Puget Systems print logo
https://www.pugetsystems.com
Read this article at https://www.pugetsystems.com/guides/1991
Article Thumbnail

Agisoft Metashape 1.6.5 - AMD Radeon RX 6800 (XT) Performance

Written on December 2, 2020 by William George
Share:

TL;DR: AMD Radeon RX 6800 & 6800 XT Performance in Metashape

Unfortunately, these new AMD Radeon cards are not able to keep up with NVIDIA's comparably priced GeForce RTX 30 Series video cards in Metashape. Better optimization of Metashape for AMD cards, especially in the Build Depth Maps step, might be able to improve their performance in the future - but for now we would recommend sticking with NVIDIA for this application.

Introduction

AMD recently launched a new generation of their Radeon graphics cards, and we have finally gotten our hands on the first two models - the RX 6800 and 6800 XT - courtesy of our friend Brian Stroh at BPS Customs. With this product line, AMD's emphasis seems to primarily be on gaming performance... but since our focus here at Puget Systems is on workstation applications, we are giving them a shot in several professional software packages. This article will look at how they handle Agisoft Metashape, the only photogrammetry program we test which does not explicitly require CUDA support (and thus a NVIDIA graphics card).

Agisoft Metashape Photogrammetry Performance Review for AMD Radeon RX 6800 & 6800 XT

If you want to see the full specifications for the new Radeon RX 6000 Series, we recommend checking out AMD's product page. But at a glance, here are what we consider to be the most important specs:

VRAM Cores Boost Clock Power MSRP
GeForce RTX 3070 8GB 5,888 1.70 GHz 220W $499
Radeon RX 6800 16GB 3,840 2.10 GHz 250W $579
Radeon RX 6800 XT 16GB 4,608 2.25 GHz 300W $649
GeForce RTX 3080 10GB 8,704 1.71 GHz 320W $699

While specs rarely line up with real-world performance, it is nice to see AMD including 16GB of VRAM on their new Radeon 6800 cards. However, based on our past testing, this really shouldn't affect Metashape performance in any meaningful way.

From a pricing standpoint, these cards are right in between the NVIDIA GeForce RTX 3070 and 3080 - although currently, supply is so poor on all of these cards that you would be lucky to find one in-stock for anywhere near the manufacturer's suggested retail price. We typically use the MSRP as a baseline, though, in order to rule out fluctuations due to different brands, sales, and scarcity.

In the end, what this means is that we will be primarily looking to see if these cards are able to out-perform the RTX 3070 and whether or not they can beat the more expensive RTX 3080. That is also why we excluded higher (and lower) models from this comparison; if you want to see a wider range of NVIDIA cards, check our our article from the RTX 30 Series launch.

Metashape Workstations

Puget Systems offers a range of powerful and reliable systems that are tailor-made for your unique workflow.

Configure a System!

Labs Consultation Service

Our Labs team is available to provide in-depth hardware recommendations based on your workflow.

Find Out More!

Test Setup

Here are the specifications of the system we used for our Metashape testing, with components selected to minimize other bottlenecks:

Test Platform
CPU AMD Ryzen 9 5950X
CPU Cooler Noctua NH-U12S
Motherboard Gigabyte X570 AORUS ULTRA
RAM 4x DDR4-3200 16GB (64GB total)
Video Card AMD Radeon RX 6800 XT 16GB
AMD Radeon RX 6800 16GB

NVIDIA GeForce RTX 3080 10GB
NVIDIA GeForce RTX 3070 8GB
Hard Drive Samsung 960 Pro 1TB
Software Windows 10 Pro 64-bit
Metashape Professional 1.6.5

To test each video card, we used our in-house Metashape benchmark - both the standard and extended versions. Each was run twice per GPU, and the best results were included in the charts below. Here is some basic info about the image sets in these projects:

  • Rock Model - 36 photos at 20 megapixels each
  • School Map - 51 photos at 18 megapixels each
  • School Model - 439 photos at 18 megapixels each
  • Park Map - 792 photos at 18 megapixels each

We make these benchmarks publicly available under the Creative Commons BY-ND license, so if you use Metashape Professional (or get the 30-day trial) you can download them from our website and compare your system's performance with the results shown here.

Benchmark Results

Here are charts showing the performance of the new AMD Radeon cards (in red) compared to NVIDIA's competing GeForce models (green) and no active GPU (grey - why we included that will become apparent in the analysis):

Performance Analysis

The overall spread in our smaller Metashape image sets isn't huge, with no more than a few percent variance there between the four cards we tested, but if you are working with larger projects (hundreds of images or more) then the differences between GPUs become very pronounced. And the delta isn't so much between tiers of cards within the same product family, though that is measurable, but rather between AMD's Radeon line and NVIDIA's GeForce. In both of our larger image sets, the AMD cards were 60-70% slower than their NVIDIA counterparts - which is a massive drop in performance, unfortunately.

But where is such a big difference coming from? To answer that, we have to look at the how these cards fare in each processing step:

Metashape 1.6.5 Radeon RX 6800 & 6800XT vs GeForce RTX 3070 & 3080 Performance Table

Table showing processing time per step (lower is better)

Pouring over the table above provides a few interesting insights:

  1. The AMD Radeon cards are as fast or faster than NVIDIA's GeForce when it comes to Align Photos, Build Dense Cloud, Build Mesh, and Build Texture steps. They are just a hair behind in Build Depth Maps, but not enough to be a big concern.
  2. In Map projects, both the Build DEM and Build Orthomosaic steps appear to be CPU only (since they perform the same across all four video cards as well when no video card is selected in Preferences). Build Dense Cloud may fit this category as well, but it does seem to vary a little between the AMD and NVIDIA cards.
  3. Decimate Mesh is, oddly, slower with AMD cards than with no GPU selected at all (and faster on NVIDIA cards). It is an extremely short part of the overall photogrammetry process, however, so it isn't worth worrying about.
  4. The big performance issue with AMD cards, then, comes from a single step: Build Depth Maps. Having one of these Radeon cards does reduce processing time by 5-10% in this step on our large image sets, compared to no GPU acceleration, but having a NVIDIA GeForce card instead drops that time by ~80%! Since this is also one of the longest steps in Metashape's workflow, the massive lead NVIDIA cards have here carries them to a huge victory in overall processing time as well.

