CUDA Toolkit and SDK 2.3 released

Is the debugger not included in the 64-bit Linux archives? I ran the ubuntu9.04 and rhel5.3 .run installer with --list and it looks like the archives do not contain bin/cuda-gdb.

The CrossCompiling support seems good, but in Win32 installs there seems to be an Error (or the installer could not overwrite my paths?)

The Environment Variable set as Cuda path points to C:/CUDA/bin64 and C:/CUDA/lib64 as the Toolkit paths are C:/CUDA/bin and C:/CUDA/lib. I changed this manually, but I guess these paths should be only used for 64bit versions. Will install the 64bit version on my Vista64 later and try cross compiling…

PS: I tried to install the 2.3beta 2h before I got the mail about the 2.3final release :">

that is a pretty big mistake…

as far as Win32 paths go, I just told the driver guys about that too.

This is what I’m interested in as well, 2.3 doesn’t work on my macbook with its 9400M where as 2.2 did. Also in the documentation for the SDK it says to run “make” in “/Developer/GPU Computing” but there is no makefile there, I assume they mean to say run “make” in “/Developer/GPU Computing/C”.

None of the included sdk sampels worked with the “normal vista64 english 190.38 release”-drivers from the normal driver download. Had to install the general 190.38 from here to make things work. Anybody else experiencing this?

Oops, also missing the cuda-gdb bin for Linux.

In case anyone else needs the answer I found this:

in the Getting Started with Linux document. I will assume it’s up to date and applies to 2.3

Do desktop and laptop share the same driver?

I completed install cuda2.3 driver on desktop.

But error occurs on laptop - couldn’t find supported hardware…

My laptop is Lenono T61p model - window7 32bit, Quadro FX 570m

There are websites that offer “modded INFs” for laptops that allow using the latest Desktop driver releases - with the risk of disabling certain vendor specific features, such as power management.

Check the forums of www.laptopvideo2go.com

The SDK doesn’t compile on OpenSUSE 11.1 32-bit:

teslaspule: /home/user/goeddeke/nobackup/cuda/SDK/C[1034]>make

make[1]: Entering directory `/data/peppone/goeddeke/cuda/SDK/C/common'

In file included from src/bank_checker.cpp:15:

./../common/inc/bank_checker.h:34: warning: type qualifiers ignored on function return type

./../common/inc/bank_checker.h:220: warning: type qualifiers ignored on function return type

src/bank_checker.cpp:57: warning: type qualifiers ignored on function return type

src/bank_checker.cpp:259: warning: type qualifiers ignored on function return type

src/cutil.cpp:29:27: error: builtin_types.h: No such file or directory

In file included from ./../common/inc/stopwatch.h:18,

				 from src/cutil.cpp:34:

./../common/inc/stopwatch_base.h:56: warning: type qualifiers ignored on function return type

./../common/inc/stopwatch_base.h:60: warning: type qualifiers ignored on function return type

In file included from ./../common/inc/stopwatch_base.h:72,

				 from ./../common/inc/stopwatch.h:18,

				 from src/cutil.cpp:34:

./../common/inc/stopwatch_base.inl:67: warning: type qualifiers ignored on function return type

./../common/inc/stopwatch_base.inl:78: warning: type qualifiers ignored on function return type

In file included from ./../common/inc/stopwatch.h:25,

				 from src/cutil.cpp:34:

./../common/inc/stopwatch_linux.h:46: warning: type qualifiers ignored on function return type

./../common/inc/stopwatch_linux.h:50: warning: type qualifiers ignored on function return type

./../common/inc/stopwatch_linux.h:126: warning: type qualifiers ignored on function return type

./../common/inc/stopwatch_linux.h:143: warning: type qualifiers ignored on function return type

In file included from src/cutil.cpp:34:

./../common/inc/stopwatch.h:35: warning: type qualifiers ignored on function return type

In file included from src/cutil.cpp:35:

./../common/inc/bank_checker.h:34: warning: type qualifiers ignored on function return type

./../common/inc/bank_checker.h:220: warning: type qualifiers ignored on function return type

make[1]: *** [obj/release/cutil.cpp.o] Error 1

make[1]: Leaving directory `/data/peppone/goeddeke/cuda/SDK/C/common'

make: *** [lib/libcutil.so] Error 2

My own apps compile and run well with 2.3, so the installation went fine.

