Puget Systems print logo

https://www.pugetsystems.com

Read this article at https://www.pugetsystems.com/guides/1625
Article Thumbnail

Agisoft Metashape CPU Performance: Intel Core X-10000 vs AMD Threadripper 3rd Gen

Written on November 25, 2019 by William George
Share:

Introduction

For years, choosing between Intel and AMD processors has been relatively straight forward. Intel held a strong lead in per-core performance, and combined with a moderate number of cores that gave their processors an edge in lightly to moderately threaded applications. AMD excelled in pure core count, making them great for heavily threaded workloads, but with lower per-core performance and architectural design limitations that led to poor results in some situations. This changed quite a bit on the consumer side when AMD launched their 3rd gen Ryzen CPUs, but still held true when it came to the HEDT (High End Desktop) processors from both brands.

Now, both Intel and AMD are launching new HEDT processors that have the potential to shake things up dramatically!

On Intel's side, the new Core X-10000 series processors are only slightly faster than the previous generation - but Intel has cut their prices nearly in half. For example, while the MSRP of the Core i9 9980XE was $1,979 the new Core i9 10980XE is just $979.

On AMD's side, the new Threadripper 3rd Gen CPUs are actually a little more expensive than earlier models (by $100-200) but their raw performance is expected to be quite a bit better. The gains are not in core count, though, but rather in clock speed, performance per core, and architectural improvements which eliminate some bottlenecks that the first and second generation Threadripper CPUs had.

Intel Core X-1000 series vs AMD Threadripper 3rd Gen in Metashape

For this article we are taking a look at how these new processors compare against each other, the previous generation they are replacing, and their mainstream desktop counterparts. The focus is on photogrammetry processing in Agisoft Metashape, using a benchmark tool we developed here at Puget Systems. More info about this utility is available in the Benchmark Details.

If you would prefer to skip over our test setup and benchmark sections, feel free to jump straight to the Conclusion.

Looking for a Metashape Workstation?

Puget Systems offers a range of workstations that are tailor-made for your unique workflow. Our goal is to provide the most effective and reliable system possible so you can concentrate on your work and not worry about your computer.

Configure a System!

Test Hardware

We only have one license for Metashape Pro at this time, the full benchmark run takes several hours, and I prefer to run it more than once per system to filter out any abnormal results. As such, I was not able to test as many CPUs as I would have liked.

Here are the specifications of the different platforms that I was able to use for my testing:

AMD Ryzen Test Platform
CPU AMD Ryzen 9 3950X ($749)
AMD Ryzen 9 3900X ($499)
AMD Ryzen 7 3800X ($399)
CPU Cooler Noctua NH-U12S
Motherboard Gigabyte X570 AORUS ULTRA
RAM 4x DDR4-2933 16GB (64GB total)
Intel 9th Gen Test Platform
CPU Intel Core i9 9900K ($499)
CPU Cooler Noctua NH-U12S
Motherboard Gigabyte Z390 Designare
RAM 4x DDR4-2666 16GB (64GB total)
AMD Threadripper 2nd Gen Test Platform
CPU AMD TR 2990WX - DLM on ($1,799)
CPU Cooler Corsair Hydro Series H80i v2
Motherboard Gigabyte X399 AORUS Xtreme
RAM 4x DDR4-2666 16GB (64GB total)
Intel X-9000 Series Test Platform
CPU Intel Core i9 9980XE ($1,979)
CPU Cooler Noctua NH-U12DX i4
Motherboard Gigabyte X299 Designare EX
RAM 4x DDR4-2666 16GB (64GB total)
AMD Threadripper 3rd Gen Test Platform
CPU AMD TR 3970X ($1,999)
AMD TR 3960X ($1,399)
CPU Cooler Corsair Hydro Series H80i v2
Motherboard Gigabyte TRX40 AORUS Xtreme
RAM 4x DDR4-2933 16GB (64GB total)
Intel X-10000 Series Test Platform
CPU Intel Core i9 10980XE ($979)
CPU Cooler Noctua NH-U12DX i4
Motherboard Gigabyte X299 Designare EX
RAM 4x DDR4-2933 16GB (64GB total)
Shared Hardware/Software
Video Card NVIDIA GeForce RTX 2080 Ti 11GB
Hard Drive Samsung 960 Pro 1TB
Software Windows 10 Pro 64-bit (version 1903)
Agisoft Metashape Pro 1.5.5
Puget Systems Metashape Benchmark

A quick note about memory specifications: the latest CPUs from AMD and Intel support varying speeds of RAM depending on how many modules are installed. In past articles we've sometimes tested at multiple speeds, but to help keep things more uniform going forward we are standardizing on the speed that each processor / platform supports when all memory slots are used (even if we aren't filling them completely for our testing). Along with using the same total RAM capacity, that should help to keep results fair and consistent.

Benchmark Details

For benchmarking photogrammetry applications we now have four image sets that we own the rights to, covering both smaller and larger size Model and Map projects. The smaller image sets are included in our public Metashape benchmark, which you can download and run if you want to compare your system's performance to what we measured in our testing.

