cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

nou
Exemplar

AMD APP SDK 2.5

is out?

so download page of SDK get updated to 2.5. but i cant download it. a get some access denied message.

0 Likes
24 Replies
nou
Exemplar

bump

0 Likes

On the download page:

Launch Date - 08/03/2011...

0 Likes

derp

0 Likes

Well, it's 8/3 now and still no luck. 🙂

Does anyone have any links to release notes for 2.5?

0 Likes

i think these are parts of it: http://blogs.amd.com/developer/2011/08/02/amd-app-sdk-2-5-provides-enhanced-performance-and-major-new-capabilities/

0 Likes

Access denied here too 8(

0 Likes

http://developer.amd.com/sdks/...ads/Pages/default.aspx <-- the download page has links to 2.5.
0 Likes

Originally posted by: MicahVillmow http://developer.amd.com/sdks/...ads/Pages/default.aspx <-- the download page has links to 2.5.


Those are the links that are giving Access Denied messages.

0 Likes

I successfully downloaded Winx64 version, but there are RC2 characters in the filename. Is it release version?

0 Likes

I have to say that the Linux package is a bit unprofessional.

First, the tar archive is not rooted in a directory so all files are dumped in whatever directory you run the tar command from.  That wouldn't be *so* bad except that the archive contains a file with the exact same name as the archive. So, the original archive gets overwritten.  That's just bad tar-fu.

Second, the forceful restart of the system is a bit much.  I know the README file mentions this, but please, at least make it a prompt.  People who blindly install without reading the README are going to get a good surprise with this one.

Third, having to edit the install script just to change the installation prefix is just bad.

I know these are fairly minor issues in the overall picture, but as an administrator of GPGPU machines I find it hard to trust AMD software for Linux when the installer is so obviously a half-baked after-thought.  

0 Likes

Originally posted by: jholewinski I have to say that the Linux package is a bit unprofessional.

 

First, the tar archive is not rooted in a directory so all files are dumped in whatever directory you run the tar command from.  That wouldn't be *so* bad except that the archive contains a file with the exact same name as the archive. So, the original archive gets overwritten.  That's just bad tar-fu.

 

Second, the forceful restart of the system is a bit much.  I know the README file mentions this, but please, at least make it a prompt.  People who blindly install without reading the README are going to get a good surprise with this one.

 

Third, having to edit the install script just to change the installation prefix is just bad.

 

I know these are fairly minor issues in the overall picture, but as an administrator of GPGPU machines I find it hard to trust AMD software for Linux when the installer is so obviously a half-baked after-thought.  

 

I have to agree, lots of unpleasant surprises here.

 

0 Likes

There seems to be some inconsistencies regarding supported HW:

In AMD-APP-V2.5-RC2/AMD-APP-SDK-v2.5-RC2-lnx32/docs/opencl/AMD_APP_SDK_Getting_Started_Guide_v2.5.pdf

There is no mention of APUs as far as I can see, nor of the GPU parts used, such as  AMD Radeon HD 6300 series Graphics for the E350. 

 

At http://developer.amd.com/sdks/AMDAPPSDK/pages/DriverCompatibility.aspx the APUs are mentioned directly, but there is no mention of which particular features are supported.

It would be nice if there was a `make test` directive which could quickly capture capabilities and potential issues. 

I found AMD-APP-V2.5-RC2/AMD-APP-SDK-v2.5-RC2-lnx32/bin/x86/clinfo which seems to show very helpful information; This command should probably be mentioned in at least one of: post-install output, README.txt, documentation pdf files. 

0 Likes

Originally posted by: jholewinski

First, the tar archive is not rooted in a directory so all files are dumped in whatever directory you run the tar command from.

