I’ve been browsing the forum for answers to this question but it seems like not too many people have asked this question.
If you look at the latest benchmarks of CPUs (https://www.cpubenchmark.net/high_end_cpus.html) it’s obvious that most XEON E5’s are outperforming the highest Core i7’s by quite a bit. Does anyone have a definitive answer on how this translates to photogrammetry processing times?
So let’s assume we have a fantastic beast of a PC set up, including fast cache drives, 2 or 3 Geforce GTX 1080 Ti’s, tons or RAM and all that jazz:
To what degree does the processor architecture (i7 vs Xeon E5) speed up or slow down the processing times? During my visit to GTC in San Jose I discovered that all the super-computers designed for GPU processing are using XEON’s and QUADRO K6000 (up to 16!!). However, as nice as these machines looked I wonder if the extra expense also translates to a significant performance boost.
Let’s say we pick an architecture, how about more (but slower) cores versus less (but faster) cores? A 10-core XEON at 2.2Ghz scores way higher than a 4-core i7 at 3.6Ghz but do they process photogrammetry faster as well?
I’ve seen some discussions regarding the amount and type of GPU’s as well. Nobody has the budget for eight QUADRO K6000’s obviously, but what about let’s say 3x GTX 1080 Ti’s VS 2x Titan Xp ? Some people would argue that having (and linking) more GPU’s is better but I’ve read threads where it looks as though the software doesn’t really use them to their full extent.
Basically I want to figure out whether throwing more money at a workstation is really worth it because of software limitations.
Thanks!
first, each license of rc only supports 32 threads, and 3 gpu’s.
if you want to use more. then you need more licenses.
different parts of the processing use either gpu or cpu. so depending on which part needs more depends on scene size/detail.
for lots of photos at high res? having 2+ gpu’s help. but returns get less and less as you add them. I’d say less faster ones are better. vs more slower ones. but gpu ram isn’t a problem never uses more than 2gb.
mesh generation uses lots of cores well. (but its the only part that dose). this part is also suited to running on mulitple pc’s.
large area models, will mean you need more cpu.
but unless your going to spend on multiple licenses limited to only the lower core xeons if you want a dual setup. and even some single chip xeons will need hyperthreading disabled.
having fast single thread will help all through out the processing.
for the stuff I do, i find the cpu to be the slower part. (running 8 core, 2x1080).
Disabling hyper-threading, would that also improve performance on a Core i7?
Currently I’m running a Core i7 at 3.3Ghz, 64GB ram and a Titan X Pascal. Processing 3x36MP images + 45x18MP photographs takes about 2-3 minutes of initial alignment and 30-40 minutes of high resolution processing which is the part I’m trying to cut down. Texture generation is an additional 10 minutes which is OK, but could be faster.
I doubt it would speed up disable hyper threading. I leave it on.
its more just a license thing, if you were getting more than 16 real cpu core’s. at that point any more than that with hyperthreading on and core’s will get wasted.
so if you had just one 22 core xeon, you could have 44 threads with only 32 doing something or you could have 22 real cores working on it.
this would need some testing to see which was faster.
have the console open while processing. and see what takes longer, the depth map generation or the creating mesh part. depth maps uses gpu, mesh uses cpu.
Hi Jan
if you want to go with XEONs, then take the W series, which are optimized for workstation load. The others are Server CPUs designed to serve as many as possible cores/VMs at lower clock speed. They are not suitable for workstation workload. The present CPUs are not designed to withstand too much core communication, so you get even slower computation if you use a server CPU, compared to a workstation ( if you take into account core speed and volume ).
But in 1-2 months AMD and Intel will offer new generation of CPUs, so I highly recommend you to wait for the official release, as the speedup and price/performance can be significant.
But in 1-2 months AMD and Intel will offer new generations CPUs, so highly recommend to wait for official release as the speedup and price/performance can be significant.
Thanks for letting me know! I’d only consider a XEON if it would dramatically reduce processing times, which it doesn’t look like it would. I’ll look into the AMD thing. It seems like they’re doing some very good things over there, although I do read that even though their processors are taking some leaps forward, their chipsets aren’t yet. This would mean that running multiple GPU’s and SSD drives on PCIe wouldn’t be very efficient.