has anyone met this error in above picture? showing “pycuda._Memory: cuMemHostAlloc failed: out of memory”
I will appreciate it if there is any explanation or solution ?
Hi @chenmy
That looks like a 2D classification job. You seem to be running out of GPU memory. How many particles do you have? And are they downsampled? If they are not downsampled, I would try to get them to a box size of around 100px (https://blake.bcm.edu/emanwiki/EMAN2/BoxSize) and try the classification again. Hope this helps.
Vamsee
Thanks for your replying!
I have 500,000~1,000,000 particles for one job, box size are 240.
Actucally, i used to run10 jobs, there was ok, but recently, if I run 5 jobs, 1 or 2 jobs may faied because of this error, somtimes it can be solved by retry, but sometimes not. So, is there something wrong with my particles’s box size?
By the way, you say “downsampled” what’s meaning of it?
1 million particles at 240 box size should work usually. I generally do my 2D classifications at box size of 128 or even smaller. After I’m happy with the final particle set, I go back and re-extract. This not only increases my speed of classification, it also reduces the total size on the disk occupied by the “extract from micrographs” job.
Downsampling is fourier cropping your particle images to a smaller box size. In your case, it would be by going from 240 to 100. The link I posted in the previous message should give you a good idea of what box sizes are the typically good to use computationally.