cross-compile) and can run 64-bit kernel modules.
See [url]https://github.com/dusty-nv/jetson-scripts/blob/master/install_gcc_arm64.sh[/url] The update is nearing release.

Thanks Dusty for your shellscript.
Unfortunately apt wants to dump my CUDA environment, when I execute it?
Am I doing something wrong or can I install a 64 Bit version compatible to the 64 Bit gcc?
On my first attempt I could not find something…

The 64-bit version of gcc that the script installs is incompatible with 32-bit gcc that CUDA currently uses. Currently I compile my 64-bit code like this, then restore CUDA (although maybe you can find a way to sandbox the two packages using apt). In the pending update, there will be full 64-bit support.

I also have a pressing need for a full 64-bit supported version. Is the R24.1 Early access “BETA” 64-bit or is it still a 32-bit user space?

Hello Dusty,

more than one month has passed since your last statement on the upcoming Release with a 64 Bit Userspace.

What timeframe are you talking about? Days? Weeks? Months?
Do you have any news? Any Beta release?

Please give us some info on how Nvidia treats this topic.

Thanks,
Andreas

You may be interested in this (R24.1 at this time), which is 100% 64-bit:
[url]https://developer.nvidia.com/embedded/linux-tegra[/url]