Yeah, the Xeon “Scalable” series (the ones with metalic names, like the Gold 6144 you mentioned) don’t make any sense in a single CPU configuration. Even with two of those, I don’t think it would beat a Core i9 14-18 core chip or a similar Xeon W. And as noted above, the Quadro can perform on par with a GeForce in Pix4D, provided the right settings tweaks, but that doesn’t justify spending a lot more money for one (in the case of Pix4D).
I’m going to be publishing some articles next week, looking at the various Core i9 processors and how they stack up in Pix4D. If y’all would find that interesting, I can post links here once the articles are finished.
Bill - I use the I9 7900X with 64GB of RAM and a 1080 Ti and Pix4D works very well compared to a dual Xeon HP Z840 with 128 GB of RAM and a Quadro K5200.
I also use Faro Scene and Autodesk Recap 360 Pro with no issues.
Recently, exporting edited 4K video in Adobe Premiere Pro CC 2018 using the i9 machine far exceeded my i7 3.4GHz 27 inch iMac (late 2012) with 32 GB of RAM and NVIDIA GeForce GTX 680MX 2 GB. Mind you, the iMac is my daily working machine. I recently had the screen replaced by Apple, but thats another story! All hail the longevity of Mac’s versus PC’s…
The reason I brought the iMac into this discussion is that a year or so ago it was faster than the dual Xeon machine when I ran a side by side comparison using Pix4D.
My own conclusions in all of this are the nVidia drivers being the root cause and ultimate culprit of 99% of P4D processing lag across all machines. There were some discussions about hyper threading causing issues a year or so ago on this forum but I haven’t heard the subject some up for a while.
Thanks Bill, and do post your article links please.
Okay, here is the article about CPU performance in Pix4D:
https://www.pugetsystems.com/labs/articles/Pix4D-CPU-Comparison-Coffee-Lake-vs-Skylake-X-vs-Threadripper-1084/
Another will be coming today or tomorrow, looking at GPU performance (including the issues we’ve talked about here with Titan and Quadro cards).
1 Like
Bill, great article and data on small to medium projects. It shows good trends across varied types of data.
I look forward to seeing the GPU data and how it matches with my single vs. dual GTX1080Ti tests.
Okay, the GPU article is up now: https://www.pugetsystems.com/labs/articles/Pix4D-GPU-Comparison-GeForce-Titan-and-Quadro-1085/
I also put together a quick blog post about cloud vs local processing times: https://www.pugetsystems.com/blog/2018/01/05/Pix4D-Cloud-vs-Local-Processing-1089/
Hopefully this info is helpful! I’d love to hear any feedback y’all have about the articles, or suggestions for improving the testing for the future 
1 Like
Nice articles Bill and they match all of my hardware tests as well.
Things get a lot more separated on large gigapixel projects instead of the small tests you did but then again very few people run the size of projects I do
One note on GPUs, some options can’t run on lower VRAM that affect accuracy…not really in your scope but it becomes a differentiating factor on the GTX series when high-end settings simple crash on low-end video cards.
Thanks for the feedback, Adam! I’d love to get my hands on some larger projects, but the City one (2691 images) was the biggest I had access to.
Out of curiosity, what sort of image count and resolution are you working with? And do you know of any sources for similar-size projects?
I wish I could give you access to some of my projects but they are not for the public. I run from 2,000 to 5,000 pics per project with 42MP photos. They take from 2 days to 3 weeks to process on the best hardware at high-end settings.
“They take from 2 days to 3 weeks to process…”
0.0
Wow, I wouldn’t be able to run those in a timely manner on different systems / specs anyway
I am sort of curious what settings folks like yourself use, though. I stuck with the standard project settings (3D Model and 3D Map) for my testing, to keep things simple, but it might be interesting to see how commonly-used high-end settings impact things… especially RAM usage.
Those are great benchmarks. Thanks guys, I’m not going to order up a Titan xP then and I’ll focus on memory speeds for now.
SO being late to the game. I ordered a new machine for processing. Typical image sets are between 800-2500 images. The new machine on the strong recommendation of our company’s IT guy was a Ryzen Threadripper. I could only afford the 1920 (12core). I also bought the Titan 12GB thinking it would be better and not knowing this thread existed. SSD M.2 drive and 128GB RAM.
Obviously the machine is a disappointment so far, as it’s actually slower than my company laptop (1400 images, Step 2 was 17 hrs on the laptop,17.5 on the big new machine).
I managed to exchange the Titan for a GTX1080i and, while I’m waiting to see how long it takes in the end, one thing I’ve already noticed is that while P4D sees the 1080, the CPU is running pinned at 100%, and where the Titan was running at 20-25% the 1080 is only at 5%. With the CPU at 100% the computer is not stable, opening any program causes hesitation.
Suggestions? I’m way too deep into this machine to make it a gaming toy and buy another (I’m not a gamer for one, and don’t have the $ for another).
