What would cause this error? GPU #3: aes_cpu_init out of memory
Intensity is too high.
#crysx
Topic: [ANN] ccminer 2.3 - opensource - GPL (tpruvot) - page 35. (Read 500193 times)
Intensity is too high.
#crysx
What would cause this error? GPU #3: aes_cpu_init out of memory
replace march=native by march=core2 in configure.sh then use ./build.sh
Finally, it works now!Thank you very much!
replace march=native by march=core2 in configure.sh then use ./build.sh
That seems to be the most generic I've found.
replace march=native by march=core2 in configure.sh then use ./build.sh
Should I put
Code:
-march=native
That will just makes it worse. What you want is a generic build but, straight from gcc...
There is no -march=generic option because -march indicates the instruction set the compiler can use,
and there is no generic instruction set applicable to all processors.
There is just too much that is different between AMD and Intel processors.
Edit: If you're only compiling for that one PC you can specify -march for that CPU.
https://gcc.gnu.org/onlinedocs/gcc/x86-Options.html
Edit2: It might also be useful to mention what CPUs are inolved. If the AMD newer than Piledriver -march-core2 should work.
I have had no success with Piledriver and older.
Did you use
Nope. As I said before all I made isCode:
-march=native
3. I edited Makefile.am:
Code:
...
#ccminer_CPPFLAGS += -DUSE_LIBSODIUM
#ccminer_LDFLAGS += -Lequi/lib
#ccminer_LDADD += -lsodium
ccminer_LDADD += -lcuda
nvcc_ARCH = -gencode=arch=compute_61,code=\"sm_61,compute_61\"
#nvcc_ARCH += -gencode=arch=compute_52,code=\"sm_52,compute_52\"
#nvcc_ARCH += -gencode=arch=compute_35,code=\"sm_35,compute_35\"
#nvcc_ARCH += -gencode=arch=compute_30,code=\"sm_30,compute_30\"
#nvcc_ARCH += -gencode=arch=compute_20,code=\"sm_21,compute_20\"
nvcc_FLAGS = $(nvcc_ARCH) @CUDA_INCLUDES@ -I. @CUDA_CFLAGS@
nvcc_FLAGS += $(JANSSON_INCLUDES) --ptxas-options="-v"
...
Should I put
Code:
-march=native
Guys... Could you help me please?... What did I make wrong?
1. I installed on my Ubuntu CUDA 9.0.176.
2. I downloaded the source from the following link (ccminer-cuda-9).
3. I edited Makefile.am:
4. I compiled ccminer. And here is strange thing - I got ccminer with the size about 18 MB. But last time (when i used CUDA 7.5) I got it with 34-35 MB... Myabe is it because of my Makefile.am? Ok, screw it, right?
5. ccminer --version showed everything is OK.
Now I copied just compiled ccminer and lib directory (with 3 files into it: libcudart.so.9.0 + libcudart.so.9.0 + libcudart.so.9.0.176) to other PC (/home/user/ccminer/) without installed CUDA on it.
I did:And then:
Why?
I've compiled ccminer on PC with Intel CPU. And I copied it toPC with AMD CPU. But before you've said that CPU doesn't matter, doesn't it?
1. I installed on my Ubuntu CUDA 9.0.176.
2. I downloaded the source from the following link (ccminer-cuda-9).
3. I edited Makefile.am:
Code:
...
#ccminer_CPPFLAGS += -DUSE_LIBSODIUM
#ccminer_LDFLAGS += -Lequi/lib
#ccminer_LDADD += -lsodium
ccminer_LDADD += -lcuda
nvcc_ARCH = -gencode=arch=compute_61,code=\"sm_61,compute_61\"
#nvcc_ARCH += -gencode=arch=compute_52,code=\"sm_52,compute_52\"
#nvcc_ARCH += -gencode=arch=compute_35,code=\"sm_35,compute_35\"
#nvcc_ARCH += -gencode=arch=compute_30,code=\"sm_30,compute_30\"
#nvcc_ARCH += -gencode=arch=compute_20,code=\"sm_21,compute_20\"
nvcc_FLAGS = $(nvcc_ARCH) @CUDA_INCLUDES@ -I. @CUDA_CFLAGS@
nvcc_FLAGS += $(JANSSON_INCLUDES) --ptxas-options="-v"
...
4. I compiled ccminer. And here is strange thing - I got ccminer with the size about 18 MB. But last time (when i used CUDA 7.5) I got it with 34-35 MB... Myabe is it because of my Makefile.am? Ok, screw it, right?
5. ccminer --version showed everything is OK.
