Jump to content
Hash, Inc. Forums

A:M Bench 2017 – Benchmark for Hash Animation:Master


Recommended Posts

Hi everybody,

 

I recently showed off my new computer system based on a AMD Ryzen 1700 (see more here).
To show what it can do, I created a new Benchmark for Hash Animation:Master which I call "A:M Bench 2017".

 

Download it here now:

- A:M Bench 2017

- A:M Bench 2017 & A:M teapot 2009 in one file

 

There already is one other Benchmark which is called "ThreeTeaPots" (you can find it here) from 2009 so why creating a new one?

The problem with ThreeTeaPots is, that it does not really use any of the newer features of A:M and will only use one thread to render.

Before it did not make much sense to get all the bells and whistles of A:M in a benchmark, simply because computer have been too slow to give a reasonable result in a reasonable timeframe.

 

But with the newer generations of computers that should no longer be a bigger issue and it may be of interested to see, if certain features are rendering faster on one machine or the other.

(this is not necessarily the case but it could occur)

 

am_bench_2017.jpg

 

What will A:M Bench 2017 utilize?

  • Ambient Occlusion (CPU based)
  • SubSurfaceScattering (SSS)
  • Particle (Hair)
  • Displacement maps
  • Bump maps
  • Raytraced Shadows
  • complex, perceentual materials
  • volumetric Light
  • Reflections
  • and more...

The file for A:M Bench contains of a few things:

  • A:M project file (am_bench_2017.prj)
  • 2 x Render preset file for the A:M project (am_bench_180p.pre, am_bench_360p.pre)
  • Netrender pool file (am_bench_netrender_pool.rpl)
  • Render preset file for the Netrender pool (am_bench_180p_multi.pre)
  • 3 texture files (big_texture.jpg, small_texture.jpg, small_texture_displace.jpg)

A:M Bench 2017 containts of 3 different tests.

You can do them all or just do one or two of them. It is up to you.

 

1.) Render an image in a resolution of 320px * 180px using A:M itself.

  • Start A:M.
  • Open "am_bench_2017.prj".
  • In the chor window click on "Render to file".
  • Click on "Load Preset" in the dialog and use the "am_bench_180p.pre" file.
  • Change the output path to your desktop.
  • Click on "Render".

2.) Render an image in a resolution of 640px * 360px using A:M itself. (this is the same as above but with another preset file)

  • Start A:M.
  • Open "am_bench_2017.prj".
  • In the chor window click on "Render to file".
  • Click on "Load Preset" in the dialog and use the "am_bench_360p.pre" file.
  • Change the output path to your desktop.
  • Click on "Render".

3.) Render an image sequence of 26 frames in a resolution of 320px * 180px using Netrender.

  • Start Netrender uisng "RenderServer_64.exe" (or RenderServer.exe if you are still using 32bit version).
  • Start as many RenderMessenger using "RenderMessenger_64.exe" (or RenderMessenger.exe if you are still using 32bit version) as you can. If you hit the limit of your computer, A:M will tell you that.
  • In Netrender, you should see a list of all the available Render-Slaves in the lower window on the screen.
  • Click on "File > Open" and open "am_bench_netrender_pool.rpl".
  • Depending on if you already did it or not, you may need to reset the frames. That can be done by right-clicking on the entry "A:M Bench 2017" and choosing "Reset Frames".
  • To start the rendering-process, select all but one of the slaves in the lower window, drag and drop them into the lower part of the upper window.

How to submit your results:

At the moment, you should provide a screenshot of your render result (for 1, 2 or 3 of the benchmarks) and post it as an answer to this thread. It is possible that in future there iwll be a webinterface to upload your results to. I will updated this thread then. Additional to that, please let us know about your system. You should provide these information:

  • Which benchmark did you do? (example: "A:M Bench 2017, 180p single")
  • Which version of A:M did you use?
  • Screenshot of the result
  • Result as text. (example: "10:15 min")
  • CPU model / name (example: AMD Ryzen 7 1700 @3.9 GHz)
  • RAM amount, type and frequency (example: 16 GB DDR4-RAM 3200 GHz)
  • Graphiccard chipset / model / name (example: Saphire RX 570 8 GB)
  • HDD / SSD name (Samsung Evo 850, 256 GB)
  • everything else like mainboard, power supply etc. is optional.

So why are you still reading? Start up your computer and show us what it is capable of! :)

 

Important: It may be, that A:M Bench 2017 will only work with A:M v19b and newer. If you experience a crash at the start of rendering, this may be the case. Please use A:M v19b (out soon) then.

 

Best regards and keep on splining!

