Table of Contents
PhotoScan is a photogrammetry program: an application that takes a set of images and combines them to create a 3D model or map. This article is part of a series looking at how different aspects of computer hardware affect PhotoScan performance. For more information on this software, or to see the other entries, check out our introductory article.
Intel has released a new line of multi-CPU processors this year, called Xeon Scalable. In our testing from 2015, we found that dual Xeon systems made sense for PhotoScan in some situations – but now that single-CPU workstations can fit more cores (up to 18 on Intel's Core X series) and GPU scaling doesn't seem to benefit from more than 3 video cards, we thought it was time to re-test dual Xeons to see if they are still worthwhile.
Methodology and Test Hardware
We already know that CPU clock speed is a big factor in PhotoScan performance, so we narrowed the scope of this article down to two Xeon Scalable models: the 4-core Gold 5122 and the 18-core Gold 6154. These both have high clock speeds, for Xeons, and will give us an idea of how performance looks with both a few cores and with many. Intel Core i7 and i9 chips are included for comparison.
Because the GPU has an impact on processing times in some workflow steps, we used a single GTX 1080 Ti for the CPU comparisons. These Xeons also have quite a few PCI-Express lanes, so a dual Xeon Scalable platform is ideal for multiple GPU testing as well. We've already seen what this looks like on other CPUs, with 3 video cards appearing to be the most that could be effectively used, so testing here will either confirm our past findings or show that we have more work to do on that front.
We used relatively small photo sets to avoid the amount of memory that each CPU supports becoming an issue. Please note that RAM capacity is absolutely something to consider if you work with large image sets, though, and we will discuss it more in the analysis and conclusion sections. As with the other recent articles we have published, all tests were conducted with "High" quality settings.
If you would like more details about the full hardware configurations we tested on, and the image sets we used within PhotoScan, simply click here to expand the following section.
|Motherboard:||Gigabyte Z370 AORUS 5||Gigabyte X299 Designare||ASUS WS C621E SAGE|
|CPU:||Intel Core i7 8700K 3.7GHz
(4.7GHz Turbo) 6 Core
|Intel Core i9 7900X 3.3GHz
(4.3/4.5GHz Turbo) 10 Core
Intel Core i9 7940X 3.1GHz
(4.3/4.4GHz Turbo) 14 Core
Intel Core i9 7980XE 2.6GHz
(4.2/4.4GHz Turbo) 18 Core
|2x Intel Xeon Gold 5122 3.6GHz (3.7GHz Turbo) 4 Core
2x Intel Xeon Gold 6154 3.0GHz (3.7GHz Turbo) 18 Core
|RAM:||4x Crucial DDR4-2666 16GB (64GB total)||12x Crucial DDR4-2666 ECC Registered 32GB (384GB total)|
|GPU:||NVIDIA GeForce GTX 1080 Ti 11GB|
|Storage Drive:||Samsung 960 Pro M.2 PCI-E x4 NVMe SSD|
|OS:||Windows 10 Pro 64-bit|
|Software:||Agisoft PhotoScan 1.4.1|
|Image Sets (from PhotoScan website)|
|Monument (32 photos)||Building (50 photos)|
Here are results for the Building image set:
And here are results for the Monument image set:
GPU Scaling Results
Looking at GPU scaling, here are the results showing from 1 to 4 GeForce GTX 1080 Ti cards with dual Xeon Gold 5122 processors:
And the same set of GPUs with a pair of Xeon Gold 6154 processors:
Looking at the CPU performance charts, we can see how the Xeon Scalable processors fared:
- Xeon 5122s are not ideal for PhotoScan. They don't have enough cores to do well in the heavily threaded Build Dense Cloud step, nor enough clock speed to make up lost time during the Build Mesh step.
- Xeon 6154s did pretty well, with good speeds across all the test steps. If you add up the times, as we have done in the summary chart below, they came in very close to the Core X series… technically even a little bit ahead. However, the pair together cost several times more than even the most expensive Core X processor.
Moving on to GPU scaling, we see similar results to what Threadripper showed in our previous article:
- A good improvement going from 1 to 2 video cards
- A measurable improvement going from 2 to 3, though not nearly as pronounced
- Almost no difference when adding a 4th GPU; technically it is a 1-3% reduction in times, but that could be within the margin of error
Even if that last bump from a 4th GPU is not an error, it is so small that it cannot justify the cost of another GPU or the various other things that come with that (larger chassis, higher wattage power supply, etc). Even the jump from 2 to 3 video cards is somewhat questionable, but in situations where the Build Dense Cloud step is a more substantial part of the overall calculations – for example, when using 'Ultra High' quality settings on that step – it could be worthwhile.
Another factor that isn't shown in the graphs above is RAM usage. With the relatively small image sets we used that isn't an issue, but with more images and higher quality settings it could be important. That is one edge which the Xeon Scalable processors have since they can support hundreds of GB of memory each (depending on how many slots the motherboard has)… but there are less expensive options which can fit lots of RAM as well. We haven't tested them with PhotoScan yet, but Intel's single-socket Xeon W chips can accommodate up to 512GB of memory and should perform on par with the Core X since they are based on the same architecture.
Given the results above, it looks like the Xeon Scalable processors are not worthwhile for PhotoScan. Their CPU performance is so close to the Core X series that those less expensive processors are a much better choice if you want to run three GPUs (four isn't worth it either) and need up to 128GB of memory. If you need more RAM, the Xeon W processors should perform on par with the Core X and will handle up to 512GB of RAM. On the flip side, for smaller projects where 64GB is sufficient, Intel's Core i7 8700K is a great value.
Puget Systems offers a range of powerful and reliable systems that are tailor-made for your unique workflow.