Now I copied just compiled ccminer and lib directory (with 3 files into it: libcudart.so.9.0 + libcudart.so.9.0 + libcudart.so.9.0.176) to other PC (/home/user/ccminer/) without installed CUDA on it.
I did:
Code:
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/user/ccminer/lib
Code:
/home/user/ccminer/ccminer --version
Illegal instruction
Why?
I've compiled ccminer on PC with Intel CPU. And I copied it toPC with AMD CPU. But before you've said that CPU doesn't matter, doesn't it?
Did you use
Code:
-march=native
Guys... Could you help me please?... What did I make wrong?
1. I installed on my Ubuntu CUDA 9.0.176.
2. I downloaded the source from the following link (ccminer-cuda-9).
3. I edited Makefile.am:
4. I compiled ccminer. And here is strange thing - I got ccminer with the size about 18 MB. But last time (when i used CUDA 7.5) I got it with 34-35 MB... Myabe is it because of my Makefile.am? Ok, screw it, right?
5. ccminer --version showed everything is OK.
Now I copied just compiled ccminer and lib directory (with 3 files into it: libcudart.so.9.0 + libcudart.so.9.0 + libcudart.so.9.0.176) to other PC (/home/user/ccminer/) without installed CUDA on it.
I did:And then:
Why?
I've compiled ccminer on PC with Intel CPU. And I copied it toPC with AMD CPU. But before you've said that CPU doesn't matter, doesn't it?
1. I installed on my Ubuntu CUDA 9.0.176.
2. I downloaded the source from the following link (ccminer-cuda-9).
3. I edited Makefile.am:
Code:
...
#ccminer_CPPFLAGS += -DUSE_LIBSODIUM
#ccminer_LDFLAGS += -Lequi/lib
#ccminer_LDADD += -lsodium
ccminer_LDADD += -lcuda
nvcc_ARCH = -gencode=arch=compute_61,code=\"sm_61,compute_61\"
#nvcc_ARCH += -gencode=arch=compute_52,code=\"sm_52,compute_52\"
#nvcc_ARCH += -gencode=arch=compute_35,code=\"sm_35,compute_35\"
#nvcc_ARCH += -gencode=arch=compute_30,code=\"sm_30,compute_30\"
#nvcc_ARCH += -gencode=arch=compute_20,code=\"sm_21,compute_20\"
nvcc_FLAGS = $(nvcc_ARCH) @CUDA_INCLUDES@ -I. @CUDA_CFLAGS@
nvcc_FLAGS += $(JANSSON_INCLUDES) --ptxas-options="-v"
...
4. I compiled ccminer. And here is strange thing - I got ccminer with the size about 18 MB. But last time (when i used CUDA 7.5) I got it with 34-35 MB... Myabe is it because of my Makefile.am? Ok, screw it, right?
5. ccminer --version showed everything is OK.
Now I copied just compiled ccminer and lib directory (with 3 files into it: libcudart.so.9.0 + libcudart.so.9.0 + libcudart.so.9.0.176) to other PC (/home/user/ccminer/) without installed CUDA on it.
I did:
Code:
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/user/ccminer/lib
Code:
/home/user/ccminer/ccminer --version
Illegal instruction
Why?
I've compiled ccminer on PC with Intel CPU. And I copied it toPC with AMD CPU. But before you've said that CPU doesn't matter, doesn't it?
its the purpose of cuda-9 branch (was present before cuda 9 release)
yep, i dont put it for now because that breaks previous cuda builds... and increase the binary size a lot for nothing on windows
you maybe should add build ONLY for cuda 9, it would be easier for a lot of people.
yep, i dont put it for now because that breaks previous cuda builds... and increase the binary size a lot for nothing on windows
61 (and 60) will be slower than 52
I discovered that if you run 52 code on pascal, it actually makes 61 code on the fly when you run the kernel. The reason it is faster, I don't know. Thing is, with cuda 9 the difference is smaller or even reversed. A little tweaking here and there and it will be faster than before :-)
i already answered 5 posts ago
if that worked i would not asked again, tried that did not work ,still same error. I am missing something.
i already answered 5 posts ago
61 (and 60) will be slower than 52
But could you tell me please if I want to compile ccminer for GTX 1050 only regardless any CPU what should I do?
As I get all I need is add into Makefile.am the following flag:
Correct?
As I get all I need is add into Makefile.am the following flag:
Code:
nvcc_ARCH = -gencode=arch=compute_60,code="sm_60,compute_60"
and remove all other flags concerning nvcc_ARCH.Correct?
61, not 60.
THANK YOU! --
For the last three hours, I have been putting off digging into an old compile directory for the snippet of code that The Immortal just provided for himself. --scryptr
Jump to: