All Discussions - NVIDIA Developer Forums

Tesla vs GTX560M this is wierd!

Bleiner — Wed, 16 May 2012 16:32:03 -0400

Hello Everyone,

So I have been working on porting over a lot of the more computationally intensive portions of my code from matlab to .ptx gpu code. I have been doing the development on my laptop where I have made HUGE increases in speed. Achieving at least a 10X speed up in a lot of areas. Now all my work is double precision, and involves millions of objects. My laptop with a gaming GPU (that by design has hampered double precision performance, and fewer processing cores) can complete a task in about 46 seconds.

I have a desktop with a Tesla C2075, that typically outperforms my laptop by a factor of two. When I bring this code over to the Tesla machine it is running anywhere from 42-48 seconds to complete the same task. Does anyone have any idea why this would be?

The only thing that comes to mind on this is that I upgraded my laptop to the 301.40 driver version to use Nsight visual studio, while the Tesla machine is still on 301.32 (or someething like that. When I attempted to upgrade the tesla machine it appears that the 301.40 drivers have been removed from the website. When I upgraded to the Beta Cuda 5 version drivers 301.53 (i think) Matlab no longer recognizes that there is a GPU attached to the system.

Could the issues be the driver? Did it improve that much from 301.27 to 301.4? Does it have anything to do with the GPU on the laptop being compute level 2.1 while tesla is 2.0? Is there a memory manager issue that 2.1 does a lot better?

The strange thing is that TESLA USED to perform at twice the speed of the laptop (and that includes all the overhead in matlab that is taking place on the CPU. So Tesla must have been significantly more than 2x faster.

Any thoughts?

Thanks
Ben

As an after thought here are the specs of the machines in question..
Tesla Work station
Dell Precision T5500
Xeon E5620 2.4ghz
12gb ram
Telsa C2075 with 6gb ram driver 301.32

Laptop
Asus G74S
Core I7 2670QM 2.2ghz
12gb Ram
GTX 560M 3gb ram Driver 301.40

Solve linear system in OptiX / Dynamic memory

xindon — Wed, 16 May 2012 15:28:28 -0400

Hey OptiX experts :-),

in order to implement a moving least squares point set surface algorithm I need to solve a linear system for each ray. The matrix sizes are variable and can change from frame to frame.

1) As OptiX does not allow dynamic memory allocation (free/malloc or new/delete), how/where should the data of the linear system be stored best?

Each ray (or more precise: each instance of the intersection program) needs its own local memory for the linear system. I thought of allocating a huge global buffer, but I'm not sure how to do the indexing such that the different programs don't interfere.

2) Do I have to implement the solver manually or are the better/faster/easier implementations already available? (which maybe also solve the above local memory problem).

Thank you very much in advance!

Regards
Tim

Non-blocking writing/reading not working

edodas — Tue, 01 May 2012 11:40:20 -0400

Hi,
Is there any reason that non-blocking data transfer wouldn't work? I am using a Quadro FX580 and when I set the blocking parameter to CL_FALSE in clEnqueueWriteImage or clEnqueueReadImage, I still have the impression that the data writing/reading is blocking.

DisplayLink BVU195 USB adapter

tnoonkes — Wed, 16 May 2012 12:54:21 -0400

I have a Lenovo T410 with the NVIDIA NVS 3100M. I recently bought a USB to DVI adapter so that i could use a 3rd monitor with my laptop. *(Laptop monitor is primary with 2 other samsung monitors attached, one via DVI input from laptop dock, second from the USB adapter.) I have updated my video drivers to the latest drivers but when i install the DisplayLink software for my USB to DVI, my computer crashes every time when i reboot windows. Please let me know if there is any fix for that adapter and NVS 3100M. Thanks.

downloading of the installer doesn't complete

mathiapeter — Wed, 16 May 2012 05:59:53 -0400

Hi, I had Nsight Visual Studio RC2 installed and everything worked fine, but now whet I try do download the final version 64bit, it shows that the installer has about 950MB but the downloading ends always at 51% so I can't run the installer. What would be the problem? I have tried 3 internet browsers but still the same.

APEX "Bake Clothing", frame 0 bug. DCC 2.85

Macha — Wed, 16 May 2012 12:45:59 -0400

In APEX Clothing midifier after using "Bake Clothing", when bake is finished, on frame 0 objects with APEX Clothing change their position to something diffrent than what was before the bake

http://imageshack.us/photo/my-images/546/beforebake.jpg/

http://imageshack.us/photo/my-images/339/afterbake.jpg/

Same thing is with other samples

Other frames does not seem to be affected

Win7 x64 SP1
3dsmax 2013 x64
NVIDIA_PhysX_For_Autodesk_3ds_Max_2013_64bit_2.85.00427.02000

Kernel launch fails if it contains rtTrace call

cneumann — Thu, 10 May 2012 12:21:35 -0400

It seems I can't get optix to actually trace any rays. As soon as I have a rtTrace call in my ray generation program the kernel launch fails (with exceptions enabled every launch index reports a RT_EXCEPTION_INTERNAL_ERROR). Not tracing rays and simply writing a color value to the output buffer works fine.
I can also successfully run the SDK examples (both precompiled and self built).

Since we've been struggling with this for a couple of days (at first suspecting a problem in my C++ wrapper code), we also wrote a logger library that intercepts all OptiX calls (it's LD_PRELOAD'ed) and shows arguments and return values, a log of a reproducer run is attached as well.
Comparing the log of our program with one from running an SDK example I can't find any substantial difference, but apparently there is something subtle going on ;)

Any suggestions on how to narrow down or solve the problem are very much appreciated. Thanks!

Platforms:
1) GTX 8800, Fedora 16 x86_64, driver 295.40, OptiX 2.5.0
2) GTX 560M, Win 7 64bit, driver 295.73, VS 2010, OptiX 2.5.0

OptiX 2.5.0 Ubuntu 10.10 64bit download is missing some examples

cneumann — Wed, 16 May 2012 11:23:15 -0400

The OptiX 2.5.0 download for Ubuntu 10.10 64bit (http://developer.nvidia.com/sites/default/files/akamai/tools/files/OptiX_2.5.0/NVIDIA-OptiX-SDK-2.5.0-ubuntu10.10-64.run) is missing the following directories/examples:

SDK/hybridShadows
SDK/isgShadows
SDK/isgReflections

as well as the precompiled versions of these.
If they were omitted intentionally SDK/CMakeLists.txt should be adjusted to not mention these directories in add_subdirectory() calls.

