So weird! It seems Nvida GTX1080 is slower than GTX1060

PIX4Dmapper PIX4Dmapper Questions/Troubleshooting
1

I just bought two PCs for drone image processing using Pix4d. The specs of the PCs are as follows:



it’s obviously that PC1 should be much faster than PC2, which is also verified by the overall performance test from a third party software.

However, it turned out PC2 is faster than PC1 when i run a test dataset (47 12M images) with same configurations with two PCs using Pix4d (see
follows).

its said the fist step is most GPU consuming. How come a slower GPU (GTX1060) works even faster than a faster one (GTX1080)? it’s noted that both  GPU are upgraded to the latest driver.

Anyone come across the similar issue? ANY comments would be greatly appreciated.

 

 

2

We have done in house testing with a pc1 i7 6700 with a M.2 and a gtx 970 compare to pc2 dual xeon 2670 with Samsung ssd and a gtx 1070. Pc1 will eat pc2 in step 1 by almost half of the time . The only difference was the storage and that quickly changed when I installed a pcie ssd into pc2 . Now step one is a bit quicker on PC2 as it should be. I think you issue may lay there as the samsung ssd’s are usually quicker then most other ssd.

3

Thanks for your comments, Arturo sosa.

In my tests, i just run the first step, of which CPU and GPU is very important while the hard disk is of low usage (Plz refer to “Hardware Components Usage when processing with Pix4Dmapper”  https://support.pix4d.com/hc/en-us/articles/202559519-Hardware-Components-Usage-when-processing-with-Pix4Dmapper). .%C2%A0)

therefore, brands of ssd don’t matter as for the processing of the first step. Pix4dMapper recommend GTX 750, GTX 970 for the processing, which seems quite out of date. Thus, I come to wonder whether the software can full exploit the power of GTX 1080.

4

Yes I am am aware of the hardware component usage list. Pound for pound the GTX 1080 is better. More cuda cores and faster clock speed. Just going hardware specs you posted, the ssd is the only thing I see that can limit here. Yes I know there is low usage , but I know Pix4d access the ssd in step one in burst where the speed will matter.

5

Thanks again for your comments.

It seems i have to update the SSD in PC1.  

6

Hello Hui and Arturo,

Indeed, PC1 is much more powerful than PC in all aspects (CPU, GPU, RAM) and processing should be faster than in PC2.

It is true that the hard disc is not the most important component in step 1, but there is still the need to load and write files on the disc, so the SSD can indeed make a difference on the performance. Also, even though many processes of step 1 are computed in the GPU, the speed up is visible in big projects of high resolution and high texture. In this case, you are comparing a machine with a powerful GPU with a machine with a more powerful GPU. Therefore, we do not expect to see big differences in the performance between the two, especially for a project of 47 images. Having said that, we also do not expect PC1 to be slower than PC2. 

As a side note, please make sure to start your test projects from scratch deleting the results folder (if any) each time you process on the various machines. If a project is already processed, when it is reprocessed the keypoints and the matches are loaded from the disc instead of being recomputed. Therefore, the comparison between a project processed from scratch and a reprocessed project is not valid, especially when talking about GPU processing, since these two substeps (keypoint extraction, matching) are processed in the GPU. 

Best regards,