Bomber+ is very slow compared to reg bomber

RENDER TIME:

Bomber: 0.43 secs
Bomber+: 5.74 secs

I have begun turning the anti-aliasing off for Bomber+ projects. Then use FXAA afterwards. Overall a big performance improvement. My 2 cents.

Here's my informal anti-aliasing (5 samples) testing on a Mac with 4 cores:

old bomber - AA off 1.91 sec
old bomber - AA on 2.81 sec

Bomber+ - AA off 2.84 sec
Bomber+ - AA on 5.41 sec

This confirms what I have noticed: AA slows down the Bomber+ component more than the old bomber.

Will test this and do as told above and see what times do I get

Is supposed that the new bomber should be faster and not slower than the old bomber

Quote
Skybase wrote: LexArt, "new" doesn't always imply faster.

I agree with you that "new" is not always better or faster, and it may be worse or have less perfomance, but I thought that as it was told that it would be an improved bomber they could at least make it have the same performance as the old bomber or even make it faster, although really as I do not know what is really inside the new bomber+, I mean how it works and how it has to work with all the new slaves and features, I can not say that it not optimized, and it could be slower because of the new things

Quote
Skybase wrote: In this case The Bomber+ is giving us access to more data to use. I think the performance is an issue, so I hope this can be solved, but I wouldn't keep my hopes up.

I agree and know that the new bomber has more data to use, and only has one unique input and up to 12 different slaves to deal with, but as said above I thought that they could have made a possible optimisation of it and get at least the same speed as the old bomber, although maybe as you say, I also hope and wish that this may be a possible bug or a performance issue, and that maybe this could be solved, but as you, I do not keep my hopes up until we get an answer from the FF team

I have made finally some tests to see what is the real difference (at least on Mac version) and considering also the AA setting with OFF, 5 samples and 25 samples. Used 600x600 pixels and 5000x5000 pixels

I have put the similar values in similar colors so you can compare them faster and easier

Using FF 5.07 Pro Multicore and using the first filter given by uberzev in this thread

Seamless and Non Seamless in reference to checkbox in Settings

Quote
Rick Duim wrote: Then use FXAA afterwards.

Have not tested with this FXAA because I do not know how to use it, if you can explain how to use it, I can test it with it

Thanks

5.0007 (on another newer computer - I'll test on the old one later):

Bomber: 0.99 sec
Bomber:+ 5.39 sec

(I'm not using multipass - it is set to Off in options).

Quote
SpaceRay wrote: and also have seen the very well done and extensive benchmark test done by LexArt, and want to thank for the work done

Thanks for the comment, as I said on the welcome post, I wanted to help if I can, so I have done this chart

Quote
Skybase wrote: It's interesting seeing how seamlessness mildly affects render time.

YES, At first I was not going to put the seamless/non seamless but as it was curious to see the slightly difference, I wanted to put it also

Quote
Skybase wrote: As a temporary solution, disable multi-pass preview and switch it to legacy or off for rendering the final image. It may speed render times up.

Ok, thanks, for the suggestion, I will test it and see if it changes something and could be faster

I have put here the table in RTF format so if any of you want to use and put your own results you can do it as it is editable and then take a screenshot of the text.

Would be good to have more perfomance and speed comparison to see more the differences between computers and OS

Thanks for the help you may give putting your own values

IMPORTANT: Download the file and replace the .ffxml extension with the .rtf extension

_Bomber_VS_Bomber_plus.ffxml

Quote
We are currently rewriting sample cache specifically for Bomber+.

You people are wonderful. Have this emoticon.

Quote
GMM wrote: Yes, this is a known issue. We are currently rewriting sample cache specifically for Bomber+.

Very good to know that this is really a problem and that can be solved in some way. Thanks very much

Quote
Skybase wrote: I have a gut feeling the Mac version is actually faster than the Windows version. Given all your results

Well I have made a new table with the results taken by me and the ones that SpaceRay have put from Windows and I have to say that you are right only in reference to the new Bomber+, and considering that it seems to be a bug or problem, it could be not faster when it may be fixed.

You can see that the comparison of old bomber is faster in Windows in most cases, and is a bad thing for us that have Mac

ALSO consider that the Windows computer from SpaceRay has a Intel i3 and I have an i5.

Results test done with FF 5.07 PRO

iMac Late 2013 - Intel Core i5 3,2 Ghz 24 GB RAM MacOS X Yosemite 10.10.4 with HD disk

Windows 7 SP1 64 bit Home Premium Intel Core i3 550 3,20 Ghz 4GB RAM

Using FF v6.001 here and Bomber Plus is still super-slow compared to the regular Bomber.

Here is the weird part:

On my machine, rendering with...

... Bomber uses 100% of the available cores/threads @ 100% CPU usage.

... Bomber+ uses 50% of the available cores/threads @ 10% CPU usage.

Something appears to be off here, no?

System specs:
Win7 SP1 64bit
Intel Xeon CPU E5-2630 v3 @ 2.40GHz (2 CPUs, 8 cores each)
64GB RAM

[Attached my test filter below]

Bomber_Bomber+ Comparison.ffxml

GMM,

is there a rough ETA for the sample cache rewrite? Will it make it into one of the beta 6 stages?

So the threading/cpu usage issue with Bomber+ is related to the sample cache, or is it actually a different issue?

Here's what I see when I toggle the checkbox... (Left: Bomber - Right: Bomber+)

Anyone?