How can nvcc stop working when you install a win patch?? And if it would happen, you should probably go to microsoft to complain that a user-space program stops working after a patch.
Funny, how people that complain about the level of support of NVIDIA (which is quite good as far as I am concerned, especially on the developer site, but also here) are buying win7, where you basically get no support at all…
As a commercial software developer, I admit CUDA does have its raw points being a new technology, but I have no complaint with the support or tools themselves.
There are clearly annoyances, which is not unexpected. There are even more in OpenCL…
But actual BUGS are rare. I have reported 4, and all four were fixed. One minor bug did take 5 months, but the rest were reasonably fast.
Other companies I deal with have their own SDKs which I depend on, have had severe, showstopper, bugs which persisted for YEARS.
That said, the remainder of your concerns are slightly valid. I also agree the profiling tools need work… to be honest I haven’t had much success using them profitably, instead I tend to do my own crude profiling hacks. And the debuggers, especially in Linux, are still crude, but also have progressed immensely even in the past few months so I cannot complain.
So, most of the concerns about use of CUDA in commercial or professional apps aren’t due to poor tools or support, but just because of the bleeding edge of the whole GPGPU field. I happen to LIKE that raw exciting edge, but it could indeed be overwhelming to a traditional programmer who tries to map his CPU experience onto the GPU and gets frustrated when the concepts don’t transfer.
Case in point: CUDA 3.0 installed with Visual Studio 2008 SP1 on Windows XP machine, everything works fine.
Now, to get the benefit of Nexus/NInsight, CUDA 3.0 installed with Visual Studio 2008 SP1 on Windows 7 machine (32 bit).
NVCC doesn’t work. Bug was reported on this forum; also, we applied for “Registered Developer Status”, no reply.
This means we are stopped until someone at NVIDIA decides to help us.
Are you seriously suggesting that this is an issue for Microsoft?
With regard to Microsoft, in 10 years of windows development we have only had 3 instances of problems which could not be solved via MSDN; in these cases, a phone call to Microsoft resulted in a systems engineer walking us through debugging steps until the problem was identified; and in at least one of these cases, a hotfix was delivered in 1 day.
I have CUDA 3.0 installed with VS2k8 SP1 on Windows 7 (x64), no problems.
How does NVCC not work? What are you trying to do, what are you expecting to happen, what actually happens?
I see your anecdotal evidence and raise you this one: exact same system on my end and it works.
Actually: yes. If a user-space program does not work, and it worked on a previous version of the OS. Then the supplier of the OS has apparently made an OS that is not compatible with previous versions. NVCC is just a compiler, nothing fancy with drivers or other OS-interaction that can be the reason.
It can however also be user-error. Have you tried reinstalling the toolkit?
I can tell you that when I started trying CUDA, I ‘accused’ NVIDIA of having a crapty driver that rebooted my linux box. After much complaining from my side it finally turned out that my CPU was instable. And given the fact that you are the first to report that NVCC does not work on Win7, I think there is a big chance that there is something wrong with your install.
Regarding bugs:
I am a registered developer and have some bugs to report. I am unsure however if those weren’t reported earlier. There is supposed to be a list of all reported problems, but it seems to be empty – at least on my panel. Is there really no open/closed/fixed bugs in your database?
Also a bit of guidance on how to set some of the parameters could be helpful (e.g. Project). Some FAQ could be nice :)
The point is not to have a debate about who’s configuration is good and who’s isn’t.
The point is that in a case like this, there is no alternative but to try to solve the problem yourself, or wait until someone takes enough of an interest to help you. This is not suitable for a corporate environment.
Thank you for offering your assistance. We are running Window 7 32 bit.
We are attempting to build the “BlackScholes” sample from the SDK. When we try to do the build (Debug or Release, no difference), the NVCC generates an incomplete PTX file. Further steps fail because the PTX file is incomplete – it looks chopped off at the end. According to a trace using the “verbose” flag, this happens sometime during the NVOPENCC phase. When I trace it using System Internals, I don’t see any unusual error conditions; NVOPENCC is simply forking BE.exe, which is writing the PTX code to the file; just not all of it.
Reinstalled the toolkit twice, reinstalled the OS once as well (also tried it on Windows 7 64 bit).
NVCC was a port of a program originally designed for LINUX, shoe-horned onto Windows using Cygnus/Mingw, and again, my point is that it seems to be unsupported by NVIDIA.
Saying that NVCC is “just” a compiler is like saying that PostgresSQL is “just” a DBMS. Take a look at the posted source code – it’s a spaghetti monster.
Still, it is a user-space program. But I read that is actually works, just not good. Saying it is unsupported is a bit harsh, but it is certainly true that the presence of NVIDIA-people on these forums has diminished a lot. Tmurray was very, very active, but has another role nowadays, and I have no idea who in NVIDIA is responsible of monitoring there forums nowadays (I think tim is still doing it unofficially).
The bugs you see are limited to the ones you’ve posted. Pretty sure that a lot of people would complain if you get details of their projects…
DavidNY, PM me the email address you used for the registered developer signup and I’ll try to find out what has happened.
Also, I don’t think I’m less active now than I have been.
I have that impression, but me I am less active on the forum (only reading two subforums at this time), so it might be just my impression ;)
I think it’s more that there are some really good posters (you probably know who you are, good posters!) who jump in on things that I would answer before.
I see your point, thanks for claryfying it :)
I understand, there is no way to see if the bug I have at hand has already been reported or even fixed, so I’ll just put what I have and see what happens…
If you represent yourself, you can just say so in that field (independent consultant, etc).