  • 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

Benchmark Results

There is a lot going on in the chart below, so before getting into it here is a reference key for the color scheme:

  • Light blue = Intel consumer CPUs (9th Gen Core series)
  • Dark blue = Intel HEDT CPUs (X-9000 series)
  • Dark blue with glow - NEW Intel HEDT CPUs (X-10000 series)
  • Light red = AMD consumer CPUs (Ryzen 3rd Gen)
  • Dark red = AMD HEDT CPUs (Threadripper 2nd Gen)
  • Dark red with glow - NEW AMD HEDT CPUs (Threadripper 3rd Gen)

Analysis

As expected, Intel's new Core X 10980XE is on-par with the previous 9980XE - but at half the price. That would be good news if either of those processors performed well in Metashape, but there is something about the way this software works which causes the Core X line to perform poorly.

Whatever that is, it also appears to affect the Threadripper line. AMD's new 3rd Gen chips are far better than their predecessors, shaving 25-35% off the processing time compared to the 2990WX, but they still lag behind AMD's more mainstream Ryzen CPUs.

I have seen performance like this before, which is why our standard Metashape recommended system has been a mainstream Core i9 for a long time now - and will remain on that CPU for the time being. I had assumed it was a problem with core count, but now I am re-thinking that because the 16-core Ryzen 9 3950X does not seem to have the same performance problem that Intel's 18-core chips exhibit. I plan to investigate this in more detail, when I have time, looking at a wider range of processors, testing the impact of SMT, etc.

Is Intel X Series or AMD Threadripper 3rd Gen CPUs better for Metashape?

Neither of these processor lines are a good fit for Metashape at this time. For most users either Intel's Core i9 9900K or one of AMD's Ryzen 9 processors will be a much better choice, as they provide the best performance in Metashape as of version 1.5.5.

Looking for a Photogrammetry Workstation?

Puget Systems offers a range of workstations that are tailor-made for your unique workflow. Our goal is to provide the most effective and reliable system possible so you can concentrate on your work and not worry about your computer.

Configure a System!

Tags: Intel 9th Gen, Intel X-series, Intel vs AMD, AMD Ryzen 3rd Gen, AMD Threadripper 2nd Gen, Intel X-10000, AMD Threadripper 3rd Gen, Threadripper, CPU, Metashape, photogrammetry
Nodjo

Could you please try using only 8 or 16 threads, making sure you use only one thread per core (ie on a 16C/32T chip, you set the CPU affinity to cores 0,2,4,8,...,30 for metashape) on a TR 3 to see how it behave?

Posted on 2019-11-27 09:53:26

I plan to dig into Metashape performance quirks on these high-end CPUs soon, and will certainly include things like disabling SMT (effectively the same as setting CPU affinity to use only one thread per physical core). I'm also in touch with folks in Agisoft's support department, and will be working with them to try and figure out what is going on. It will be slow going, though, because I can only run the test on one system / CPU at a time (as I only have one license of Metashape Pro).

Posted on 2019-11-27 17:11:43
Nodjo

Great, thanks. I'm definitely looking forward to reading about your findings. I'll need to build a machine dedicated to Metashape in a couple of months and I want to get the most out of it. I suspect that the best deal is to run several instances on Metashape on the same machine. This is suggested by your test results and also discussed on the Agisoft forums with promising results but I've never really seen a serious benchmark and can't run it myself as I currently lack the required hardware, so that theory still sounds like a risky bet.

Posted on 2019-11-27 20:10:20
GNK

Thanks for the articles and testing!

Disabling hyper-threading on i7 7820x (8 high speed cores vs 16 more variable speed virtual cores ) gets slightly improved performance across the downloaded tests.

I suspect something about L1/2/3 cache structure and usage ... 7820x might be extreme case for Skylake(v1) due to gimped L3 cache size ... IDK > contention for cache data causes misses and re-writes when core:cache ratio gets to a critical level?

Disabling CPU-cache functions (hardware pre-fetch, adjacent line prefetch, virtualization etc) also seems to minimally improve the HT-ON performance, but not as much as compared to HT=OFF > performance with HT=OFF + BIOS cpu-cache-fetch functions =ON seems to be best on these tests for this CPU ...

Running Metashape only on real cores reduces 'total cpu usage' to c. 50% and dramatically improves heat > this enables high core clocks to be maintained > also worth noting that CPU package-ring is always very busy and toasty during the tests, even when actual cores are less so, again hinting at role of cache and other non-core elements on the die...

Because of this I dont think multiple instances of Metashape on a single/virtual machines could work well in most cases, except perhaps on very high end gear - and as I understand it one would need multiple Metashape licenses for that.

[note: HT is ON in BIOS, but software is controlled/limited to non-hyperthreaded cores (i use the excellent ProcessLasso). HT=OFF globally in BIOS probably compromises other performance overall, but maybe not so much with this CPU? - not tested]

Posted on 2019-11-28 21:44:50
Aroldo Yoc

Thanks for the articles. because they no longer used 128gb of RAM in the INTEL + Gigabyte Z390 Designare configuration

Posted on 2019-12-01 21:09:33

Thanks :) We specifically went with 64GB across all test platforms in this series because 16GB was the largest we could get 2933MHz memory modules in, and we wanted to run that speed for both the new AMD and Intel Core X processors (as higher supported memory speed is one of the advantages of those new chips over the previous generation).

Posted on 2019-12-03 17:20:45