Hoping to get some guidance for trying to build a machine for some of our users. I see a lot of i9 7900x benchmarks but not as many Ryzen… with that being said is there anything we should worry about with Ryzen? We are looking to build a 2950X machine with a GTX-1060 and 32gb of RAM with a 500GB NVMe drive. What are some things I need to consider when building their machines, I seem to have gotten conflicting information for different sources and just need to straighten out how much to budget for projects of about 1500-2000 images.
What type of processing will you be doing? 3D maps, 3D models, Ag multispectral?
From my own test and what I have gathered you need everything in your computer as fast as you can afford. I would recommend getting a 1080Ti and at lease 64GB Ram if you are going to build a 2950x machine. Processing speed does make a difference too but it all depends on the size of the projects that you are processing.
I can run a set of your images and give you a benchmark of what my machine will do if you would like and maybe someone with a Ryzen could do the same.
From what I have gathered from your data that you shared is as follows.
Processing step 1. - Processor Speed is king. On the 3D eagle model benefits from overall processor clock speed. My machine did it in 99 seconds and the fastest one on your chart did it in 106 which is the same processor but not overclocked. For the 3D building Map the top 25 were all the same basic processor. Again my machine did it in 115.5 seconds and the fastest one on your chart did it in 130 seconds.
Processing step 2. - Cores and processor speed together are king.The GPU on all of the top 20 were a high end gpu. 1080Ti, Titan V, Quadro
Processing step 3. - Higher core count looks to be the fastest. More of a mix of GPUs in step three so I would say they do not matter as much in this step but a NVIDIA GPU does do better then a machine without one.
There is a nice 1950x Threadripper machine in the spreadsheet but it never breaks the top 20 results.
Brian thanks for the info. Personally, I’m not doing any of the processing. The team responsible has asked me to source them a desktop that will be their workhorse (I’m just the IT guy).
So for clarification, what is the GPU doing? It seems a lot of these builds have a higher end card, is it actually having a major impact going from 1060 to a 1080ti?
Also with Intel seeming to score a little better with these, is it safe to say this is more worries about single core performance?
This is a response from one of the members of the team doing these maps
I use the Pix4Dmapper module and I have been looking into using the Pix4Dmodel too. I use the colored point cloud, classified point cloud, orthomosaic, DTM, DSM, DEM, 3D textured mesh, index map, thermal map, contours, reflectance maps, stockpile volume calculations, measuring tools. I am sure there are more parts of the software I use, but those are what come to mind.
Thanks for reporting back. Interesting results. This supports the advice we give that the best-generalized CPU for use in Pix4D is one that balances both clock speed and core count.
If anyone else has a look and finds any actionable insights make sure to share!
Wow. Judging from the data in the spreadsheet this looks like a significant improvement over the Threadripper 19 lineup. Did you use a GPU for processing?
Threadripper 1950x versus 2950x improvement is only a die shrink and with a slightly better XFR. The two, especially with high speed memory, are fairly close to each other since the X399 platform is the basis of both. The improvements in Pix4D and other HEDT focused programs have come from better Windows drivers and BIOS changes over the last year on the platform. I’ve been saying this for a year, Threadripper is viable with Pix4D dollar on dollar to what Intel is offering. Core for core is a different story.
The keys to building a good Threadripper system are +3000 mhz memory, Raid 0 NVMe drives or ultra performance PCI-E drives, and up to date BIOS and Windows 10. I also would not recommend the CPUs above the 2950x due to the way 4 die Threadripper chips handle the memory through a separate controller on two of the dies.
I’ll have to remember to run my 1950x and 2080 when I get back to post updated numbers.
@ Sterling. That would be really interesting to have some data to support. Hopefully, soon we will have some larger datasets to benchmark with so that we can get a better idea under workload conditions.
Hey, Anand thanks for sharing! These are interesting results as the Eagle model seems to be performing well but the building map is over what I would expect. I would ensure that you have the most up to date GPU drivers and maybe try with only a single GPU to see if you get any performance gain. Also, it would be interesting to test without GPU entirely to see what performance gain is achieved with a single RTX card.
I ran the data from ary sanjaya’s lembongan.zip and got 280 seconds per the Quality Report on Pix4D v4.3.33 (3D map template)
CPU - Intel Xeon W-3175X 28c/56t @ stock (3.1GHz)
CPU Cooler - Custom
GPU - GeForce RTX 2080 Ti Founder’s Edition x2
RAM Amount - 192GB in (12) 16GB chips
RAM Speed - DDR4 DRAM 2933 C17 (XMP)
Hard Drive or SSD or M.2 - Samsung 970 Pro
Motherboard Model - ASUS Rog Dominus Extreme
Operating system - Windows 10 Pro
This was run with the CPU on XMP but stock speed (set to auto in bios). I will try again on 4.5GHz. I allocated 32GB of ram and 50 threads for this render
Hey Benjamin. In order to be able to compare your results with the others, you should test with the same dataset. Check the Puget site for links to download the imagery. This will give you an apples to apples comparison.
Just got my new Pix4D/Laserscanning rig up and running:
Threadripper 3970X 32c/64t
Asus ROG Zenith II
128Gb DDR4 3000Mhz
MSI RTX 2080Ti Sea Hawk
2 x EVO 970 500Gb m.2 Raid 0
2 x Sandisk Ultra 3D 500Gb SSD Raid 0
1 x WD Black m.2 1Tb
1 x Samsung PM981 SSD 512 Gt M.2 OS disk
3 x WD Red 4Tb Raid 0
These cookies are necessary for the website to function and cannot be switched off in our systems.
They are usually only set in response to actions made by you which amount to a request for services, such as setting your privacy preferences,
logging in, or filling in forms. These cookies do not store any personally identifiable information.
These cookies allow us to count visits and traffic sources so we can measure and improve the performance of our site.
They help us to know which pages are the most and least popular and see how visitors move around the site.
All information these cookies collect is aggregated and therefore anonymous.
If you do not allow these cookies we will not know when you have visited our site, and will not be able to monitor its performance.
These cookies may be set through our site by our advertising partner (Google).
They may be used by Google to build a profile of your interests and show you relevant adverts on other sites.
They do not directly store personal information but are based on uniquely identifying your browser and internet device.
If you do not allow these cookies, you will experience less targeted advertising.