AWD for N-wheeled cars

romanmarchenko — Thu, 10 May 2012 09:52:07 -0400

Hi guys. I need to simulate M1126 Stryker ICV, which is 8-wheeled car. But eDIFF_TYPE_LS_4WD or eDIFF_TYPE_OPEN_4WD its not enough to make it truly all-terrain. If I will stop at hill, when front and rear wheels are hanging in air (all vehicle mass pushes on middle wheels), it is almost impossible to move my vehicle from this position - driven wheels are spinning freely. So my question is: Is it possible to apply torque to all wheels and make N-wheeled car truly NWD? Or maybe this feature will be added in future releases (I am using physx 3.2-beta3)? Or maybe it is possible to adapt tank simulation model somehow (but steering mode, according to documentation, is not really applicable to ordinary vehicles)?

Thanks in advance.

PhysX 3.2 Vehicle and Heightfield = crash

Vladimir62 — Sun, 13 May 2012 17:42:55 -0400

Hi!
I have found a bug with vehicle on heightfield.
When vehicle falls down on flat terrain (no matter falling height) - when wheels contact the heightfield - my application crashes with segmentation fault in PxVehicleSuspensionRaycasts() function call at physx::PxsBoardPhaseContextSap::batchUpdate() (from disassembler).
Vehicles work fine with flat plane as ground and another obstacles.
I have found what if the vehicle falls on the terrain not parallel (i.e. with small z axle angle - on the front or rear wheels only) - there are no crash happens and car lands on the terrain successfully and then moves ok. If vehicle lands on 4-wheels at same time, or rear and front wheels (x axle angle) - app crashes.

I'm using SampleVehicleWheelRaycastPreFilter from demo sample and heightfield generation from documentation sample.
Qt + GCC 4.4.3 + Ubuntu 10.04 lts + PhysX LINUX SDK 3.2 beta 2 and 3 (both crashes)

Autodesk Softimage 2012/13 and nVidia Apex

putranguan — Wed, 16 May 2012 10:41:26 -0400

Greetings

I'm goint to start using apex but I'm using Autodesk Softimage 2012 and I've read that this last don't need physx integration; it comes built in. But what about APEX? I'd like to make use of destruction, clothing and the rest of its power in Autodesk Softimage 2012.

Thanks

Cuda, PTX and Debugging symbols

Bleiner — Wed, 28 Mar 2012 13:08:21 -0400

Hello everyone,

I have a question about .ptx files and debugging in Visual studio 10 professional. I am attempting to write some ptx code to integrate into Matlab. However I am ONLY writing the __global__ functions and I have not written a main function or any host code. The functions are fairly simple, but what I want to do is be able to debug my code when matlab is running it. I am compiling my code using nvcc -ptx 'functionname.cu'. and when I try to use nvcc -G -ptx 'functionname.cu' to get debugging information nothing else is returned except the .ptx file. I should mention that I am relatively new to visual studio and that is why I am compiling using the command line.

I believe that if I had the symbols generated I could attach the process to Matlab and debug it. Since .ptx is supposed to be just in time compiled code does it not allow me to have debugging symbols?

Any help would be greatly appreciated
Ben

CUDA: clock cycles for division of one float value

hahnerehm — Wed, 16 May 2012 08:55:14 -0400

Hello,
you can find a table about how many arithmetic instructions per clock cycle a multiprozessor can perform in the CUDA C Programming Guide.
A division is performed by a couple of multiply-add instructions. But how many multiply-add instructions are needed to perform a division on 32 bit float value?

Thanks in advance,
Johannes

LiveKernelEvent

Michi10 — Wed, 16 May 2012 07:11:50 -0400

Hello support NVIDIA

Since 3 weeks, I have a big problem in the video or games comes with me on the screen such a picture.

[url=http://www.abload.de/image.php?img=foto0089s2dri.jpg][img]http://www.abload.de/thumb/foto0089s2dri.jpg[/img][/url]

After a few minutes, this image comes back.

[url=http://www.abload.de/image.php?img=foto00903gckl.jpg][img]http://www.abload.de/thumb/foto00903gckl.jpg[/img][/url]

Then I looked at the Fehlerrebort me and understand only station so I can make only so enter it you:

Produkt
Windows

Problem
Grafikkartenfehler

Datum
16.05.2012 11:06

Status
Nicht berichtet

Beschreibung
Aufgrund eines Videohardwareproblems ist Windows nicht mehr voll funktionsfähig.

Problemsignatur
Problemereignisame: LiveKernelEvent
Betriebsystemversion: 6.0.6002.2.2.0.768.3
Gebietsschema-ID: 1031

Dateien zur Beschreibung des Problems
WD-20120516-1106.dmp
sysdata.xml
Version.txt

Weitere Informationen über das Problem
BCCode: 117
BCP1: 87FA1510
BCP2: 92358ACE
BCP3: 00000000
BCP4: 00000000
OS Version: 6_0_6002
Service Pack: 2_0
Product: 768_1

I was with my computer at stores where I bought the parts and the graphics card and now even the technician is verwiert and white but not what he do sol my graphics card is only 1 month old. The burden has not revealed the whole PC and the driver version is 296.10 this software I de 5 times and installed but always the same result.

What can that this be it suddenly is pink and one freezes the whole computer?

From the UIG wonder what can be even?

The technician said to me the computer is fine, but what is the error?

I defend glad if you please me an only for the problem very soon can give a response?

With friendly regards
Michael Hirschegger

nvv start error code=13 (linux)

Da Ma — Tue, 15 May 2012 08:04:16 -0400

Dear all,

i want to profile my cuda program. In the good old time i used computeprof without any problem. Since update to cuda 4.1 i must use the new profiler nvv.