*Fuchur*

am_bench_2017_n_am_teapot_2009.zip

am_bench_2017.zip

Link to post
Share on other sites

...and here we go with my results:

 

A:M Version:

v19 b

 

Computer:

- AMD Ryzen 1700 @3.9 GHz

- 16 GB DDR4-3200 RAM

- Saphire Nitro+ RX 570 8 GB

- SSD Crucial MX300, 525 GB

 

A:M Bench 2017 360p (Single Thread):

01h 17min 23s

 

A:M Bench 2017 180p (Single Thread):

00h 19min 12s

 

A:M Bench 2017 180p (Multi Thread, 26 frames):

00h 38min 34s

 

Best regards

*Fuchur*

am_bench_2017_360p_1slave_speed_big.jpg

am_bench_2017_180p_1slave_speed_big.jpg

am_bench_2017_180p_15slaves_multi_speed_big.jpg

Link to post
Share on other sites

...and here we go with my results of my old Machine:

 

A:M Version:

v19 b

 

Computer:

- AMD Phenom II 1090T @ 3.6 GHz

- 16 GB DDR3-1800 RAM

- GigaByte HD 7870 2 GB

- 250 GB SAMSUNG HD103SJ

 

A:M Bench 2017 360p (Single Thread):

01h 58min 56s

 

A:M Bench 2017 180p (Single Thread):

00h 31min 32s

 

A:M Bench 2017 180p (Multi Thread, 26 frames):

02h 27min 07s

 

Best regards

*Fuchur*

pII_1090T_180p_1thread.jpg

pII_1090T_360p_1thread.jpg

pII_1090T_180p_5thread.jpg

Link to post
Share on other sites

Using the latest version of v19 (the one from May). I only ran the smallest one, but it tested better than expected. Not sure if v19b will make a difference, but I'll re-test it once it's available.



A:M Version:


v19.0 RC SSE4 (macOS Sierra 10.12.6)



Computer:


-iMac (Retina 5K, 27-inch, Late 2015)


-4 GHz Intel Core i7


-24 GB 1600 MHz DDR3


-AMD Radeon R9 M390 2048 MB


-2.12 TB Fusion Drive



A:M Bench 2017 180p (Single Thread):


00h 21min 51s



Screen Shot 2017-08-26 at 9.56.56 AM.png


Link to post
Share on other sites

Hey Mark,

 

thanks for running it :).
It is totally fine to only run the single-thread-180p one only. I doubt that 19b is faster than 19 RC, but we will see :).

The results are astonishing, but than I had a look what CPU is running in your Mac and it explains a lot ;).

Your Mac is running an Intel i7 6700k (more or less the same performance like the Intel i7 7700k, which is one of the strongest 4-core-cpus available even today but is only a refresh of the 6700k with a higher clockspeed). There has been very little performance increases for Intel-CPUs in the last years.

 

Here you can see a comparission between a 7700k and a 6700k:

> http://www.pcgameshardware.de/Core-i7-7700K-CPU-264844/Tests/7700K-vs-6700K-1217266/#BenchmarkCollection_1217266_432

 

There is may be a difference of 6-9% between those two which is quite astonishing too. (that is really close to nothing in day to day work).

 

The Ryzen 7 1700 is a direct competitor of the 7700k but has more cores.
For the multi-cpu-test you would properly see a bigger margin between the i7 6700k and the 1700, but for single-core stuff, they will come out more or less the same (difference here is a little bigger at about 13-14%) as we can see here. :)

 

Thanks a lot for testing. That is very interesting for me too :).

 

Best regards

*Fuchur*

 

PS: Everybody is very welcome to post his/her results here too :).

  • ____ 1
Link to post
Share on other sites

Yes, it will be out soon... but it is in beta testing currently.

I thought, 19a would have been out anyway, but seems like Jason did not release it yet.

I am sure a or b will be out soon ;).

 

See you

*Fuchur*

Link to post
Share on other sites

Don't "winky" face. This issue is neither funny nor cute. If Jason is the only one that can handle such matters then someone has to get him on the ball or if someone else can be then someone else should be given the job. Besides this matter their is also the times I've seen with people or a person was stuck waiting for a new license. A fall back needs to be assigned or hired to keep these things from happening.

 

You may have your slightly bug fixed update but that most of the rest here don't makes it rude to announce your special case/privilege so flagrantly.

Link to post
Share on other sites

Hang in there. You raise valid points and we (collectively) do need to look at options going forward to prevent the bottlenecks.that can occur

I think Fuchur is as frustrated as you and was winking because he thinks v19b will likely arrive before v19a... most unfortunate and frustrating. But also nothing that is under his control.

 

I'm hanging with you here eagerly awaiting the next release too.

Link to post
Share on other sites

Using a smiley face there is just to keep the mood a little bit up because I can't change it but I am certain there is a solution on the way. This is called "Galgenhumor" (gallows humor) in Germany.
Very sorry if you did not get that... may be a cultural thing...

