GPU mode using CPU!

[alameen...@gmail.com]

I don't have a GPU in my machine.

I noticed if I add '-gpu 0' at the end of the finetuning command, execution time reduces significantly (compared to when '-gpu 0' is omitted)!

As I said I don't have a GPU and I'm running Caffe on CPU! 

Anyone can explain this weird behavior?

thanx

[Steven Clark]

Is the output / behavior of the net the same in both cases? Or is it just aborting early when -gpu is specified?

[alameen...@gmail.com]

test results are only slightly better (+1e-6) better when '-gpu' is used. the thing is that it doesn't abort when '-gpu' is used. I don't have a GPU! How come?!

[Yi Liu]

I'd have expected for you to run into various issues compiling a GPU enabled caffe with the standard procedures if you don't have one installed on your machine. Are you running on a Mac? Maybe you are using an onboard GPU of some sort?

[alameen...@gmail.com]

No I'm running Caffe on Ubuntu 12.4

[alameen...@gmail.com]

This is still a mystery to me! how I'm running Caffe on GPU mode even though I have no GPU installed in my machine!

[Darren Garvey]

On 11 September 2015 at 17:01, <alameen...@gmail.com> wrote:

This is still a mystery to me! how I'm running Caffe on GPU mode even though I have no GPU installed in my machine!

caffe device_query -gpu 0

That should tell you what (if any) GPU you have available.

Cheers,

Darren

[alameen...@gmail.com]

Here is the output of ./build/tools/caffe device_query -gpu 0 :

I0913 16:39:16.202206 31974 caffe.cpp:73] Querying device ID = 0

I0913 16:39:16.205225 31974 common.cpp:157] Device id:                     0

I0913 16:39:16.205245 31974 common.cpp:158] Major revision number:         2

I0913 16:39:16.205253 31974 common.cpp:159] Minor revision number:         1

I0913 16:39:16.205260 31974 common.cpp:160] Name:                          GeForce GT 630

I0913 16:39:16.205274 31974 common.cpp:161] Total global memory:           4294115328

I0913 16:39:16.205282 31974 common.cpp:162] Total shared memory per block: 49152

I0913 16:39:16.205288 31974 common.cpp:163] Total registers per block:     32768

I0913 16:39:16.205294 31974 common.cpp:164] Warp size:                     32

I0913 16:39:16.205304 31974 common.cpp:165] Maximum memory pitch:          2147483647

I0913 16:39:16.205312 31974 common.cpp:166] Maximum threads per block:     1024

I0913 16:39:16.205320 31974 common.cpp:167] Maximum dimension of block:    1024, 1024, 64

I0913 16:39:16.205327 31974 common.cpp:170] Maximum dimension of grid:     65535, 65535, 65535

I0913 16:39:16.205338 31974 common.cpp:173] Clock rate:                    1620000

I0913 16:39:16.205348 31974 common.cpp:174] Total constant memory:         65536

I0913 16:39:16.205353 31974 common.cpp:175] Texture alignment:             512

I0913 16:39:16.205363 31974 common.cpp:176] Concurrent copy and execution: Yes

I0913 16:39:16.205369 31974 common.cpp:178] Number of multiprocessors:     2

I0913 16:39:16.205375 31974 common.cpp:179] Kernel execution timeout:      Yes

[Detlef Schmicker]

I am not sure, if it is connected, but I do have a strange experience too:

When I updated caffe framework from December to actual version my GPU does not get warm anymore, if I am using classification network from my app.

My guess is, that GPU is not used anymore in classification?! 

[Youssef]

Hi,

I'm facing the same issue. I'm using Ubuntu 12.04 and caffe rc2. Did you manage to find solution to this issue.

[Yi Liu]

what does lspci say?