Some system information's:
Linux 3.2.1-gentoo-r2 x86_64
nvidia-drivers-295.41
dev-util/nvidia-cuda-sdk-4.1
dev-util/nvidia-cuda-toolkit-4.1
dev-java/sun-jdk-1.6.0.31
dev-java/sun-jre-bin-1.6.0.31

When i start nvv from comandline i get

JVM terminated. Exit code=13
/opt/cuda/libnvvp/jre/bin/java
-jar /opt/cuda/libnvvp/plugins/org.eclipse.equinox.launcher_1.1.0.v20100507.jar
-os linux
-ws gtk
-arch x86_64
-showsplash
-launcher /opt/cuda/libnvvp/nvvp
-name Nvvp
--launcher.library /opt/cuda/libnvvp/plugins/org.eclipse.equinox.launcher.gtk.linux.x86_64_1.1.1.R36x_v20100810/eclipse_1309.so
-startup /opt/cuda/libnvvp/plugins/org.eclipse.equinox.launcher_1.1.0.v20100507.jar
-exitdata a848011
-data @user.home/nvvp_workspace
-vm /opt/cuda/libnvvp/jre/bin/java
-vmargs
-jar /opt/cuda/libnvvp/plugins/org.eclipse.equinox.launcher_1.1.0.v20100507.jar

It told me nothing because i'm not that java-expert ... maybe someone can tell me whats went wrong. Should i use a newer version of sun java?

Thanks a lot in advance.

Best,
David

Cuda FreeBSD

DJs3000 — Wed, 16 May 2012 06:18:09 -0400

Can I use Cuda on FreeBSD? There is a GeForce 460.

Physx NxSceneStats2 * getStats2() no result if create fluid with flag .flags &= ~NX_FF_HARDWARE

manking — Wed, 16 May 2012 05:45:28 -0400

Physx 2.8.4
If create fluid with .flags &= ~NX_FF_HARDWARE there is collision, but in
getStats2() all values is 0.

If comment //.flags &= ~NX_FF_HARDWARE
getStats2() return correct result, but there is no collision.

Cg.framework 3.1 not working on XCode 3.2.6

jdxjohn — Mon, 14 May 2012 17:57:01 -0400

I'm totally pulling my hair out on this one... I've installed Cg framework using the installer on my Snow Leopard Intel MacBook and from the command line running "gcc -c test.cpp" on the following works fine:

#include 

void f() {
 CGContext context = cgCreateContext();
}

But in Xcode 3.2.6 I simply cannot get it to work, it cannot find . I've created a C++ console app and simply added #include to the hello world cpp file. I tried adding the framework to the target but it makes no difference.
If I start typing #include Xcode auto-suggests the file, or for any other installed framework, but not Cg.

I'm totally out of ideas why gcc with no command line parameters can find the framework but Xcode can't. I even tried moving the framework from /Library/Frameworks to System/Library/Frameworks but no change... it also doesn't show up in the default list of frameworks to add to the project.

I'm not an experienced Mac developer and I don't know if thre's any magic about frameworks, but to all intensive purposes XCode simply doesn't seem to think /Library/Frameworks/Cg.framework is a framework.

If anyone can shed some light on this I'd be hugely grateful, I've been banging my head on this wall for days now and can find nothing on the web about it and am basically sitting here with code I can't build for no obvious reason!!

I am using CG 3.1, v3.1.0010 installed on MacOSX 10.6 with XCode 3.2.6.

[PhysX 3.1.1] Creation of deformable triangle meshes

reiserzx — Tue, 03 Jan 2012 08:54:25 -0500

Hello,

at first i wish some good 2012.

Beside that important note i also have issues when trying to create deformable meshes with PhysX 3.1.1. I got my vertice and index buffer which i cook a deformable mesh with. The cooking itself runs fine. Yet when i use it to create the PxDeformable i receive an exception from within the call of: CPhysX::g_pPhysics->createDeformable(desc);

The exception is: First-chance exception at 0x503c78f9 in .exe: 0xC0000005: Access violation writing location 0x00000000.

The question i'd like to ask is the following. When the cooking runs fine, is it guaranteed that the resulting PxDeformableMesh is a valid input for createDeformable ? Im just wondering what might be wrong here.

Thanks for the support,
Oliver

Here some adjusted code lines:


PxDeformable* CreateDeformable(PxVec3* vertices, PxU32* indexBuffer, CONST UINT& verticeCount, CONST UINT& indexCount){

	 //prepare mesh description
	PxDeformableMeshDesc desc;
	desc.vertices = pxVertices;
	desc.primitives = pxIndices;
	desc.numVertices = verticeCount;
	desc.numPrimitives = pxPrimitiveCount;
	desc.vertexStrideBytes = sizeof(PxVec3);
	desc.primitiveStrideBytes = sizeof(PxU32) * 3;
	desc.primitiveType = PxDeformablePrimitiveType::eTRIANGLE;

	PxDeformableMesh* dMesh= NULL;
	if(desc.isValid()){
		PxToolkit::MemoryWriteBuffer streamOut;
		if(CPhysX::g_pCook->cookDeformableMesh(desc, streamOut))
		     dMesh = CPhysX::g_pPhysics->createDeformableMesh(PxToolkit::MemoryReadBuffer(streamOut.data));
	}


	 if(result){
		PxDeformableDesc desc = PxDeformableDesc(CPhysX::g_pPhysics->getTolerancesScale());
		desc.deformableMesh = dMesh;
		desc.mass = 1.0F;
		desc.deformableReadDataFlags = PxDeformableReadDataFlag::ePOSITION_BUFFER;

		if(desc.isValid()){
	 	 	//this is the call where the exception occur
			return CPhysX::g_pPhysics->createDeformable(desc);
		}
	}
	return NULL;
}

Variables have "no value" in NSight

robosmith — Tue, 15 May 2012 20:32:56 -0400

Several variables which are in scope at the breakpoint are shown as "no value at the target location." Sometimes they have a value at the assignment point, but change to "no value" after executing another line. How can I see these values?

This is using Dev Kit v4.2 & NSight v2.2 release with VS2008 on GTX 525 with Optimus.

