Archives Discussions

[ryta1203]

I'm trying to run a 4096x4096 sample of the Black Scholes problem but I'm getting an CL_OUT_OF_RESOURCES error on the clEnqueueReadBuffer for both the putPrice and callPrice, any ideas?

I've tried to change the work size but no luck.

[ryta1203]

BTW, I'm running a 5870 with 1GB of ram. I only mention this because the Black Scholes runs fine at 2k*2k problem size and I only get this error when moving up to 3k*3k or 4k*4k.

[genaganna]

Originally posted by: ryta1203 BTW, I'm running a 5870 with 1GB of ram. I only mention this because the Black Scholes runs fine at 2k*2k problem size and I only get this error when moving up to 3k*3k or 4k*4k.

Ryta1203,

       I am able to run without any problem. It looks like you are facing problem with CPU resources. Please try to close other applications running on your system.

[ryta1203]

Wouldn't I see a "failed to allocate host memory"? I'm not seeing this.

Also, I've closed down every other non-essential program and I still get the same error:

KernelFailed.CL_OUT_OF_RESOURCES.

I heard that error is usually due to lack of available registers or local memory, which is why I tried reducing the work size but that didn't help.

Are you running Win Xp 32 bit?

My system:

WinXP 32 bit

AMD Phenom x4 II 3.2Ghz

ATI 5870

4GB RAM

Any other ideas? This is very annoying. I can't run 4k*4k for Mersenne Twister either.

But I can run 4k*4k for DCT, for instance.

[genaganna]

Originally posted by: ryta1203 Wouldn't I see a "failed to allocate host memory"? I'm not seeing this.Also, I've closed down every other non-essential program and I still get the same error:

KernelFailed.CL_OUT_OF_RESOURCES.

I heard that error is usually due to lack of available registers or local memory, which is why I tried reducing the work size but that didn't help.

Are you running Win Xp 32 bit?

My system:WinXP 32 bit

AMD Phenom x4 II 3.2Ghz

ATI 5870

4GB RAM

Any other ideas? This is very annoying. I can't run 4k*4k for Mersenne Twister either.

But I can run 4k*4k for DCT, for instance.

I am running on XP64.  Is this error coming from clEnqueueNDRangeKernel or clEnqueueReadBuffer?  Sample checks of resources before calling clEnquueNDRangeKernel.

One more thing sample not using local memroy at all.  why register size increase based on input size.

[ryta1203]

Originally posted by: genaganna

Originally posted by: ryta1203 Wouldn't I see a "failed to allocate host memory"? I'm not seeing this.Also, I've closed down every other non-essential program and I still get the same error:

KernelFailed.CL_OUT_OF_RESOURCES.

I heard that error is usually due to lack of available registers or local memory, which is why I tried reducing the work size but that didn't help.

Are you running Win Xp 32 bit?

My system:WinXP 32 bit

AMD Phenom x4 II 3.2Ghz

ATI 5870

4GB RAM

Any other ideas? This is very annoying. I can't run 4k*4k for Mersenne Twister either.

But I can run 4k*4k for DCT, for instance.

I am running on XP64.  Is this error coming from clEnqueueNDRangeKernel or clEnqueueReadBuffer?  Sample checks of resources before calling clEnquueNDRangeKernel.

One more thing sample not using local memroy at all.  why register size increase based on input size.

1. it's coming from clEnqueueNDRangeKernel

2. It doesn't make any sense that the registers would increase, but I know that the SDK 2.2 has register allocation issues. I just wanted to let you know that I tried reducing the work size to no avail, though you are right, I didn't think this would help (and in a normal stable predictable system I might not have tried that but AMD OpenCL SDK is anything but).

[ryta1203]

Originally posted by: genaganna

Originally posted by: ryta1203 Wouldn't I see a "failed to allocate host memory"? I'm not seeing this.Also, I've closed down every other non-essential program and I still get the same error:

KernelFailed.CL_OUT_OF_RESOURCES.

I heard that error is usually due to lack of available registers or local memory, which is why I tried reducing the work size but that didn't help.

Are you running Win Xp 32 bit?

