CUDA error in func 'init' at line 245 : out of memory

[vinid223]

I get the following error message

cu 18:11:22|ethminer Using grid size 8192 , block size 128
m 18:11:22|ethminer Waiting for work package...
m 18:11:22|ethminer Got work package: #a0b08556
ℹ 18:11:22|CUDA0 set work; seed: #c5f7a893, target: #0000000225c1
ℹ 18:11:22|CUDA0 Initialising miner...
m 18:11:23|ethminer Speed 0.00 Mh/s gpu/0 0.00 [A0+0:R0+0:F0] Time: 00:00
m 18:11:24|ethminer Speed 0.00 Mh/s gpu/0 0.00 [A0+0:R0+0:F0] Time: 00:00
cu 18:11:24|CUDA0 Using device: GeForce GTX 780 Ti (Compute 3.5)
m 18:11:24|ethminer Speed 0.00 Mh/s gpu/0 0.00 [A0+0:R0+0:F0] Time: 00:00
CUDA error in func 'init' at line 245 : out of memory.
./start.sh: line 1: 10952 Segmentation fault (core dumped) ./ethminer -U -F

For some reasons after I rebooted the computer I get this error. I tried to reinstall the NVIDIA Driver and rebooted again and I still get this error message. I am on Ubuntu 17.10 with the NVIDIA Driver 384.

It was running fine for more than 3 days in a row and even restarted some time and now I get this error and I don't really know how to fix that.

I also tried using OpenCL with the -G command instead to try it out and I get the error
cl 18:08:18|cl-0 Loading kernels
cl 18:08:18|cl-0 Writing light cache buffer
cl 18:08:18|cl-0 Creating buffer for header.
cl 18:08:18|cl-0 Creating mining buffer
cl 18:08:18|cl-0 Generating DAG
✘ 18:08:18|cl-0 clEnqueueNDRangeKernel ( -4 )
cl 18:08:18|cl-0 Switch time 4947 ms / 2242203 us
✘ 18:08:18|cl-0 OpenCL Error: clEnqueueNDRangeKernel -4

I have a 780 TI 3Gb
Any Ideas?

[orkblutt]

Same here on 2 rigs:
I have to use --duda-devices x y z to see what's happening
i suspect dag size a bit bigger and cuda alloacting not only in gpu memory but in ram too...
I think my rigs needs more ram. can you confirm?

[orkblutt]

on 1070s x 6 rigs : i can't start the 6 card., only 5 at same time.

os: win10 64 with 4 Go of RAM

[vinid223]

I have 16 Go of ram so I don't think that is the problem

[emily-pesce]

If on Windows, make sure virtual memory is huge. I had to go to 24GB before I could get a 7 card rig to run. (I have 4GB of RAM)

[orkblutt]

Thank you Michael!! It's working back now by increasing virtual memory (swap)...
Have a good day!

[JamesTheAwesomeDude]

This precise error, same line number and everything is occurring on a GT 720M on Debian 9.2.1 with kernel 4.9.0-4 and the NVidia OpenCL drivers

clinfo

Adding 20GiB of SSD swapspace did not change anything

[Algomorph]

I think I'm getting the same error.
The rig is on (currently) latest master 6fbf438, Ubuntu 16.04, 16GB RAM, GeForce 760 GTX, I get:
CUDA error in func 'init' at line 247 : out of memory.

I'm using vanilla CUDA 9.1.85-1 from the official Nvidia repo, kernel is 4.4.0-104.

[JamesTheAwesomeDude]

I've heard from a few places that you're just out of luck if you have a 2GB card due to DAG technicalities

[vinid223]

Like I said in the issue @JamesTheAwesomeDude I have a 3Gb GPU

[emclab]

I have same issue with 6GB GTX 1060 card on EthOS 1.2.9.

[DeadManWalkingTO]

Try latest version and feedback please.
Thank you!