Breakpoint don't be hit using Nsight Visual Studio

froure — Tue, 15 May 2012 08:49:16 -0400

Hi!

I'm trying to debug with Nsight. I put some breakpoints but debugger don't stop the execution there. During the debugging, this breakpoints appers with a warning icon, and when I put the mouse over, it says: "The breakpoint will not currently bye hit. CUDA: no source correspondence for breakpoint".

I try to find some solution inside the other forums, but I can't.

Can you help me please?

Thanks!! :D

Can't use GPU local debugging with Nsight

laz007 — Sun, 15 Apr 2012 03:58:50 -0400

Hello!
I'm using GF 8400M GS on a x64 Windows 7 machine with Visual Studio 2008.
I installed the new beta driver (301) and Nsight 2.2. After I tried the tutorial nothing happened when I put breakpoints in CUDA kernels.
What could be the possible reason for that problem?
Does my GPU support local debugging? Because it was written that only selected GPUs support it.

Linker Issue in building a CUDA application in Visual Studio 2010

gladman — Thu, 16 Feb 2012 03:56:49 -0500

Reposted from old forum.

I am trying to build an application with Visual Studio 2010 and Nsight 2.1 hosted on Windows 7 (x64) using the 4.0 and/or the 4.1 build customisation rules.

I am using object file output format with the default filename for the output object files - $(IntDir)%(FileName)%(Extension).obj, which correctly generates all the object files in the $(IntDir) directory. But after all the files have compiled correctly, the linker build step then fails with the message:

LINK : fatal error LNK1181: cannot open input file '..\.obj'

It appears that the file listing inputs for the linker is not being generated.

If I perform the linker step manually, the application builds correctly and runs without problems. So it seems that there is an issue with the way the build customisations are producing the input file lists for the linker.

I would appreciate any advice that people can offer on what might be going wrong.

Best way to tackle a specific MIP with CUDA

Caelen — Thu, 10 May 2012 04:57:51 -0400

Hi every one,

I am actually working on a project and I would like to have your opinion
The aim of this project is to solve a problem using GPGPU programming (so CUDA)

So problem is quite easy, I have to put in a truck a lot of container, with different weight and size. But the Truck is large enough to accept all containers so it is not a bin packing problem. Also, the truck is divided in N area in which you can put exactly one container.
So, each container has an order of “preference place” and I have to optimize this preference. Also you have some other constraint; some container must be neighbor, one container can take one and half place etc…

The problem can be model by a Mixed Integer problem, with an objective function.
Actually, I was thinking to put for each container N binary variables and put 1 at the variable i if the container is in the area I, 0 otherwise.

The problem is that, I will have a spare matrix (density under 5%). So, from what I've read so far, it won’t be very useful to use GPGPU

I also think to put for each container a variable which can have the value 1 to N. Then the matrix will be denser.

My major problem is that I don’t know what is the best way to tackle the problem :
Using branch & Band Solver and during the relaxation, use a LP solver programming in GPGPU.
Or use Constraint Solver in GPGPU
Or, use metaheuristic solver in GPGPU.

So, I would like to have your opinion about my problem

Thank’s a lot

Flashing L4T on cardhu dev tablet

madmaze — Wed, 28 Mar 2012 12:57:40 -0400

I have the latest version of the Linux for Tegra development kit,
I am stuck on flashing the dev tablet with the new kernel/bootloader.

The documentation says the following:

You must first put the target board into reset/recovery mode. Do so by first powering 
on the board and then holding the recovery button and pressing the reset button.

When i power on the tablet it announces:


...
Checking for RCK.. press key  in 5 sec to enter RCK
OS will cold boot in 10 seconds if no input is detected
Press  to select,  for selection move

(options are USB and Android)

if i press Vol- when the RCK message pops up it gives me a distorted screen with an android exclamation mark in it. Then no other input it possible, not even turning it off.

if i select to boot USB then it claims "Starting Fastboot USB download protocol"
If at this point I execute flash.sh it will hang after "Nvflash started"

any suggestions on what I could/should try?
Is there a way to get debug output from NVFLASH about where its stuck?

Thanks,

Matthias

Booting device from internal eMMC drive aka mmcblk0p5

PouryaShirazian1 — Tue, 15 May 2012 18:50:07 -0400

Sorry if my question is already posted somewhere in ther forum. I have successfully booted the device from a prepared USB drive and now want to flash the device to be able to boot from its internal eMMC memory drive. How can I do this properly?

The error that I receive is:

file not found: bootloader/system.img
failed executing command 2147483647 NvError 0x4
command failure: create failed
Failed to flash cardhu.

Thanks

Trouble installing 2.7 Physx plug-in for Max2012

CDIcthomas — Wed, 14 Mar 2012 14:11:52 -0400

I have downloaded the 2.7 build for max2012-64 but when trying to install the application it fails to find an installation of max2012-64 installed. I do have a licenced version working on this system, but it is on the D: drive. I had a similar problem with the Fume_FX 2012 install but it allowed for me to point to the installation folder to complete the install.

Any work arounds that you know of? I didn't have this problem with 2011 and it is on the same drive as 2012.

Thanks for any help to get this working in max 2012-64

Cuda 5 Preview: GPUDirect RDMA ?

skybuck2000 — Tue, 15 May 2012 18:01:15 -0400

Hello,

On the cuda toolkit 5.0 preview download page it mentions the following but I cannot find any
further documention about what this is ?

"
GPUDirect RDMA provides fastest possible communication between GPUs and other PCI-E devices
• Direct memory access (DMA) supported between NIC and GPU without the need for CPU-side data buffering
• Significantly improved MPISendRecv efficiency between GPU and other nodes in a network
• Eliminates CPU bandwidth and latency bottlenecks
• Works with variety of 3rd party network and storage devices
"

Are these just general hardware/driver improvements or can I somehow program cuda to communicate with network cards (nics) ?

Sounds somewhat interesting so I would like to know more about it...

Are these perhaps tesla only features or so ? Does fermi support it or only the new kepler ?

I find this link/doc:

http://www.cc.gatech.edu/~vetter/keeneland/tutorial-2011-04-14/13-cuda_advmpi_keeneland.pdf

It's somewhat older (?) and it mentions linux only, has this now changed, and is it available for windows too ?