My system:WinXP 32 bit

AMD Phenom x4 II 3.2Ghz

ATI 5870

4GB RAM

Any other ideas? This is very annoying. I can't run 4k*4k for Mersenne Twister either.

But I can run 4k*4k for DCT, for instance.

I am running on XP64.  Is this error coming from clEnqueueNDRangeKernel or clEnqueueReadBuffer?  Sample checks of resources before calling clEnquueNDRangeKernel.

One more thing sample not using local memroy at all.  why register size increase based on input size.

Any other ideas?

[ryta1203]

I also don't have a problem running this size on URNG.

No problem with MatrixTranspose either.

I have GPU_INITIAL_HEAP_SIZE=100 and GPU_MAX_HEAP_SIZE=100

For MersenneTwister, if using a 4k*4k size then:

seedBuf = 256MB

resultBuf = 512MB

Is this a problem? I have a 5870 with 1GB memory.

[ryta1203]

Originally posted by: genaganna

Originally posted by: ryta1203 Wouldn't I see a "failed to allocate host memory"? I'm not seeing this.Also, I've closed down every other non-essential program and I still get the same error:

KernelFailed.CL_OUT_OF_RESOURCES.

I heard that error is usually due to lack of available registers or local memory, which is why I tried reducing the work size but that didn't help.

Are you running Win Xp 32 bit?

My system:WinXP 32 bit

AMD Phenom x4 II 3.2Ghz

ATI 5870

4GB RAM

Any other ideas? This is very annoying. I can't run 4k*4k for Mersenne Twister either.

But I can run 4k*4k for DCT, for instance.

I am running on XP64.  Is this error coming from clEnqueueNDRangeKernel or clEnqueueReadBuffer?  Sample checks of resources before calling clEnquueNDRangeKernel.

One more thing sample not using local memroy at all.  why register size increase based on input size.

For the twister sample I'm getting an INVALID_BUFFER_SIZE for resultBuf.

It appears that despite having set those environment variables, my max buffer size is still only 256MB, this is such a huge limitation.

[ryta1203]

Originally posted by: genaganna

Originally posted by: ryta1203 BTW, I'm running a 5870 with 1GB of ram. I only mention this because the Black Scholes runs fine at 2k*2k problem size and I only get this error when moving up to 3k*3k or 4k*4k.

Ryta1203,

       I am able to run without any problem. It looks like you are facing problem with CPU resources. Please try to close other applications running on your system.

No, I don't think this is it. I think I am facing an issue of exceeding the max heap size (which is a GPU problem).

At 4096*4096, the total buffer allocation is 768MB (256+256+256) for the BlackScholes problem, which I'm assuming is beyond what is allowed.

I have tried to play with the env variables but that is not working, where can I find documentation on those variables?

[genaganna]

Originally posted by: ryta1203No, I don't think this is it. I think I am facing an issue of exceeding the max heap size (which is a GPU problem).

At 4096*4096, the total buffer allocation is 768MB (256+256+256) for the BlackScholes problem, which I'm assuming is beyond what is allowed.

I have tried to play with the env variables but that is not working, where can I find documentation on those variables?

Ryta1203,

I am able to run same on HD4850, XP64. I am not able to understand why you are getting that error.  I am not using any env to increase or decrase heap size.

[ryta1203]

Originally posted by: genaganna

Originally posted by: ryta1203No, I don't think this is it. I think I am facing an issue of exceeding the max heap size (which is a GPU problem).

At 4096*4096, the total buffer allocation is 768MB (256+256+256) for the BlackScholes problem, which I'm assuming is beyond what is allowed.

I have tried to play with the env variables but that is not working, where can I find documentation on those variables?

Ryta1203,

I am able to run same on HD4850, XP64. I am not able to understand why you are getting that error.  I am not using any env to increase or decrase heap size.

So you can also run MersenneTwister for 4k*4k size?

[ryta1203]

Originally posted by: genaganna Ryta1203,

I am able to run same on HD4850, XP64. I am not able to understand why you are getting that error.  I am not using any env to increase or decrase heap size.

Could this be a problem with XP32? Because that is what I am using, are you able to run it on a XP32 version?