Here is a gallery of charts showing the individual step processing times in the Park Map project, to help visualize these differences:

Are AMD's Radeon RX 6800 and 6800 XT Good for Agisoft Metashape?

Unfortunately, no - these new AMD Radeon cards are not able to keep up with NVIDIA's comparably priced GeForce RTX 30 Series video cards in Metashape. Better optimization of Metashape for AMD cards, especially in the Build Depth Maps step, might be able to improve their performance in the future - but for now we would recommend sticking with NVIDIA for this application.

As always, please keep in mind that these results are strictly for photogrammetry in Agisoft Metashape. If you have performance concerns for other software in your workflow, we highly recommend checking out our Hardware Articles (you can filter by "Video Card") for the latest information on how a wide range of programs perform with various GPUs, CPUs, and other hardware.

Photogrammetry Workstations

Puget Systems offers a range of poweful and reliable systems that are tailor-made for your unique workflow.

Configure a System!

Labs Consultation Service

Our Labs team is available to provide in-depth hardware recommendations based on your workflow.

Find Out More!
Tags: Hardware Acceleration, NVIDIA, RTX 3080, GPU Acceleration, GPU, Video Card, Metashape, Agisoft, photogrammetry, RTX 3070, AMD, Radeon, Radeon RX 6800, Radeon RX 6800 XT, GeForce
adrian

Something is very strange with those numbers William.
Metashape uses GPU for alignment and depth map generation.
In the alignment step (especially for the larger models) the 6800XT is on par with the 3080.
But for depth map generation, it looks like the GPU is not even used, since the computation time is only marginally faster then NO GPU...

Posted on 2020-12-03 22:08:47

It must be using the AMD GPUs somewhat for building the depth maps, since they are 5-10% faster than no GPU (the same Radeon card installed, physically, but unchecked in the Preferences within Metashape) during that step on our large projects. But the NVIDIA cards are *much* faster, and since that is one of the longest steps on those big image sets it puts the NVIDIA cards in a huge lead.

I didn't go into it in the article's text, since small (<100) image sets are not usually as big of a deal for most folks, but oddly the AMD cards were *faster* in Build Depth Maps with our two little projects. If I had a wider range of project sizes, it might be interesting to look at where the flip happens: at what image count do the AMD cards go from being somewhat faster to a LOT slower.

Posted on 2020-12-03 22:26:08
adrian

Having larger image sets would be hugely beneficial for photogrammetry benchmarking in general. Professionals that use these rigs for photogrammetry usually work with thousands or tens of thousands of images.
I can provide you with significantly larger image datasets. How can I get a hold of you to talk off-line?

Posted on 2020-12-03 22:41:51

Larger image sets are not so much my concern - these four projects together already take multiple hours to run, and we do them twice on each combination of hardware we want to test... so it adds up, and making that even longer wouldn't really be viable most of the time :)

I have also done some spot checking in the past with large (1000-2000 image) projects, and they have generally followed the same performance patterns that we see on the 400-800 image sets.

In this specific situation, the question I have is more about the mid-size projects: where do the AMD cards go from being faster to slower? At 100, 200, or 300 images? Certainly they are faster with ~50 images, but far slower by the time we get to 400+ images. Most of our customers are working in the hundreds to thousands of images per project, so that is why I tend to focus on the performance of our larger tests as the best predictors of end-user results.

*EDIT* - It just occurred to me that I could probably chop the Park Map project set we use up into smaller chunks, if I am careful about selecting photos, and make projects from it at multiple smaller image counts. I may look at doing that in the future, though I have a lot of other testing to do this month (even just within Metashape: SMT on vs off on AMD Ryzen CPUs and GeForce RTX 30 Series Multi-GPU Scaling are both in the works).

Posted on 2020-12-03 22:53:44
Michael Freeman

William, your work with testing these different set ups is amazing. Thank you! Your various CPU and GPU tests on Metashape have litterally provided me with a raise at my work. I look forward to your next multi GPU tests.

One day I want to buy a new set up with the recommendations you provide and start my own small business.
Thank you.

Posted on 2020-12-04 14:22:57

Thank you for those kind words, and congrats on your raise - that is awesome :)

Posted on 2020-12-04 17:44:26
adrian

Also, perhaps you could try the 1.7 pre-release version of Metashape and see if it performs any differently.

https://www.agisoft.com/for...

Posted on 2020-12-03 22:53:55

I've had enough stability issues with Metashape full releases, I don't think I feel up to trying a pre-release version that is described as potentially unstable :-)

I didn't mention it in this article, but I actually had to slightly modify the way I run our benchmarks to get them to behave on AMD's cards. Normally our two big image sets run back to back via scripting, and that works just fine on Nvidia cards, but for some reason these new Radeon models would cause Metashape to hang early on in the second project. If I ran them separately they did just fine, so I really don't know what's going on there but I definitely don't want to introduce more potential instability :-)

Posted on 2020-12-03 23:16:25
adrian

I get it.
Very strange this behaviour...

Posted on 2020-12-03 23:35:01