Ok I found some official documentation/link here:

http://developer.nvidia.com/gpudirect

Hmm, so far it seems to be for tesla/data center products and linux...

I guess this is not for desktop/windows (yet?) and probably requires special network card and/or drivers ? hmm...

Bye,
Skybuck.

Can I redistribute nvmex and nvopts.sh?

jeffblanchard — Tue, 15 May 2012 17:04:08 -0400

I want to release a CUDA-C (nvmex) package I have nearly finished. I am working on the licensing and such. I can not seem to find a definitive answer regarding including nvmex and nvopts.sh in my package. Am I permitted to include these files in my package?

Kepler Quadro?

Bleiner — Tue, 15 May 2012 15:43:41 -0400

Hello Everyone,

Not to be a starting rumors, but has anyone heard anything about when the new Kepler Quadro cards will be coming out? I am responsible for selecting new laptops for the design department at my company. The 5010M looks to be a great card don't get me wrong, but with my code starting to rely more on CUDA, I want to get the most powerful GPU for double precision work I can find. I would hate to drop a LARGE sum of money on a laptop only to have the GPU be obsolete a month later, especially when the GPU is a major component of the cost.

Any rumors? Any employees around who can deny knowing anything while winking at the same time?

Thanks
Ben

Matlab, Mex and Cuda

Bleiner — Tue, 15 May 2012 15:33:54 -0400

Hello Everyone,

I had a question about integrating Cuda code with C code. Here is my dilema: I am working in a matlab environment, and I want to introduce some Cuda C code to help speed up some applications. I can do that using .PTX files without much trouble, but that is strictly for GPU only code called my matlab. But what if I wanted to use Host C code to do some operations faster than matlab and move stuff to the GPU and process data during the C code.

I can compile C code in matlab using the MEX function, but as far as I know Mex won't call NVCC as a compiler (only C compilers). So my question is. Is there a way to call precompiled (in so much as ptx is compiled) from my C function such that my C code will compile using a standard C compiler? or is there another format I can compile my CUDA code to such that I can just include it during compilation (without the need for the CUDA to be recompiled)? I am new to C and CUDA programming and so I am unfamiliar with a lot of the compiler issues and tricks.

Thanks
Ben

How do I use "prof_trigger" (user profile triggers) from my kernel?

m4dc4p — Tue, 15 May 2012 13:07:03 -0400

The CUPTI Event API provides counters for user profiling (prof_trigger_00 through prof_trigger_07).

I can figure out how to read those counters, but how do I write to them or in some other use those counters from my kernel?

I am using a Tesla (v1.1 capability) GPU.

Thanks!

Cg Runtime : CG_FILE_READ_ERROR

eliyam30 — Sun, 13 May 2012 08:34:27 -0400

Hi all,
Im having a trouble with CG context error:
when trying to create an effect I get :
err = "Cg runtime error: The file could not be read."
What Is necessarily needed for the success of creating an effect?

Im trying to create the effect from the lump code i pasted down,
but for some reason it fails.
(The problem occurs only when trying to compile it from a specific call stack order)

std::string test=
"// Vertex attribute interface:\n"
"struct APP2VP\n"
"{\n"
" float4 position : POSITION;\n"
" float4 Normal : NORMAL;\n"
" float4 UV : TEXCOORD0;\n"
"};\n"
"// Varyings interface:\n"
"struct VP2FP \n"
"{\n"
" float4 position : POSITION;\n"
" float2 UV : TEXCOORD4;\n"
"};\n"

"// VERTEX PROGRAM\n"
"VP2FP vp(APP2VP IN)\n"
"{\n"
" VP2FP OUT;\n"
" OUT.position = IN.position;\n"
" OUT.UV = IN.UV.xy;\n"
" return OUT;\n"
"}\n"
"// FRAGMENT PROGRAM\n"

"float4 fp(VP2FP IN) : COLOR\n"
"{\n"
" \n"
" return float4(1,0,0,0);\n"
"}\n"
"technique render\n"
"{\n"
" pass both\n"
" {\n"
" VertexProgram = compile latest vp();\n"
" FragmentProgram = compile latest fp();\n"
" }\n"
"}\n";

Thanks!

Update:
If I paste the create from lump call :


cgfxHdl = CgFx::createFromLump( test ,m_searchPaths,err,failOnTexLoad );


cgfxHdl = CgFx::createFromLump( test ,m_searchPaths,err2,failOnTexLoad );

Than the second call succeeds!

How to get the result in a data form?

milkiii — Mon, 14 May 2012 23:00:44 -0400

The demos feedback the result of ray tracing through a picture. If I now have a mesh which is consisted by several vertex, and I want to use a form to represent each vertex's light intensity, does any function or method to get this kind of result? Or, since PerRayData.result stores the result after ray tracing, can I get it in some way? It is so important to me, hoping I can get the answer. Thank u very much!

2 questions reguarding PhysX

dartwing17 — Tue, 15 May 2012 10:03:00 -0400

1. Can cooked triangleMeshes be particle drains?

When setting shape->setFlag(physx::PxShapeFlag::ePARTICLE_DRAIN, true) on the TriangleMesh shape, nothing happons

2. How do I get the fluid particle densityBuffer?

I have set the particleReadDataFlags when creating the fluid (particleSystemDesc.particleReadDataFlags |= physx::PxParticleReadDataFlag::eVELOCITY_BUFFER | physx::PxParticleReadDataFlag::eDENSITY_BUFFER;)
and get the density per particle like this:

physx::PxParticleFluidReadData* rd = pf->lockParticleFluidReadData();
if (rd->validParticleRange > 0)
{
physx::PxStrideIterator particleFlags(rd->flagsBuffer);

// iterate over valid particle bitmap
for (physx::PxU32 w = 0; w <= (rd->validParticleRange-1) >> 5; w++)
{
for (physx::PxU32 b = rd->validParticleBitmap[w]; b; b &= b-1)
{
physx::PxU32 index = (w << 5 | physx::lowestSetBit(b));
const physx::PxF32& density = rd->densityBuffer[index];

if (particleFlags[index] & physx::PxParticleFlag::eVALID)
{
particleDensityArray.push_back(density);
}
}
}
}
rd->unlock();

but doing this causes the runtime error:

c:\p4\px31_installers\sw\physx\release\physx-3.1\standard_build\physxsdk\sdks\physx\src\particles\NpParticleFluidReadData.h (64) : invalid operation : PxParticleReadData access through PxParticleBase::lockParticleReadData() while its still locked by last call of PxParticleBase::lockParticleReadData().

