Hi everyone,
This article contains very useful information on how Pix4D uses the computer resources.
In addition to that:
-
CPU
Pix4D does utilize multicore-multithread technology during various steps of processing.
-
Initial processing is highly multi-threaded so multicore matters.
-
The ortho and reflectance map is heavily multi-threaded too
-
Densification is also multi-threaded
It is better to invest in a computer with a single CPU with many cores rather than dual CPU systems
- GPU processing
Pix4D can use a GPU to accelerate processing, especially for feature extraction/matching.
The GPU usage is not in the log file but can be tracked with Nvidia Tools http://www.nvidia.com/object/system_monitor.html
More information on Pix4D GPU usage and recommended GPUs here.
We have seen better performance with GeoForce GPUs over Quadro or Titan cards.
Only one GPU and CPU are used. There is not any advantage of using two CPU or two GPU graphic cards.
Regards.
Hello,
I am about to build a system with the following specification so please guide me if any changes required as I’ll be processing large no of images, probably around 20000.
AMD Ryzen Thread Ripper 2990wx with 32 cores
Asus x399 MB
Corsair Vengence 3000MHz 64GB
WD Gold 12 TB
Samsung 960 Evo PCIE 500 GB
Corsair HX 1200 SMPS
Nvidea 1060 DDR5 6GB
Thank you
That isn’t enough system or video RAM.
1 Like
If I’ll add 2080TI 11GB and 128 GB ram will it be enough?
Can you suggest me some configuration?
the 2080ti is unknown at this point. The 1080ti is proven and works well so if budget permits you may want to consider that until more information about the 2080ti cards is out there. 128GB of system RAM is definitely your best bet to get the projects to process. Even then there is no guarantee. Check out the article below for some tips on processing large datasets.
https://support.pix4d.com/hc/en-us/articles/202558579-Processing-Large-Datasets
Holden Thanks,
Is there any information available abt AMD Ryzen Thread Ripper 2990wx?
How does it perform with 32 cores?
If there is any comparison available, it would be really helpful to me.
I have not seen any testing data for the 2990wx. Puget Systems tested the 1950x but that is all that I have seen. The additional cores will probably not have as much benefit on reducing processing time but would be useful if you wanted have other applications open while processing. That is likely the main benefit that I could see to having that many cores.
https://www.pugetsystems.com/labs/articles/Pix4D-CPU-Comparison-Coffee-Lake-vs-Skylake-X-vs-Threadripper-1084/
Holden Thank you,
I will rather wait for some time to see if any comparison or test results show up.
We must be need 8 or 16 GB of RAM and 15 and 30 GB of free SSD space. If you want to do larger projects however you will need 32GB of RAM and as much as 120GB of free SSD space.
@Adam Jordan
I’ve been following your post for a while and a was meaning to ask… Where do you recommend to buy a pc like yours?? I’m from México City and it’s very difficult to find PCs like yours. Can you help me? or even sell me one of yours?
As of today… If money was no object what PC specs would everyone recommend getting to get a 3d photogrammetry model out as fast as possible? I am open for not just one system, but maybe multiple systems working on different collection sets. Any help would be much appreciated.
Ive been running some new test on a RTX2080.
I built a new machine and the specs are as follows.
CPU - 8700k overclocked to 5.2Ghz with a avx offset of -4 (when avx instructions are running it clocks to 4.8Ghz)
CPU Cooler - Corsair 115i Pro
GPU - NVIDIA RTX 2080
RAM Amount - 64GB
RAM Speed - Corsair 3200
Hard Drive - 1TB M.2 Samsung 970 EVO
Motherboard Model - Asus ROG Maximus X Hero (WiFi)
Operating system - Windows 10
Pix4D Version - 4.2.27
I am testing if 64GB ram is better then 32GB and the differnce between the 2080 and the 1080ti that I had.
I should be getting a 2080ti soon as well to test out.
Brian
Also, If anyone would like me to run a set for them please let me know. All I have to run are the test files in this thread and my own farm files that we process using the ag multi spec settings.
So is the main performance from the CPU mostly pure clock speed, and not the amount of cores?
@Brian Young. Wow thats pretty top of the line hardware. If you are willing it would be really interesting to see what results you get from the Puget Systems benchmarking tool. See link below. If you are able to test let us know what results you get!
https://www.pugetsystems.com/labs/articles/Puget-Systems-Benchmark-Tool-for-Pix4D-1155/
@Holden and anyone else interested. I will run Puget benchmark tonight.
Here are the Puget System Benchmark test results from my current system posted above.
Eagle 3D Model:
Step 1 was 99 seconds
Step 2 was 337.7 seconds
Total time 436.7 seconds
DJI Building Map:
Step 1 was 115.5 seconds
Step 2 was 276.6 seconds
Step 3 was 405.9 seconds
Total time 798 seconds.
Here is Puget Systems own tests copied from the link above.
Here are sample results from workstations built here at Puget Systems in the first few weeks of our using this benchmark internally:
| System Specs | Eagle Model Step 1 | Eagle Model Step 2 | Eagle Model Total | Building Map Step 1 | Building Map Step 2 | Building Map Step 3 | Building Map Total |
| Core i7 8700K +
GTX 1050 Ti 4GB | 152 | 383 | 535 | 163 | 302 | 465 | 930 |
| Core i7 8700K +
GTX 1080 8GB | 117 | 373 | 490 | 140 | 300 | 440 | 880 |
| Core i9 7960X +
GTX 1080 Ti 11GB | 119 | 278 | 397 | 147 | 226 | 292 | 665 |
| 2 Xeon E5-2690v4 + Titan Xp 12GB | 148 | 317 | 465 | 183 | 256 | 341 | 780 |
I will post my own results from the projects we have here in this thread in a day or so.
I’m not near my HEDT right now but I’ll run the benchmark later. My GTX 2080 is behind a Threadripper 1950x. At the time it was a better dollar to dollar buy than what Intel was offering though with the new 9000 series chips from Intel the line between consumer and HEDT processors is blurring. I’d wait to see what the 9000 series looks like since it’s the first major shakeup to Intel’s lineup in over a year.
One other piece of technology I upgraded to was a 9 series Optane drive. There is a lot of benefit to going to a true PCI-E drive for operations that take a lot of read/writes. NVME doesn’t have the power output, literally in terms of wattage, to support the fastest versions of solid state memory and the 2x-5x speed increase has been worth it to me though I haven’t done a Pix4D benchmark yet.
1 Like
Thanks for sharing Brian. Impressive results. Really blows the stock 8700k and GTX 1080 out of the water. 82 seconds may not seem like a lot but at scale that could be a real time saver. Thanks for sharing. A good indicator of RTX performance.
Can someone confirm that RAM speed it’s something to consider? I’m making some upgrades on my PC and I’d like your inside on changing my stock 64 Gb 1400 MHz to 64 Gb at 2600 MHz. Thx!
Hi Raul. I have not seen any data to suggest that memory speed has a significant impact on processing time. Not to say that it doesn’t. We just don’t have testing data to show that it does. I am going to share some raw benchmarking data that you can take a look at. I would consider everyone who is interested to take a look and see if there are any insights that people can gain from the data. Be sure to share any interesting conclusions!
https://docs.google.com/spreadsheets/d/1WUCH3aTcaAglRwdH4j88B-iChhFyugFKCjuGcPPzMsY/edit?usp=sharing
1 Like