CUDA_ERROR_OUT_OF_MEMORY

[Xiaoying18]

Hi,
When I tried to run Lux with
./pagerank -ll:gpu 1 -ll:fsize 12000 -ll:zsize 20000 -file hollywood.lux

I got an error like this:

CU: cuMemAlloc(&fbmem_base, size) = 2 (CUDA_ERROR_OUT_OF_MEMORY): out of memory
pagerank: ../legion/runtime/realm/cuda/cuda_module.cc:2345: void Realm::Cuda::GPU::create_fb_memory(Realm::RuntimeImpl*, size_t): Assertion `0' failed.
Aborted (core dumped)

My nvidia-smi information is like this:
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 390.116 Driver Version: 390.116 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce GTX 108... Off | 00000000:02:00.0 Off | N/A |
| 53% 74C P2 263W / 280W | 6201MiB / 11178MiB | 98% Default |
+-------------------------------+----------------------+----------------------+
| 1 GeForce GTX 108... Off | 00000000:04:00.0 Off | N/A |
| 0% 37C P5 16W / 280W | 0MiB / 11178MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 2 GeForce GTX 108... Off | 00000000:83:00.0 Off | N/A |
| 0% 33C P5 22W / 280W | 0MiB / 11178MiB | 0% Default |
+-------------------------------+----------------------+----------------------+

May I get any help from you, please?

[jiazhihao]

Hi,

The -ll:fsize flag specifies the GPU memory allocation (in MB) per GPU. You can try to set it to a number before 11178, say 10000.

[Xiaoying18]

Yeah!! It does work!! Thank you for your reply!!^ - ^

…

On Sat, Apr 6, 2019 at 12:58 PM Zhihao Jia ***@***.***> wrote: Hi, The -ll:fsize flag specifies the GPU memory allocation (in MB) per GPU. You can try to set it to a number before 11178, say 10000. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#6 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AlVTHbWOB0hcFVykMj5w5SYLZl6f0Azcks5veNJNgaJpZM4cgB_9> .

[Xiaoying18]

Hi again,

I am very interested in the" graph repartitioning to get load balancing" part in the paper, but I am not very sure which part of the code is especially for the "dynamic graph repartitioning". May I get any suggestions from you about this, please?