I dont understand why I get this error. I do the same procedure for getting position and velocity, and that works flawless.

- Andy

PS: I use Physx 3.1.1

Update: Forget question nr. 2. I think I understand now why it doesnt work. The densitybuffer is only valid after the particle fluid has been simulated. Meaning if I for example put a boolean in my last if-check which is false at startup, and set it to true while the simulation is running then it works.

But im still wondering if its possible to make a trianglemesh a particle drain... still cant make that work

Help adding external force to particles

dartwing17 — Sat, 12 May 2012 10:29:45 -0400

Hi,

I have run a wind simulation in Flow-3D and have exported a set of wind-vectors, which describe
a position and a velocity. I want to use these wind-vectors in my particle system, so that when a particle is inside a wind-vector position (lets say +- 0.5), the velocity of the wind-vector should be added to the particle.

I have implemented a way to do this, but it is very inefficient. I was just wondering if anyone had any suggestion on how to make the algorithm more efficient.

This is my implementation (sorry for the ugly code):

glm::vec3 checkWindPosition(glm::vec3 particlepos)
{
for(int i = 0; i < flowWindGrid.size(); i++)
if(particlepos.x < flowWindGrid[i].position.x+0.5f && particlepos.x > flowWindGrid[i].position.x-0.5f
&& particlepos.y < flowWindGrid[i].position.y+0.5f && particlepos.y > flowWindGrid[i].position.y-0.5f
&& particlepos.z < flowWindGrid[i].position.z+0.5f && particlepos.z > flowWindGrid[i].position.z-0.5f)
return flowWindGrid[i].velocity;
return glm::vec3(99,99,99);
}

void update()
{
physx::PxParticleFluidReadData* rd = pf->lockParticleFluidReadData();
if (rd->validParticleRange > 0)
{
physx::PxStrideIterator particleFlags(rd->flagsBuffer);
// iterate over valid particle bitmap
for (physx::PxU32 w = 0; w <= (rd->validParticleRange-1) >> 5; w++)
{
for (physx::PxU32 b = rd->validParticleBitmap[w]; b; b &= b-1)
{
physx::PxU32 index = (w << 5 | physx::lowestSetBit(b));
const physx::PxVec3& position = rd->positionBuffer[index];

if (particleFlags[index] & physx::PxParticleFlag::eVALID)
{
glm::vec3 newvelocity = checkWindPosition(glm::vec3(position.x,position.y,position.z));
if(newvelocity != glm::vec3(99,99,99))
{
windIndexBuffer.push_back(index);
windForce.push_back(physx::PxVec3(newvelocity.x,newvelocity.y,newvelocity.z));
}
}
}
}
}

rd->unlock();

physx::PxStrideIterator forceBuffer(&windForce[0]);
physx::PxStrideIterator indexData(&windIndexBuffer[0]);
pf->addForces(windForce.size(), indexData, forceBuffer, physx::PxForceMode::eFORCE);
}

Errors with cuda RC2 visual profiler

tcelvis — Fri, 16 Dec 2011 19:31:09 -0500

1. Execution starts normally (as indicated by output lines. After approx 20 seconds progress status switches to collecting results and then aborts with the following error dialog:
"Unable to locate CUDA libraries and establish connection with CUDA dirver
Error com.nvidia.viper.jni.CuException: CUDA_ERROR_INVALID_VALUE"

Some reruns give error ... CUDA_OUT_OF_MEMORY

2. Notices that when nvvp is launched 42 processes show up all looking identical. "top" output for each line is as follows:
/usr/local/cuda/libnvvp/jre/bin/java -jar /usr/local/cuda/libnvvp/plugins/org.eclipse.equinox.launcher_1.1.0.v20100507.jar -os linux -ws gtk -arch x86_64 -showsplash -launcher /usr/local/cuda/libnvvp/nvvp -name Nvvp --launcher.library /usr/local/cuda/libnvvp/plugins/org.eclipse.equinox.launcher.gtk.linux.x86_64_1.1.1.R36x_v20100810/eclipse_1309.so -startup /usr/local/cuda/libnvvp/plugins/org.eclipse.equinox.launcher_1.1.0.v20100507.jar -exitdata 6e000f -data @user.home/nvvp_workspace -vm /usr/local/cuda/libnvv

3. Visual profiler users guide included with RC2 still references computeprof and not nvvp. computeprof was not part of this distribution.

4. This application runs file as the execuatable when not using visual profiler. The version 4.0 computeprof also worked fine on this application.

5. Using Centos 5.5 linux on a quad-hex core chassis containing 8 Fermi 2090 GPUs.
Utilizing stream ids running 24 threads with 3 threads sharing each GPU

cudaFree returning cudaErrorMemoryAllocation - bug?

ajsimmonds — Mon, 14 May 2012 13:38:35 -0400

I have been encountering a strange problem using the cuda 4.2 tools where our application eventually receives a cudaErrorMemoryAllocation error when trying to perform a cudaFree on cudaMalloc'd memory. The number of allocations and deallocations performed varies but the problem can be reproduced in the app relatively easily. Once the error has been received once further frees and also cudaMemGetInfo continue to return the error.

To further narrow down the error I have also written a test program that simply allocates areas using cudaMalloc and when this returns an out of memory error, releases one or more of the previously allocated areas to make space. This program, which launches no kernels, fails with the same symptoms. I have also tried this with the 4.0 tools and still receive the same error condition.

If I limit the number of iterations such that the error is not encountered then it is quite likely that the free memory value returned by cudaGetMemInfo is larger than the value the program started with.

This looks all the world to me as if there is a problem with the tracking of memory allocations within the SDK, so can anyone confirm this or possibly point me at things I may be doing wrong?!

