2d Classification failed by out of GPU memory

2D Classification
1

Hi Everyone,

I am trying to process big data set.
My system is 4 RTX2080 8G GPUs, 2 Xeon CPUs, 56G memory, 2T SSD, 11T data hdd drive.

Here is my case,
My molecule size is around 200 A.
Using CUDA-8

The first trial was box size 350 and 3.8m particles.
It worked for 2D classification. But 2D was still blurry after several rounds of classification.
Some 2D class edge were cropped.

The second trial was box size 384 and 2.9m particles.
It worked for 2D classification. 2D was still blurry after several rounds of classification.
Some 2D class edge were still cropped.

So I increased box size as 450 according to calculation posted in here.
Then I tried, but 2D classification was failed by out of GPU memory.

Is there someone can give suggestion for this?

Thanks,

Jinseo

2

Here are some updates,

Since I upgrade CUDA-8.0 to CUDA-10.0,
The 2D classificiation starts to work normally with 2.8m particles.
Also classification speed is a little bit faster than I think.

Jinseo

1 Like