CUDA 11.8

[blueberry]

It seems CUDA 11.8 is out. I hope @jcuda can release the support soon. Of course, I offer to build linux binaries, as always.

[jcuda]

Thanks for the heads-up! I'll try to schedule an update. (Maybe @corepointer also wants to be notified here - I'm still unsure about the best approach for that GLIBC issue that we stumbled over in some previous releases)

[corepointer]

glad to help =)

[jcuda]

The update for CUDA 11.8.0 has been done. The native libraries for Linux should be built based on the tag version-11.8.0-RC00, which is currently the same as the master state.

^{(Note: There are extensions of the API that are not supported (e.g. updates/extensions for the graph-based execution models for core CUDA that have been introduced in 11.6, and callbacks for cuSolverDn that have now been added in 11.8). Keeping all this up to date is a task that is just too time-consuming for me to do for free in my spare time. But having updated binaries for newer CUDA versions should be a reasonable baseline for the updates)}

@blueberry and @corepointer Considering the GLIBC issue, we might try the same process as last time, where @corepointer built the binaries and @blueberry tested the result, if this is OK for both of you.

[blueberry]

Thank you. OK, I'll wait for @corepointer's binaries.

On Sat, Dec 3, 2022 at 5:31 PM Marco Hutter @.***> wrote:

The update for CUDA 11.8.0 has been done. The native libraries for Linux should be built based on the tag version-11.8.0-RC00, which is currently the same as the master state.

(Note: There are extensions of the API that are not supported (e.g. updates/extensions for the graph-based execution models for core CUDA that have been introduced in 11.6, and callbacks for cuSolverDn that have now been added in 11.8). Keeping all this up to date is a task that is just too time-consuming for me to do for free in my spare time. But having updated binaries for newer CUDA versions should be a reasonable baseline for the updates)

@blueberry https://github.com/blueberry and @corepointer https://github.com/corepointer Considering the GLIBC issue, we might try the same process as last time, where @corepointer https://github.com/corepointer built the binaries and @blueberry https://github.com/blueberry tested the result, if this is OK for both of you.

— Reply to this email directly, view it on GitHub https://github.com/jcuda/jcuda-main/issues/55#issuecomment-1336193820, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAAZ2JBAOXLIA3DDRMEMQCLWLNYUPANCNFSM6AAAAAASGTRUXY . You are receiving this because you were mentioned.Message ID: @.***>

[corepointer]

Sorry for the delay. Binaries are here Seems Nvidia is way quicker than we are (ok, they've got the person-power and the monetary incentives) - CUDA 12 is out.

[jcuda]

Thanks for the binaries (and the heads-up about CUDA 12). I'll try to do the release of 11.8 before the holidays (cannot promise, but will try).

And ... *sigh* ... I'd say that I'd try to do the update for 12 during the holidays, but ... whenever there's a few ""free"" days, I tend to squeeze so many TODOs into these days that I only get a fraction of that done after all...

[jcuda]

JCuda 11.8.0 is on its way into Maven Central, and should soon be available under the usual coordinates:

<dependency>
    <groupId>org.jcuda</groupId>
    <artifactId>jcuda</artifactId>
    <version>11.8.0</version>
</dependency>
<dependency>
    <groupId>org.jcuda</groupId>
    <artifactId>jcublas</artifactId>
    <version>11.8.0</version>
</dependency>
<dependency>
    <groupId>org.jcuda</groupId>
    <artifactId>jcufft</artifactId>
    <version>11.8.0</version>
</dependency>
<dependency>
    <groupId>org.jcuda</groupId>
    <artifactId>jcusparse</artifactId>
    <version>11.8.0</version>
</dependency>
<dependency>
    <groupId>org.jcuda</groupId>
    <artifactId>jcusolver</artifactId>
    <version>11.8.0</version>
</dependency>
<dependency>
    <groupId>org.jcuda</groupId>
    <artifactId>jcurand</artifactId>
    <version>11.8.0</version>
</dependency>
<dependency>
    <groupId>org.jcuda</groupId>
    <artifactId>jcudnn</artifactId>
    <version>11.8.0</version>
</dependency>

@blueberry If you have a chance to test it and see whether everything works as expected, that would be great.

[jcuda]

@blueberry If you had a chance to try out 11.8.0, I assume that there are no issues. I'll still leave this open for a while, because the update for CUDA 12 is not even really in the pipeline yet, due to too many other tasks piling up.

[neocoretechs]

@blueberry If you had a chance to try out 11.8.0, I assume that there are no issues. I'll still leave this open for a while, because the update for CUDA 12 is not even really in the pipeline yet, due to too many other tasks piling up.

You might be able to build CUDA 12.1.0 for *nix from the ugliness that is https://github.com/neocoretechs/jcuda-all Or it may not help at all. But it WILL compile.