Many thanks
Andrew

Move only child in the Joint drive.

Booster — Fri, 11 May 2012 14:36:31 -0400

Hello everyone.
Is it possible to move only child actor of joint through joint drive? Thanks in advance.

ArrayFire + Nsight???

sizheng — Wed, 09 May 2012 21:08:48 -0400

i'm trying ArrayFire, but it seems that i cannot debug arrayfire code by nsight~~~

anybody knows how to?

Parallel Nsight 2.2 RC2 - No Source Available

wdrozd — Fri, 04 May 2012 16:35:28 -0400

When selecting CUDA Debugging with memory checker enabled I get a kernel crash with a window in Nsight that says "No Source Available". When I click on the link "Browse to Find Source", a message says "The source code cannot be displayed".

My application compiles fine, so clearly it can find the source (for both the C code and the Cuda code)

Also I have no problem stopping at a breakpoint set in my Kernel prior to the crash (grid launch failure)

The call-stack says "No active Cuda Kernels".

Can you please me determine how to set Nsight to detect the source?

Thanks.

Please Help - error C0501: type name expected at token "sampler2D"

n0valyfe224 — Mon, 14 May 2012 05:07:48 -0400

I'm new to this forum and a complete noob at writing shaders, I'm following a tutorial series by Eat 3D.. I'm writing this shader for Maya and am currently trying to get my first textures in..

whenever I try and load the shader into Maya I get the following errors:

cgfxShader -e -fx "D:/Tutorials/Shader Production/05_Textures/05_Textures.cgfx" Fresnel_Shader;
// Error: The compile returned an error. //
// Error: D:/Tutorials/Shader Production/05_Textures/05_Textures.cgfx(76) : error C0501: type name expected at token "sampler2D"
D:/Tutorials/Shader Production/05_Textures/05_Textures.cgfx(110) : warning C7011: implicit cast from "float4" to "float3"
D:/Tutorials/Shader Production/05_Textures/05_Textures.cgfx(140) : warning C7011: implicit cast from "float4" to "float3"
D:/Tutorials/Shader Production/05_Textures/05_Textures.cgfx(76) : error C0000: syntax error, unexpected type identifier, expecting ',' or ';' at token "sampler2D" //
// cgfxShaderNode::cgErrorCallBack
// Error: The compile returned an error. //
// Error: D:/Tutorials/Shader Production/05_Textures/05_Textures.cgfx(76) : error C0501: type name expected at token "sampler2D"
D:/Tutorials/Shader Production/05_Textures/05_Textures.cgfx(110) : warning C7011: implicit cast from "float4" to "float3"
D:/Tutorials/Shader Production/05_Textures/05_Textures.cgfx(140) : warning C7011: implicit cast from "float4" to "float3"
D:/Tutorials/Shader Production/05_Textures/05_Textures.cgfx(76) : error C0000: syntax error, unexpected type identifier, expecting ',' or ';' at token "sampler2D" //
// Error: cgfxShader : "Fresnel_Shader" unable to load effect "D:/Tutorials/Shader Production/05_Textures/05_Textures.cgfx" //

Here is the shader itself:

//////////////////////////////////////////////////////////////
// CGFX Matrices Section
float4x4 WorldViewProjection : WorldViewProjection < string UIWidget = "None"; >;
float4x4 WorldInverseTranspose : WorldInverseTranspose < string UIWidget = "None"; >;
float4x4 ViewInverse : ViewInverse < string UIWidget = "None"; >;
float4x4 World : World < string UIWidget = "None"; >;

// Maya Description
string description = "Basic Texture Shader";

//////////////////////////////////////////////////////////////
// Parameters section

float3 shaderColor : Diffuse
<
string UIName = "Shader Color";
> = {1.0f, 1.0f, 1.0f};

float3 lightDirection : Direction
<
string UIName = "Light Direction";
> = {0.0f, 1.0f, 0.0f};

float3 ambientColor : Diffuse
<
string UIName = "Ambient Color";
> = {0.15f, 0.15f, 0.15f};

float fresnelPower
<
string UIWidget = "slider";
string UIName = "Fresnel Power";
float UIMin = 0.0f;
float UIMax = 10.0f;
float UIStep = 0.01f;

> = 3.0f;

float specularPower
<
string UIWidget = "slider";
string UIName = "Specular Power";
float UIMin = 0.0f;
float UIMax = 100.0f;
float UIStep = 0.01f;

> = 30.0f;

float3 FresnelColor : Diffuse
<
string UIName = "Specular Color";
> = {1.0f, 1.0f, 1.0f};
// Textures

// Diffuse
texture diffuseTexture
<
string name = "default_color.dds";
string UIName = "Diffuse Texture";
string TextureType = "2D";
>

sampler2D diffuseMap = sampler_state
{
Texture = ;
AddressU = Wrap;
AddressV = Wrap;
};

//////////////////////////////////////////////////////////////
// Structs section
// input from application
struct a2v {
float4 position : POSITION;
float2 texCoord : TEXCOORD0;
float4 normal : NORMAL;
};

// output to fragment program
struct v2f {
float4 position : POSITION;
float2 texCoord : TEXCOORD0;
float3 worldNormal : TEXCOORD1;
float3 eyeVector : TEXCOORD2;
};

//////////////////////////////////////////////////////////////
// Vertex Shader
v2f vShader(a2v In)
{
v2f Out;
Out.position = mul(WorldViewProjection, In.position);
Out.worldNormal = mul(WorldInverseTranspose, In.normal).xyz;

float3 worldSpacePos = mul(World, In.position);
float3 worldCameraPos = float3(ViewInverse[0].w, ViewInverse[1].w, ViewInverse[2].w);
Out.eyeVector = worldCameraPos - worldSpacePos;