 

Just to make sure we are on the right track here:
I'm not hired in any way and I am voluntarily doing beta testing of the software so that other people get more bug proove software. And having a beta program or even a software build which is not going public is quite a standard behaviour concerning any kind of software (happens all the time for instance at Microsoft with their Insider Beta Program). When I do something with a newer version, it is a sneak peak for the future for others. If you do not want that to happen, let me know and I will stop contributing to the community with such things (which at least are hopefully of use a little bit later for everybody)

 

I am willing to help where I can (on the forums, with video tutorials and more because I like the software and the people here). I'm not in any way employed or anyhow attached with Hash financially.

So talking with me about that kind of stuff on this forum is really of no use, because I can not change anything about it and I only have good intentions when contributing here.

 

I find it a little sad, that this positive thread is going down such a bad road. I wanted it to be a good thread helping everybody to see if a new computer would benefit him/her with A:M.

Sorry that you see it differently.

 

If you do not want me to contribute to the community anymore with this benchmark or anything else – just tell me again and I'll get this thread deleted by the moderator or admins as soon as possible.

 

Best regards

*Fuchur*

Link to post
Share on other sites

Like I said. unfortunate and frustrating.

 

Don't read too much into that post Fuchur.

Read between the lines and know that the underlying issue isn't about you.

 

I wonder if the powers that be (Jason? Martin?) would approve a different system for distribution of associated files and executables?

Perhaps even target those items that never change as being separate from the primary installation to streamline updates even more.

And yes, I'm talking github here.

Post a disclaimer on the download page that tells users to 'Use at own risk' and then... off we go.

 

Jason?

Martin?

 

(And of course... all of this would also depend on what Steffen is willing and able to do)

Link to post
Share on other sites

I find it a little sad, that this positive thread is going down such a bad road. I wanted it to be a good thread helping everybody to see if a new computer would benefit him/her with A:M.

 

I appreciate what you're doing, Gerald! It's helping me decide if I want to use a Ryzen processor in my next build...I'm debating setting up a dedicated renderer.

Link to post
Share on other sites
  • 4 weeks later...

Hi, below are test results from my PC....

 

A:M Version:
v19b

Computer: (Purchased Jan 2015)
- Intel i7-5930K @ 3.5GHz
- 16 GB DDR4 2133 MHz RAM
- nVidia GeForce GTX 980 4GB GDDR5 SLI 2x (v385.69)
- SSD Samsung PM851 128GB
- HD Toshiba 2TB 7200RPM
- Windows 10 Home 64-bit v1703 b15063.608

A:M Bench 2017 180p (Single Thread):
00h 18min 3s
A:M Bench 2017 360p (Single Thread):
01h 12min 36s

A:M Bench 2017 180p (Multi Thread, 26 frames, 6 cores, 11 slaves):
00h 53min 55s
A:M Bench Teapots (Single Thread):
00h 1min 33s

 

A:M Bench 2017 180p (Single Thread):

AM Bench 2017 320x180 2017-09-22.JPG

 

A:M Bench 2017 360p (Single Thread):

AM Bench 2017 640x360 2017-09-22.JPG

 

A:M Bench 2017 180p (Multi Thread, 26 frames, 6 cores, 11 slaves):

AM Bench 2017 320x180 Multi 2017-09-22.JPG

 

A:M Bench Teapots (Single Thread):

AM Benckmark Teapots 2017-09-22.JPG

Link to post
Share on other sites

Hey, thanks for participating.

Wow, that system must have cost a small fortune... even today in 2017 that processor is really expensive (a new one would cost about 500-600 Euros over here) and a dual GPU/SLI setup is expensive by itself.

 

But maybe well deserved. Those are nice results especially in the single core performance. :)

 

Best regards

*Fuchur*

Link to post
Share on other sites
  • 3 years later...

Today I updated my CPU to a new AMD Ryzen 9 5900x with a new Mainboard (Asus Prime B550 Plus) and wanted to see how much that will help with rendering in A:M.

The result:

  • It is about 1.7x faster for single-core 180p rendering time:
    My AMD Ryzen 7 1700 was at around 19:12 with 180p single, the new 5900x is at 11:19.
  • The rendertime for 360p single-core:
    AMD Ryzen 7 1700 was at 1h 17m 23s, the 5900x is at 45m 21s
  • ...and of cause the 5900x has even more cores (12 core vs 8 cores) and does quite much better there too:
    AMD Ryzen 7 1700 was at 38m 34s, the 5900x is at 26m 17s.

But why isn't it even doing better with 12 cores then 8 cores (here it only gets up to about 1.5x as fast)?
Simple answer: The newer versions of A:M do not allow to use all threads but only cores at the moment. So I can only render with 12 instances while I rendered with 14 with the 1700 here.

