Hi there - we currently use Pix4d and run it on a ‘Fast’ PC which was designed for gaming and has a CUDA compatible GPU. Projects process reasonably quickly but with the amount of data-sets we are currently processing for ourselves and other companies, we are looking to invest in a server type platform in order to reduce processing times.
Are there any recommendations on the best way to go hardware wise? We currently use Mac Pro’s/iMac 5K’s as our office computers from which we could harness some serious processing power if we could network process (hopefully this feature comes soon to the software)… But are happy to invest in PC based systems if required; has anyone tried a bitcoin mining setup to process Pix 4d data?
Thanks in advance all,
Matt
I am in a similar situation to Matt, currently using a high end gaming machine that is beginning to bog down with data sets of ±2000 - ±10mp images. I am looking at a Xeon (Xeon E5-2697V3) based machine with Quadro graphics (NVIDIA Quadro M5000 8GB), but before I fork out the many $$$ required it would be good to have a direction from the software developers. Further, would a dual Xeon processor machine with dual (or more) graphics be advantageous. I understand the software must be designed to use multiple processors, is this the case with Pix4d??
Look forward to some direction
Bill
This is an on going question for many and becomes costly and complex very quickly
P4D advice is to divide jobs into smaller sub-jobs processed in parallel.
At some point volume jobs will become impractical for local processing, (even for local parallel processing techniques) P4D has work-flow solution for off site processing that could reduce hardware and time for processing cost. I have no experience with this work-flow but have looked at this for some time. Photometric data capture has become so cheap that I can’t see hardware processing cost catching up any time soon (the essence of the question…capture pipe cost vs processing pipe cost vs output cost).
FWIW…I us i7-49xx cpus and K5xxx gpus, Think I would be better served using GPX-9xx gpus and putting the money into more ram and local hardware for now any ways.
I wouldn’t waste your money on a Quadro video card, invest in your CPU and ram, lots of fast ram. Most of the heavy lifting is done by your cpu and ram. Our processing workstation is a Haswell-e 5820K, on a MSI X99 m/b, with 128GB of 3000MHz ram, upgrading to a 5960X as they’ve come down in price. I’d advise using a somewhat decent CUDA Video Card, 970+ but don’t waste your money on a Quadro as Pix4D depends on the video card very little, particularly the mesh stage.
I’m getting decent results, and make sure you buy the fastest SSD you can afford, makes a difference. M.2 drives have come down in price and utilize a 32Gb/s transfer rate. A 256GB drive should suffice.
Good luck
I am interested in some more info on these types of setups. My issue is I have been running mac for 10 years now so I am out of the PC lingo and specs.
What are the average processing times (say per 100 images) I am wondering if an in house oc or server can be purchased and run (cost effectively) versus running a could computing server like AWS or DroneData? I would ideally have processing on site but not if the cost is far too high.
I have hear most of you say good ram and good processor with OK video card should be good.
Would something along the lines of this : http://www.memoryexpress.com/Products/MX46512 be a decent machine, maybe double the ram and get a superfast ssd for the working drive?
If thats a bit frivolous, what would some specs to tell a computer builder be to get a system that should last and run nicely. I tried the beta on my mac and I dont think 6-8 hours of processing is realistic for a business to run well.
Cheers,
Todd
I know this is an old thread, but wanted to post a comment on video cards and processing time, pasrticularly in light of a statement above that video cards do not make much difference.
We are using a Xeon based system that originally had a K4000 Quadro card in it (~750 CUDA cores) which we updated to a GTX-1080 card (~2500 CUDA cores), and this had a significant impact on processing times. The baseline job I have been using for testing is a 300 picture job, with large 36 MP images. On the original system, the job took just under 5 hours to process at medium point cloud density, ~80% image overlap. With the video card upgrade to the GTX 1080, the processing time improved by about 55 minutes or so - this was a 20% increase in processing speed. Most of the benefit (more than half) was realized in step 1, but other steps were faster as well.
Our system is a Dell T3610, with a E5-2667 V2 CPU (fast 8 core), 128GB of RAM, SSD’s and the above mentioned video cards.
Tom
Hello all. Puget Systems has developed an easy to use benchmarking tool for Pix4D that you may find interesting. You can find it HERE
As this is not a Pix4D product, it is not an official benchmarking tool but you can use it at your own discretion. Also, the information on the page is helpful if you have questions about how your system compares to another setup.
Feel free to share your times and hardware setups on this thread!