Benchmarks

[Nedo]

I just did the new Benchmark, can we post it somewhere to compare the hardware ect. ?
This is my Benchmark:

[Roberto]

Thank you for sharing the results! I think it's good if we can use this topic and the Academy Facebook group to share the results. The benchmark is still experimental and the scores and the benchmark itself will be certainly adjusted in the future. We plan to unify the data collection with future versions of Zephyr and make it publicly available and sorted in a clever way.

[Roberto]

And here's my benchmark results!
Laptop: https://benchmark.3dflow.net/2064696...11dfd62c01770a
Workstation #1: https://benchmark.3dflow.net/62bad52...428428957382d5
Workstation #2: https://benchmark.3dflow.net/0bcb9ef...dfe64ad649fe00

[Filippo]

Mine!

https://benchmark.3dflow.net/d6b73cdae14fe827f77a8f76c11604bb

Not too shabby

[Nedo]

so far, looks like the 1080(ti) is the way to go!

[AvionAvionics]

My workstation is worse than Roberto laptop, excuse me while I go sulk in a corner

[Andrea Alessi]

Don't worry, this is my laptop

https://benchmark.3dflow.net/00303e2d7521f62fd764f4d1d7769c4f

[Roberto]

Well, to be honest my "laptop" weighs about 4kg

[bernamorillo]

[Roberto]

Hi Bernamorillo,

Yes, the score is related to running times. The benchmark is done with a small and very well-textured synthetic example, running at high details settings. Of course, this does not cover all the possible cases, and results may vary with different datasets. For instance, if you are planning to use a lot of images, you need more RAM than the amount required to run the benchmark.
As for the hardware suggestions, it depends on your budget. In general, the cost never scales linearly with the running times. For example, a GTX 1060 is half (maybe less) the price of a GTX 1080. The GTX 1080 is of course faster, but not twice as fast. A workstation has its advantages over a laptop. For example, it's expandable in the future, you have less overheating problems, and you can install two or three GPUs instead of a single one.

[bernamorillo]

Thank you for the answer.

My point is that:

It is better 2 1060 GPU than 1 1080 GPU and it is cheaper.

When we are working with several scans the best idea is to invest in 3 x 1050 GPU better than 1 x 1080 GPU

is it correct?

Thanks

[Andrea Alessi]

Yes, usually that would be a better choice. The only big difference would be the amount of video ram (i.e. the 1080 is available in 8GB, while the 1050 is not). In some *rare* cases, you might not have enough ram on the 1050, for example when doing the photoconsistency step on very high resolution images (say, 50MPixels).

If you're working with "standard" images, i would go for 3x1050 rather than one 1080

[manocheese]

I'm not scoring very highly on the benchmark and in normal use, it doesn't seem to be making great use of the GPUs, I'm not sure where the bottleneck is. During processing, my CPUs are maxed out most of the time, the GPUs spike occasionally but don't go over 30% usage each most of the time. The cache is on an nvme drive.

https://benchmark.3dflow.net/a09a5719cf5566280a4d2f92a023ee82

[Andrea Alessi]

Hi manocheese,

could you please run the benchmark again and send us the log (or if you can find the previously saved log from the previously run benchmak, that works too) ? this way we can check each phase what is happening.

http://www.3dflow.net/technology/documents/3df-zephyr-tutorials/how-to-locate-3df-zephyr-log-files/

(Also keep in mind that the benchmark uses a very small dataset - your rig has a lot of ram so it's going to outperform most machines when dealing with huge datasets, but that is not going to be linearly shown in the benchmark scores - that said however, i agree that your score looks odd on a first sight)