I’ll await the final results of the re-processing of step 2, but early indications suggest the Titan was at least contributing.
Matthew,
First, never run anything else with Pix4D, especially in Step 2.
Second, 100% CPU is expected in Step 2 with little to zero GPU usage…no idea why the Titan showed any usage.
You have the right hardware to have the best Pix4D machine out there…mine is Intel but otherwise identical. I have used my machine to process a 200 gigapixel project so what you have is certainly capable. Lastly, running the highest accuracy, highest detail, and lowest noise is never the quickest…you have the hardware to max the settings and thus cheaper hardware with inferior settings could easily be quicker.
Matthew,
Moving away from the Titan is a good move! Its a fine card for most other applications, but something about Pix4D’s usage never triggers the Titan (or Quadro) cards to really kick up to their full speed. The GeForce 1080 Ti should be much better for you in that regard
As for the CPU, unfortunately Threadripper isn’t the best option out there… but it isn’t the worst, either. I wrote a CPU comparison article recently that might be of interest to you, and may help steer any future hardware purchases you make for this application:
https://www.pugetsystems.com/labs/articles/Pix4D-CPU-Comparison-Coffee-Lake-vs-Skylake-X-vs-Threadripper-1084/
And echoing Adam, I would not plan on running anything else alongside Pix4D. If you do, it will be sluggish since Pix4D is trying (during most steps) to use as much CPU power as you have available. There are ways to manually reserve a couple of CPU cores / threads for Windows and other applications to use - search online for the term ‘affinity’ if you want to learn more about that - but doing so will slow Pix4D down as well.
Thank you Adam. I appreciate the response. The big machine and the laptop were running the same settings on the same dataset through the same step. I would think the Threadripper, SSD, and 8x the RAM would minimally have matched it. Perhaps it was all the Titan (?).
Question: anyone have an idea if the Titan will be supported within the next month? I have a huge project I’ve bid on which will be important I do as efficiently as I can. The computer generally ran well with the Titan, and I think it is generally better hardware. It would be a shame to abandon it with improvements in Pix4D coming which perhaps could make use of it.
I just checked the computer and it has crashed out of P4D again. Is it possible it’s the demo version (subscription expired)?
I need to see if there’s a GPU swap step I’ve missed.
Get rid of the Titan and never go back…GTX 1080Ti with 11GB is enough for 42MP pics.
I am about to downgrade because the latest official release has crashed 7 times in a row…ridiculous.
The issue between Titan and Pix4D is due to the CUDA libraries within the Pix4D application.
@Matthew Joyce
- Under normal circumstances, your trial license ending would not have caused your project to end processing prematurely or crash. If you would like us to investigate the source of the crash, please let us know at https://support.pix4d.com/hc/en-us/articles/hc/en-us/requests/new.
- We are investigating the source of the extended processing time with the NVIDIA Titan Xp and NVIDIA Titan X (Pascal) GPUs and will release a patch as soon as possible. I cannot provide you with an expected release date or timeframe but can assure you that a patch will be included in a future build if we identify the cause of the problem and resolve it.
As Philip Angell mentioned, we have reason to believe that the processing slowdown with the NVIDIA Titan Xp and NVIDIA Titan X (Pascal) GPUs, especially during Step 3, is due to the use of an outdated library. However, I cannot stress enough that the investigation is still underway and that I will comment here again as soon as I have an update on the status of the investigation.
If you must purchase new processing hardware in the meantime, please consider a different series of GPU. I recommended considering one of the following GPUs from NVIDIA. Please note that this list is not exhaustive.
- GTX 750
- GTX 970
- GTX 1060, 1070,1080
If you have experienced extended processing times during Step 3 and would like to contribute to the investigation, we would greatly appreciate your help. Please let us know by contacting us at https://support.pix4d.com/hc/en-us/articles/hc/en-us/requests/new.
You all realize that the most expensive Threadripper is the same price as the cheapest i9 right? Threadripper is just fine for Pix4D especially dollar for dollar and if you do any other HEDT tasks on your PC Threadripper is likely the better buy. If you wanted to do a build dollar for dollar for Pix4D, the i9 and Threadripper are not the best buys. The Coffee Lake i7’s are.
As others have said, Pix4D runs best alone on PC. You can go into the processing options and limit the amount of cores Pix4D uses if you need to do other things. Of course this will lead to longer processing times.
Thank you all. I tried again to run the software through strep 2 and it crashed prematurely again. I had installed the previous version (4.1.15) in hopes that perhaps it was the new version. At a bit of a loss.
Andrew, I submitted a question to the help desk and was told, in so many words, that individual help was not available to those without a current license: see this forum.
I have 7 days to shop a GPU back, before I dump the Titan - bear in mind Pix4D ran, if horribly slowly, with the Titan, but has not functioned once with the 1080i - I need to understand why this is crashing. Especially with the large project appearing more likely; I’ll be collecting the imagery the week after next, and will have a very short fuse to get this modeling done.