If someone at AMD would be so kind as to run these commands and update the linked file with the one generated here, it would solve the first issue.  Thanks!

mkdir AMD-APP-SDK-v2.5-lnx6 ;

tar -xvzf AMD-APP-SDK-v2.5-lnx64.tgz -C AMD-APP-SDK-v2.5-lnx64 ;

mv AMD-APP-SDK-v2.5-lnx64.tgz AMD-APP-SDK-v2.5-lnx64.tgz.backup ;

tar -czvf AMD-APP-SDK-v2.5-lnx64.tgz AMD-APP-SDK-v2.5-lnx64

 

0 Likes

I noticed some other strange things about the new linux SDK (2.5) packaging.  In the following directories the file permissions have changed from the previous linux SDK (2.4).

AMDAPPROOT/bin/x86/

AMDAPPROOT/bin/x86_64/

AMDAPPROOT/include/CL/

Files have permission "-rwxr--r--." rather than "-rwxr-xr-x.", which is especially strange for clinfo.

 

Also, why doesn't AMD just include a bash script in AMDAPPROOT/bin/ to set all the paths, something like this:

 

BASE=/opt/amd/app VERSION=2.5 if [ "`uname`" = "Darwin" ]; then if [ -z "${C_INCLUDE_PATH}" ]; then C_INCLUDE_PATH="${BASE}/${VERSION}/include"; export C_INCLUDE_PATH else C_INCLUDE_PATH="${BASE}/${VERSION}/include:${C_INCLUDE_PATH}"; export C_INCLUDE_PATH fi else if [ -z "${C_INCLUDE_PATH}" ]; then C_INCLUDE_PATH="${BASE}/${VERSION}/include"; export C_INCLUDE_PATH else C_INCLUDE_PATH="${BASE}/${VERSION}/include:${C_INCLUDE_PATH}"; export C_INCLUDE_PATH fi fi if [ "`uname`" = "Darwin" ]; then if [ -z "${CPLUS_INCLUDE_PATH}" ]; then CPLUS_INCLUDE_PATH="${BASE}/${VERSION}/include"; export CPLUS_INCLUDE_PATH else CPLUS_INCLUDE_PATH="${BASE}/${VERSION}/include:${CPLUS_INCLUDE_PATH}"; export CPLUS_INCLUDE_PATH fi else if [ -z "${CPLUS_INCLUDE_PATH}" ]; then CPLUS_INCLUDE_PATH="${BASE}/${VERSION}/include"; export CPLUS_INCLUDE_PATH else CPLUS_INCLUDE_PATH="${BASE}/${VERSION}/include:${CPLUS_INCLUDE_PATH}"; export CPLUS_INCLUDE_PATH fi fi case "$1" in m32) if [ -z "${PATH}" ]; then PATH="${BASE}/${VERSION}/bin/x86"; export PATH else PATH="${BASE}/${VERSION}/bin/x86:${PATH}"; export PATH fi if [ "`uname`" = "Darwin" ]; then if [ -z "${DYLD_LIBRARY_PATH}" ]; then DYLD_LIBRARY_PATH="${BASE}/${VERSION}/lib/x86"; export DYLD_LIBRARY_PATH else DYLD_LIBRARY_PATH="${BASE}/${VERSION}/lib/x86:${DYLD_LIBRARY_PATH}"; export DYLD_LIBRARY_PATH fi else if [ -z "${LD_LIBRARY_PATH}" ]; then LD_LIBRARY_PATH="${BASE}/${VERSION}/lib/x86"; export LD_LIBRARY_PATH else LD_LIBRARY_PATH="${BASE}/${VERSION}/lib/x86:${LD_LIBRARY_PATH}"; export LD_LIBRARY_PATH fi fi ;; m64) if [ -z "${PATH}" ]; then PATH="${BASE}/${VERSION}/bin/x86_64"; export PATH else PATH="${BASE}/${VERSION}/bin/x86_64:${PATH}"; export PATH fi if [ "`uname`" = "Darwin" ]; then if [ -z "${DYLD_LIBRARY_PATH}" ]; then DYLD_LIBRARY_PATH="${BASE}/${VERSION}/lib/x86_64"; export DYLD_LIBRARY_PATH else DYLD_LIBRARY_PATH="${BASE}/${VERSION}/lib/x86_64:${DYLD_LIBRARY_PATH}"; export DYLD_LIBRARY_PATH fi else if [ -z "${LD_LIBRARY_PATH}" ]; then LD_LIBRARY_PATH="${BASE}/${VERSION}/lib/x86_64"; export LD_LIBRARY_PATH else LD_LIBRARY_PATH="${BASE}/${VERSION}/lib/x86_64:${LD_LIBRARY_PATH}"; export LD_LIBRARY_PATH fi fi ;; *) echo "ERROR: Unknown switch '$1'. Accepted values: m32, m64" ;; esac

