Forum,
Looking for help and tips on very slow computing. My computer speeds up as it crumches data, but is still very slow. With Head tracking, I am getting 8.0 to 2.8 secs/frame. GPU loading for both GPUs is always less than 27%, CPU loading is always under 12%.

Any thoughts on how to make this faster are very appreciated. The only strange thing on my system is both monitors rotated to portrait (tall).

iPi Pro + Biomech 3.2.0.182

Manufacturer Dell Inc. ------------------------------------------------------
Model Precision T7610
Intel(R) Xeon(R) CPU E5-2620 v2 @ 2.10GHz, 12 core (2x, 24 core total)
Total amount of system memory 16.0 GB RAM
Windows 7 Professional 64-bit operating system

Storage --------------------------------------------------------------------------------
Total size of hard disk(s) 745 GB
Disk partition (C:) 220 GB Free (279 GB Total) OS
Disk partition (D:) 439 GB Free (466 GB Total) Data Disk

Graphics--------------------------------------------------------------------------------
Display adapter type NVIDIA GeForce GTX 770 (2x)
Total available graphics memory 4096 MB
Dedicated graphics memory 2048 MB
Dedicated system memory 0 MB
Shared system memory 2048 MB
Display adapter driver version 9.18.13.4752
Primary monitor resolution 1200x1920
Secondary monitor resolution 1200x1920
DirectX version DirectX 10

???????????????????????

Hi
What are the performance numbers when using a single GPU in Mocap Studio?

Do I need to physically take the card out, or just uncheck it in iPi Mocap studio?

With the 2nd card unchecked, processing is much better.

Head Tracking
2.051-1.161 Secs/frame

No Head Tracking
0.809-0.631 Secs/frame

GPU loading was up to 65.5% as well.

Thank you for the feedback.
We have already noticed a decrease of performance on one of our dual-GPU configurations and are working on a fix currently. We'll notify you when we have a beta version for testing.

Vasili,
Thank you for the response. Please let me know if there is any data I can post, or tests or Beta's I can help you with.

Also, a question of philosophy and your programming methods:

In general, do you recommend 1x $400 GPU or 2 $200 GPUs for performance and speed of processing?

I am not sure if this question has an answer without knowing how your team thinks about distributing load across the GPUs, and then re-combining and re-sync'ing the work streams.

I assumed 2x cards at half the price would be a better option, and built my system that way.

Generally, we recommend the opposite: one high-end GPU is better than two mid-range GPUs.
With current algorithms in our software 2 GPUs provide at most 1.6 times improvement over 1 GPU of the same model. Moreover, with "light" camera configs like 1 Kinect or 3-4 PS Eyes dual GPUs give no advantage and may actually perform worse. Because in this case computations are not so intensive to fully utilize even a single GPU, and multiple GPUs always require some overhead to distribute the payload.

V3.2.3.190 contains performance improvements for multiple GPUs which may solve this issue.