Short answer: The 5900x is pretty much a beast of a thing and it is much faster than my 1700... not exactly cutting it half but :).

Best regards
*Fuchur*

210512_amd_ryzen_9_5900x.JPG

210512_amd_ryzen_9_5900x_360p.JPG

210512_amd_ryzen_9_5900x_mulit.JPG

Link to post
Share on other sites

Hi Robert,

I just tried it... a little under a minute at about 58-59s with the AMD Ryzen 9 5900x and A:M Version 19.0 p.
I think it cuts the next best entry till now by Jason by about 16s for the teapot and is the first one under 1 minute :).

(link: https://forums.hash.com/topic/36753-cpu-render-benchmark/page/8/)

Best regards
*Fuchur*

210512_amd_ryzen_9_5900x_teaport.JPG

  • clap 1
Link to post
Share on other sites
24 minutes ago, Fuchur said:

Hi Robert,

I just tried it... a little under a minute at about 58-59s.
I think it cuts the next best entry till now by Jason by about 16s for the teapot and is the first one under 1 minute :).

That is 20x faster than my first result 11 years ago!

  • clap 1
Link to post
Share on other sites

Man that is improvement ;).
I am pretty sure it has to do with the improvements in A:M too, but still: Damn that is fast. ;)

Best regards
*Fuchur*

Link to post
Share on other sites

You could load up v15 and see how that compares.

I think about half of the speed up will be from coding improvements since v15

Link to post
Share on other sites

I just downloaded v15j+ (32bit) and got these results with it. It is between 03:24-03:30 for the teapot benchmark.
Hopefully I got everything right and compareable with the render settings, because that version has not yet the feature to import render preset files.

It is a little bit hard to compare anything like that because: Yes of cause, the software is much faster today but the reason for that is the processor itself too. If for instance the new processor has a new instruction set like SSE4.2 it can be used by the software if it is implemented.

But how to weigh that in? Is it possible because the software is supporting more or because the hardware has involved and is better?
I'd say it is both. Anyway I think 3:30 is still pretty nice for such an old version.

Best regards
*Fuchur*

210512_am_15j_32bit.JPG

Link to post
Share on other sites
1 hour ago, Fuchur said:

Hopefully I got everything right and compareable with the render settings, because that version has not yet the feature to import render preset files.

If you set Render to "Use Camera Settings" that should retain all the proper settings.

 

And I presume you have your computer set to a power management setting that doesn't let the CPU throttle below 100%.

Windows 10 does this well but i could never get Windows 7 up to 100% unless i was running multiple instances of A:M simultaneously.

Link to post
Share on other sites

I am using Windows 10 with the default powermanegement.

Anyway it would only be one core or thread no matter what and I doubt Windows is going to throttle that CPU because of about 6% usage at most ;).
The 5900x is a 12-core / 24-threads CPU. I asked Steffen recently to see if there is a reason why there are only 12 render slaves with netrender available currently if you have a 12-core CPU (I seem to remember it was the thread count not the core count and maybe it would be good to have 1 or 2 threads for net renderer to manage the others) but lets see what he says about that.

I re-did the test now and it seems to be pretty consistent even with using the Camera settings as the output.
This time it took version 15j+ to complete it: 3:26. (within margin of error I would say). CPU went up to around 4,7 GHz on one of the cores but I did not exactly get the right time for the screenshot ;).
Interestingly it seems like the computer switched the threads at the begining. It maybe that those are "the best" cores he is switching between. (that is a normal thing. The CPUs are going to be messured by the system and than there is a core which is marked "best" and one that is "second best" which are going to do most of the single-core usage because they are the fastest for that kind of task.

Best regards
*Fuchur*

210512_am_render_v15jplus_teapot_5900x.JPG

Link to post
Share on other sites

By "throttling" I really meant the CPU speed. My 3.5GHz CPU can sit around at 2.2 or even 1.6 unless I turn on a "power plan" that sets the minimum processor state to 100%

 

However... When I'm just running A:M and rendering on my 4-core machine the usage of the cores can go as high as 49% and is always above 25%. I wonder what is happening during render that uses more than one core.

Link to post
Share on other sites

Hi Robert,

ah I see... yes that is pretty much fine... 4.7 GHz is more or less max if you are not doing a overclock by hand... sometimes it can go higher than that even, but in general 4.7 to 4.9 is the max.
I am not sure what is causing the above 25% but that seems to be normal. Maybe it is about refreshing the windows or switching between cores or something.
But it is constantly going below that again after a short while. I think only Steffen could answer that, but it is very likely just some kind of overhead like that.

Best regards
*Fuchur*

Link to post
Share on other sites
Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...