0 Likes

If AMD APP SDK included a bash script like above (e.g., appvars.sh) and extracted the SDK contents to /opt/amd/app/2.5/, and made a symbolic link to the current SDK version (cd /opt/amd/app/ ; ln -s 2.5 current); then it would be very simple to set the paths system wide with a bash script in /etc/profile.d/ (e.g., app.sh).  I would prefer this method as it is more flexible updating and switching between SDK versions and would give better control over system wide vs. individual user path settings.

# # AMD APP with OpenCL support # case "$(uname -m)" in i[3-6\d]86) MBIT="m32";; x86_64) MBIT="m64";; esac if [ -e /opt/amd/app/current/bin/appvars.sh ]; then source /opt/amd/app/current/bin/appvars.sh "$MBIT" fi case $DISPLAY in '') export DISPLAY=:0;; *) ;; esac

0 Likes

Originally posted by: jholewinski I have to say that the Linux package is a bit unprofessional.

 

First, the tar archive is not rooted in a directory so all files are dumped in whatever directory you run the tar command from.  That wouldn't be *so* bad except that the archive contains a file with the exact same name as the archive. So, the original archive gets overwritten.  That's just bad tar-fu.

 

Second, the forceful restart of the system is a bit much.  I know the README file mentions this, but please, at least make it a prompt.  People who blindly install without reading the README are going to get a good surprise with this one.

 

Third, having to edit the install script just to change the installation prefix is just bad.

 

I know these are fairly minor issues in the overall picture, but as an administrator of GPGPU machines I find it hard to trust AMD software for Linux when the installer is so obviously a half-baked after-thought.  

 

Lets not forget the first line of Install-AMD-APP.sh is probably missing an echo command.

I have to say this is one of the worst Linux 'installer' I have ever come across.

Now I admit, I didn't read the README.txt but after installing a few of the previous stream SDK's I didn't think it would be bad, assuming I'd probably just be prompted for a few things like install directory etc.

Nope, I see an error message and then down goes the multi-user frontend to my GPU cluster.

I'm placing the blame for that on myself but I agree with jholewinski on everything he said. Adding, the install script is one of the worse pieces of perl I have ever seen.

0 Likes

The Linux 2.5 packages have the updated Linux Installer files, could you please try the latest/updated packages.

0 Likes

Okay, I checked windows driver. All is working. Kernel Analyzer works too now. I also checked that it is no longer needed to attach AMD accelerator to display in order to have it running. It works!

0 Likes

on windows global memory size is 800MB on linux it is 512MB. why?

also what exactly is meaned by multGPU support on windows?

0 Likes

I still can't download 32bit version of SDK for Windows. Is there any alternate download link?

0 Likes
foreignobjects
Journeyman III

The above link : http://developer.amd.com/tools-and-sdks/heterogeneous-computing/amd-accelerated-parallel-processing-...

redirects to the app-sdk generic page.

The download page contains the latest APP-SDK

The archive page contains APP-SDK v2.8 , v2.7 , and v2.6 but not v2.5

The direct download links : http://developer.amd.com/Downloads/AMD-APP-SDK-v2.5-lnx32.tgz and http://download2-developer.amd.com/Downloads/AMD-APP-SDK-v2.5-lnx32.tgz

yield 403 Forbidden messages when attempting wget.

Are there any alive mirrors with APP-SDK-v2.5 around?? Does anyone have a spare copy?

0 Likes

Wonderful! Works!

0 Likes

If anyone is looking for the windows 64 version, here it is:

http://developer.amd.com/amd-license-agreement/?f=AMD-APP-SDK-v2.5-Windows-64.exe

0 Likes