Why is my new PC 40% slower to calculate and write a grid surface?

Question

I've just updated a PC to do scientific calculations more quickly. The focus of the change was to step up to the Nvidia 1080Ti from the 980Ti.

Now I'm running a test model. I get the following results:

Stage             1080Ti PC    980Ti PC  
Set up            00:00:12     00:00:14  
Create grid       00:22:25     00:15:50  
GPU calc          00:02:24     00:03:25  
Export results    00:00:22     00:00:26

So you can see above that everything is faster except for the grid creation. The GPU stage is 30% quicker with the 1080Ti than the 980Ti which is great and the exports are 15% quicker. But the grid creation is 40% slower.

I think the grid step is mostly CPU and HDD reliant. So I did CPU and HDD benchmark suites on Phoronix Test Suite. It found that the i7-7700k CPU is better than the i7-4790 across all 14 benchmarks I ran. But when I ran the HDD benchmarks the Samsung 960 EVO NVMe SSD was mostly better than the Samsung 850 SSD except for a few tests, in particular the dBench 1 Client mark was 40% worse and the the SQLite Insertions mark was 80% worse. HDD Benchmark Comparison So I'm assuming the issue is with the HDD.

So my questions are:

1. Do you think the Samsung 960 EVO HDD has resulted in that 40% performance reduction compared to the Samsung 850 SSD?
2. Do you have any suggestions on how to fix it?

Answer 1

Credit for this answer goes to my local MSY technician. He saw this article: http://www.pcworld.com/article/3173618/linux/kaby-lake-is-unleashed-with-kernel-410.html and it made him wonder if the 4.4 linux kernel was restricting the NVMe 960 EVO HDD somehow. So I updated from kernel version 4.4 to version 4.10 and with no other changes there was a 39% boost in the performance of the grid step. As a bonus there was a 3% boost in the performance of the GPU calculation step.

Stage             1080Ti PC    1080Ti PC    980Ti PC  
                  kernel 4.10  kernel 4.4   kernel 4.4  
Set up            00:00:12     00:00:12     00:00:14  
Create grid       00:13:45     00:22:25     00:15:50  
GPU calc          00:02:20     00:02:24     00:03:25  
Export results    00:00:22     00:00:22     00:00:26

This doesn't confirm for sure that the 960 EVO NVMe SSD was the source of the problem. But it at least gets the system to the point where it is quicker than the old one for that task.