Four different results:{honeybee legacy-CPU} / {honeybee legacy-accelerad GPU} / {honeybee[+]-CPU} / {honeybee[+]-accelerad GPU}

#1

Hi,
I have a comparison of the four pairs using one same simple scene to run daylight simulation, the results are as belows:


(new user can post one picture)

#2

hi @miraclema,
Can you also share for each type, how long the simulation took ?

Best
Peter

#3

According to the accelerad manual, adding -g 0 argument will let the accelerad use CPU instead of GPU, I’ve also test the original rtrace.exe, the results are the same as that using the -g 0 argument. So the way I switch between GPU and CPU may be right.

First, I can’t tell why there are stright lines in the GPU simulation, I think it is not direct value because I can’t get rid of it using the cumValue component’s mode_2.

Second, in the CPU ones why the illuminated area are all on the left of a stright line, lets compare the different time 6-7-9-11-13-15-17:


the upper ones are CPUs, lower ones are GPUs, all using honeybee[+].

#4

And finally the third question, commpared with honeybee[+], may I conclude that the honeybee legacy is not accurate(below is the comparison of 8 am.)?


I totally have no idea why the difference…:sweat_smile:

#5

Well, I think the time saving greatly depends on what type of simulation you run, and you may find in many cases the decreased time is so small. For example the time spent for the 8 am daylight simulation above using accelerad is 5.2s while the typical time using CPU is 5.9s. I suppose the accelerad is mainly designed for the speed-up of fake-color render of daylight simulation, in other cases such as the grid based simulation the contribution of GPU is not so large.

#6

The files are uploaded here:TEST.rar (1017.9 KB)

#7

@miraclema ,

The PIT type of study is not drastically different between legacy and HB[+]. The annual metrics we’re the ones that were really overhauled.

It looks like you may not have matched all of your Radiance parameters between the simulations to do an accurate comparison. The “splotchyness” of the HB+ results indicate that you should probably increase the accuracy of some of the setting a bit if your goal is a perfect match.

Also remember that the Ray tracing methods of payment for the Radiance are stochastic so you will never get a perfect match.

#8

I think I’ve figure out why:
Firstly when I calculate the 8am. sunlight I saw the difference of GPU and CPU in honeybee legacy simulation (High parameter):


It seems the direct light is missing using CPU.
When I use honeybee[+], both CPU and GPU simulation the direct light is missing, and the distribution of light is like above:

So I cut a hole the same place of window:

Then the results match!

The reason is now clear.I wonder had I wrongly add windows? Must I cut a hole on the wall?:sweat_smile:

#9

Hi, for this kind of simulation, my new examples can show you a clue. (the time spent showed in the screenshot)

1 Like
#10

Thanks @miraclema for sharing your findings.

I am curious to know what GPU are you using?

#11

The CPU of my laptop is core i7 8750H, with nvidia gtx 1070 graph card.

#12

Cool. Good to know. Thanks!