This is the error I get when using 2.2 cuda-gdb with 2.3 files:

CUDA GDB could not access debug API version 20090303 and is incompatible with the installed CUDA driver

There was some mention in the recent past of a build of CUDA/SDK which linked against the dynamic crt for Windows (/MD switch). Whatever became of this? The application we currently use CUDA for is a C++/CLI application, so we’re required to build our app with /MD. If there is no longer any plan to release an /MD build, what do the users of CUDA.NET and people in my situation do?

just a quick heads-up that I’ve not been able to pull either of the documents below from the links in the initial post:

CUDA GDB 2.3 User Manual or
CUDA C Programming Guide 2.3

I’m getting a ‘page can not be found’ error for both.

Thanks!

Heads up for anyone using GPUWorker with CUDA 2.3: The high-level API cuda_runtime.h now specifies template functions for cudaMalloc, cudaMallocPitch, cudaHostAlloc, etc… to handle the conversion from (T**) to (void**). GPUWorker uses boost::bind to wrap up function calls, which gets very confused at these templates and does not compile complaining:

error: no matching function for call to ‘bind(<unknown type>, void**, long unsigned int)’

It also happens with bash - and as a reply to tmurray: yes, hitting enter does seem to work, both with bash and with tcsh.

this is odd, because I looked at the programming guide on my laptop this morning, but now I’m getting file not found on my desktop. will send emails accordingly.

I had a similar problem compiling CUDPP 1.1 under OpenSuSE 1.1 64-bit and had to “patch” several .h files with additional includes until I got CUDPP to compile. Mysteriously, compiling with a small Python script of my own, which mimicks what the Makefile should do, to generate a .so dynamic library for CUDPP, everything compiled and linked without a hitch. I suspect something strange is happening within common.mk. The 2.3 SDK compiled very cleanly on 64-bit OpenSuSE 11.1. I no longer have any system with 32-bit, so I can’t say much about that.

I installed the 2.3 driver, sdk and toolkit. Things seemed to go well. I added the lib64 statement to my LD_LIBRARY_PATH and source the relevant file. When I attempt to make one of the projects in the nvidia_cuda_sdk directory I get the following:

make[1]: Entering directory /home/mbibby/nvidia/nvidia_cuda_sdk/common' make[1]: Leaving directory /home/mbibby/nvidia/nvidia_cuda_sdk/common’
make[1]: Entering directory /home/mbibby/nvidia/nvidia_cuda_sdk/common' make[1]: Leaving directory /home/mbibby/nvidia/nvidia_cuda_sdk/common’
make[1]: Entering directory /home/mbibby/nvidia/nvidia_cuda_sdk/common' make[1]: Leaving directory /home/mbibby/nvidia/nvidia_cuda_sdk/common’
make -C projects/BlackScholes/
make[1]: Entering directory /home/mbibby/nvidia/nvidia_cuda_sdk/projects/BlackScholes' /usr/bin/ld: skipping incompatible /usr/local/cuda/lib/libcudart.so when searching for -lcudart /usr/bin/ld: skipping incompatible /usr/local/cuda/lib/libcudart.so when searching for -lcudart /usr/bin/ld: cannot find -lcudart collect2: ld returned 1 exit status make[1]: *** [../../bin/linux/release/BlackScholes] Error 1 make[1]: Leaving directory /home/mbibby/nvidia/nvidia_cuda_sdk/projects/BlackScholes’
make: *** [projects/BlackScholes/Makefile.ph_build] Error 2

Both the /usr/local/cuda/lib and lib64 directories contain:

lrwxrwxrwx 1 root root 14 Jul 21 22:34 libcudart.so → libcudart.so.2
lrwxrwxrwx 1 root root 16 Jul 21 22:34 libcudart.so.2 → libcudart.so.2.3
-rwxr-xr-x 1 root root 260968 Jul 21 22:34 libcudart.so.2.3

All of which looks good. I was able to make any/all of the projects using 2.2.

I am running RHEL 5.2 on a dual quad X86 machine.

Any help would be appreciated.

Malcolm

Since posting this comment/request, I have checked the environment and both the PATH and LD_LIBRARY_PATH entries are correct. Has anybody else seen this problem?

I have found the ‘problem’: the Makefile in the SDK directory needs the ’ LIB = ’ statement to be changed to lib64.

Is anyone else having problems with cudadriver_2.3_linux_64_190.16.run? It runs smooth on my desktop machine but it doesn’t seem to work on my HP8710w laptop, the previous 190.09 beta driver worked but this one just gives me a black screen upon starting X and completely freezes.

N.

There is usually a driver for notebooks but not this time. Should I just wait for it ? (I’m using vista64 on geforce 8600m gt)

Thanks