I am configuring a more powerful workstation build to decrease processing time for our larger projects. I currently have an i7 7700k with 16gb ddr4 and a 1080 and it took three days to process our last phase one camera projects. I am weighing the 10 core i7 vs a dual xeon setup and I cannot find a difinitive answer on the best setup to get and which xeon should be better than the other. Will dual xeons allow me to process two at the same time better than one xeon or will getting a crazy amount of cores not speed up my processing that much.
Taylor,
How large are your projects?
I feel your pain if you are using a Phase One camera - lots of megapixels there. We went through a lot of research several months ago on a similar path. We are using 36 megapixel images (Sony A7R and Nikon D810 cameras), so we were looking for some horsepower improvements on the processing end at this point. It seems like there is more information out there on Agisoft software than there is for Pix4D in terms of hardware recommendations. We are awaiting delivery of our new system at this time (a week or two out at this point). Some things I learned in research and in talking to people:
1. Dual processors and/or high core counts are good to a point, but overall processing speed is just as important. One of the specs to keep an eye on, and you have to dig for this information sometimes, is what the peak clock speed of the CPU is with all cores running 100%. While current Xeon CPU’s are available in core counts as high as 22 per CPU, the clock speed on these highest core count processors drops some over slightly lower CPU’s in the lineup. Based on this info, we are building a new workstation with dual E5-2697A CPU’s. These are 16 cores per CPU, but at a higher base and peak clock speed than the 2698 or 2699 processors that have higher core counts.
2. Multiple GPU’s can possibly help, but you have to have some CPU horsepower behind the system to make this work. Suggest looking at Puget Systems website for articles on this subject. I have also heard this same info from people I have talked to locally. We are going with 4 GTX-1080Ti cards in the new system.
3. RAM is hard to find good info on what is beneficial or useable. It is possible to build workstations with 1TB of RAM at this point, but it is very expensve to do so. We went with 512GB of RAM. Agisoft has a calculator available that gives RAM amounts for X amount of images at Y megapixels per image. According to this formula, we should be able to process about 2000 36MP images with 512GB of RAM. We will see how this one works out in practice, and how the Agisoft information translates to the Pix4D world.
4. Last thing is to use the fasted disk drive system you can. We have straight SSD’s in our current system and this is one of the bottlenecks. New system will have a main drive that is NVMe, with an SSD data backup drive. The speed increase on NVMe drives over a regular SSD is a pretty solid jump.
After we get our new system up and running I will post some feedback on these boards about how things are working out.
Tom
@Tom: did you have the chance to test your new system? Would be interested to hear your feedback on it.
Interested as well!!
Thanks Tom!
AHHHH! I was hoping for a follow up…
Hi Jason,
What you have to consider when you choose a computer is:
- CPU. The more cores with the better clock speed, the better. Intel Core i9 processors are the one performing the best but Threadripper CPUs have also shown a great performance and the price is a bit cheaper.
- GPU. The use of a compatible video card helps during step 1 and step 2. GeForce cards from the mid-range (GTX 1060) on up all perform well.
- RAM. That mainly depends on how big your projects are. It is mainly used for step 2 and 3. I would say that at least 32 GB is the recommendation.
- Hard disk: SSD is preferred. The main impact occurs during step 3.
More information on which hardware component is used for each step can be found here.
Regards.
Hi all. Some of these questions are addressed in the Hardware FAQ article linked below. Have a look.