Out.texCoord = In.texCoord;

return Out;
}

//////////////////////////////////////////////////////////////
// Pixel Shader
float4 pShader(v2f In) : COLOR
{
float4 outColor;

float3 worldNormal = normalize(In.worldNormal);
float3 lightDir = normalize(-lightDirection);
float3 eyeVector = normalize(In.eyeVector);
float3 reflectionVector = - reflect(eyeVector, worldNormal);

float4 colorMap = tex2D(diffuseMap, In.texCoord);

float3 specular = pow(saturate(dot(reflectionVector, lightDir)), specularPower)* FresnelColor;

float3 lambert = saturate(dot(lightDir, worldNormal));

float3 fresnel = pow(1- saturate(dot(eyeVector, worldNormal)), fresnelPower) * lambert * FresnelColor;

outColor.rgb = (lambert + ambientColor) * shaderColor + fresnel + specular ;

outColor.rgb = colorMap;

outColor.a = 1.0f;

return outColor;
}

//////////////////////////////////////////////////////////////
// Techniques
technique Simple
{
pass one
{
VertexShader = compile vp30 vShader();
PixelShader = compile fp30 pShader();
}
}

Any Help would be greatly appreciated, really want to know what I am doing wrong so I can learn from it.. been staring at it for hours and redoing the tutorial over but to no avail..

Nsight on Dell XPS 15 Optimus Notebook

den3b — Fri, 11 May 2012 00:50:34 -0400

Hey there!

I wanna buy a Dell XPS 15 which seems to be officially supported for NSight.

Does anybody have a similar notebook and can confirm NSight works smoothly on it for debugging?

Thanks a lot!

Rick

Performance advices?

jamesclarke — Fri, 11 May 2012 11:28:55 -0400

Hi, it's me again. Sorry for so many questions these days.

I am working on a project of terrain rendering. I need some performance advices, please help me. Thank you!

I am using Quadtree for LOD with two graph trees, one for rendering and the other for preparing. However, I found switching between these two trees comsumes a lot of time. I have tried "Switch" class, and it seems not a good choice. Would hints (NVSG_HINT_ALWAYS_INVISIBLE) or the TraversalMask mechanism be useful?

ps:
the whole process is : when viewpoint moves, find the terrain nodes which are not changed, from the rendering tree, and reuse them in the preparing tree; read new terrain data and build new nodes into the preparing tree; switch the rendering tree and the preparing tree.

New libs

giveen — Mon, 14 May 2012 17:01:38 -0400

Are the libs inside /system/lib of the Ventana ICS image newly made or are they a hangover from Honeycomb.

Installing Apex / physx on a Maya 2012 deployment

Vustadumas — Mon, 14 May 2012 15:24:49 -0400

When trying to install the Physx plug-ins, in Maya 2012 (64-bit), it says that Maya is not installed. Our deployment isn't installed in the Maya default folder structure. Could this be the cause of the error?

Here's the error...

PhysX For Autodesk Maya 2012 64bit is not installed.
Setup requires the version of Autodesk Maya of
matching version and bits to be installed prior to
installation of this product. Setup will now exit.

Thanks for any help!

Corey

What's the problem with materials and which side they are connected to deserialization problems?

Kosmos — Sun, 13 May 2012 11:41:08 -0400

Hello everyone
I am trying to serialize some triangle meshes in scene from one application and deserialize them in another.
So, i have next code for serialization purposes:
PxSDK is PxPhysics object pointer
mPxScene is PxScene object pointer


class FileSerialStream : public PxSerialStream //class for writing serialized file
{
public:
    FileSerialStream(std::string file_name) { ofile.open(file_name, std::ios::binary); }

    ~FileSerialStream()
	{
		ofile.flush();
		ofile.close();
	}

    virtual void storeBuffer(const void *buffer, physx::PxU32 size)	{ ofile.write(reinterpret_cast(buffer), size); }
	virtual physx::PxU32 getTotalStoredSize(void)					{ return ofile.tellp(); }

private:
	std::ofstream ofile;

    // Disable unimplemented operations
    FileSerialStream( const FileSerialStream & );
    FileSerialStream& operator=( const FileSerialStream & );
};

void SerializeScene()
{
	PxCollection *mCollection = PxSDK->createCollection();

	for(unsigned int i = 0; i < movable_entries.size(); i++) //run though physical objects in scene
	{
		if (movable_entries[i].mController != NULL || movable_entries[i].IsCameraController) //if object is character controller
			continue; //we skip it

		movable_entries[i].mActor->collectForExport(*mCollection); //add object to collection
	}

	FileSerialStream fss("serialized_map.bin");
	mCollection->serialize(fss); //serialize
	PxSDK->releaseCollection(mCollection);
}

First of all, i see major difference in file size between this serialization and creating RepX file from PVD. If i save objects in frame to RepX file size is 433kb, but with binary serialization it 14kb (i don't have any idea how it can store all that figures points in such amount of data)
Anyway... let's assume that serialization successed.

Deserialization apllication code:

void DeserializeScene(const char* filename)
{
	FILE* fp = NULL;

	if( (fp = fopen(filename, "rb")) )
	{
		fseek(fp, 0, SEEK_END);
		PxU32 fileSize = ftell(fp);
		fseek(fp, 0, SEEK_SET);

		PX_ASSERT(fileSize != 0);
		void* mem = malloc(fileSize + PX_SERIAL_FILE_ALIGN);

		void* mem16 = (void*)((size_t(mem) + PX_SERIAL_FILE_ALIGN)&~(PX_SERIAL_FILE_ALIGN - 1));
		fread(mem16, 1, fileSize, fp);
		fclose(fp);

		PxCollection* collection = PxSDK->createCollection();
		collection->deserialize(mem16, NULL, NULL);

		PxSDK->addCollection(*collection, mPxScene);
		PxSDK->releaseCollection(collection);
	}
}

When this function runs i getting next messages in console:
NpRigidStatic | 42 | 4536
NpShape | 42 | 8064
Padding: 168 bytes
C:\p4\px31_installers\sw\physx\Releases\PhysX-3.1\standard_build\PhysXSDK\SDKs\PhysX\src\NpPhysics.cpp (733) : invalid operation : resolveMaterialPtr: material not found.
PVD shows empty scene, so... objects isn't load succesfully.
From messages we see something about invalid operation : resolveMaterialPtr: material. Wat is it? What should be done about it? There is nothing about it in built-in guides.
I am using PhysX-3.1.2_PC_VC10_SDK_Core