CUDA 8 Features Revealed

Originally published at:

Today I’m excited to announce the general availability of CUDA 8, the latest update to NVIDIA’s powerful parallel computing platform and programming model. In this post I’ll give a quick overview of the major new features of CUDA 8. Support for the Pascal GPU architecture, including the new Tesla P100, P40, and P4 accelerators; New Unified…

Thanks for the great presentations today at GTC. Is CUDA 8 going to have support for Visual Studio 2015?

Yes, the compiler with Visual Studio 2015 will be one of the host compilers that nvcc supports in CUDA 8.

Please, make it come to us as soon as possible! Can't wait any more for it as gcc-5 has been there for really a long time......

Release it already!!

Thanks Mark for the post, I have a critical question. The new shared memory model seems like a great abstraction over the current model. Basically what you are saying is that only the Pascal arcitecture will support this new feature on CUDA 8 due to hardware limitations ? (even on the Tesla K40 series)

What I'm saying is that Unified Memory is supported on all CUDA GPUs starting with Kepler (including K40), but with limitations -- see Pascal adds specific hardware capabilities (large 49-bit VA and hardware page faulting / migration capability) that expand the Unified Memory capabilities and improve performance, as described in the post.

Microsoft Visual Studio 2015 Update >= 2 is not supported yet! Do you have plans to support it in the release version?

I would like to know that can I use cuda7.5 or even 7.0 to compile with GTX1070; and if GTX1070 is only compatible with cuda8, can I use it just like the other versions in matlab? Someone told me that cuda8.0 is not supported in matlab2016b which is the newest matlab.
I'm now using matconv and want to use GTX1070 as my GPU solution

So does this mean that Pascal graphic card support is coming to Macs or only unified memory? Not fully understanding this.

Microsoft doesn't even offer to download older versions of VS 2015. Please add support to new versions.

Yes, I am trying to run some tests of CUDA v8.0 but can not since newer versions of Microsoft Visual Studio 2015 are not supported and I can not download an older version. Please, add support for newer versions of Microsoft Visual Studio 2015.

We are working to support in the CUDA 8 general availability release.

When can we expect CUDA 8 for MacOS?

You can get the CUDA 8 release candidate for MacOS now if you are a registered developer. I'm using it on my Macbook Pro.

Right now it's Unified Memory only, which means Kepler-class Unified Memory. So you get the ease of programming benefits but not all the performance benefits. You can use cudaMallocManaged on all OSes now, though, which helps with portability.

Thanks. Will CUDA 8 support gcc versions later than 5.3? :)

I believe 5.4 will be supported with CUDA 8.0.

I would also like to see 1070 supported under OSX, I will buy one as soon as there are drivers.

Hi, is there a release date for CUDA 8? I really need to use it with VC14 Update 3