about the benefit of more RAM for overall performance

Forums: about the benefit of more RAM for overall performance
Autodesk Softimage XSI (Developed by Face)

about the benefit of more RAM for overall performance

Postby DemiTree » Sat Feb 22, 2014 4:05 pm

DemiTree Sat Feb 22, 2014 4:05 pm
hi i am new to both SI and octane and really excited to have found my way here.
also pleased to see your pace of development.
I am an architectural designer in frankfurt who is in transition from rhino/grasshopper + vray workflow to softimage 2014 64bit + octane.

i want to ask two quick technical questions and lastly about your plan for integration with ICE, 1) does more ram for CPU help speed up the raytracing ? 2) what is Ms/sec a measure of? I realized the 2nd question might be more a softimage question than Octane question.

very recently I added 16gigs of ram to my system (so total 32 gigs now), and noticed that the number notated as "Ms/sec" on the top left has doubled although I have not added any GPU.
currently i have 2 gtx 765m installed in my laptop with SLI DISABLED, which I can upgrade anytime because it is a schenker notebook.

This measurement, which I am guessing is a measure of how fast the rendering is taking place, and I see it decrease when I am using an additional display monitor, which makes sense.

since softimage is exporting the existing scene to Octane each it renders, i thought more ram probably helps a little before raystracing starts loading the scene into GPU, but doesnt do any work during the raytracing. but then, GPU has to pass the result back to CPU at somepoint, i guess more RAM will also help here, somewhat reducing delay time esp. if I have a big scene like a district of a city? i want to ask if this is a correct.

and lastly, its probably a question a lot of people are curious about. I am transtioning to softimage mainly because of ICE. will you be working on octane 1.5 version of softimage plugin before ICE?

thank you..! again i am excited to be here,am really interested in CUDA computing in general.
DemiTree
Licensed Customer
Licensed Customer
 
Posts: 6
Joined: Sun Feb 09, 2014 3:23 am

Re: about the benefit of more RAM for overall performance

Postby face » Sat Feb 22, 2014 7:54 pm

face Sat Feb 22, 2014 7:54 pm
Hi and welcome Demi,
more Ram is good to store some things inside Softimage but will not speed up the rendering itself.
The rendering is done in the GPU.

Ms/sec is a speed indicator.
An image with a resolution of 1280x720 has around 1MPixel.
When you get 50Ms/sec on a scene with a resolution as above, each pixel is rendered 50x in one second.
Is the max samples set to 500 samples, your rendering is done in 10 seconds.

Also the double of the speed after you have added ram, shouldn´t be correct.
Only when the GPU has shared ram, but i don´t think so.
The ram is only used to convert and transfer data from and to the GPU.
You´re also right with your remark that more ram will help you on big scenes.

ICE will be a part after the next release and after some issues with this release are solved.

face
Win10 Pro, Driver 378.78, Softimage 2015SP2 & Octane 3.05 RC1,
64GB Ram, i7-6950X, GTX1080TI 11GB
http://vimeo.com/user2509578
User avatar
face
Octane Plugin Developer
Octane Plugin Developer
 
Posts: 3204
Joined: Sat Mar 06, 2010 2:10 pm
Location: Germany

Re: about the benefit of more RAM for overall performance

Postby DemiTree » Sun Feb 23, 2014 12:56 pm

DemiTree Sun Feb 23, 2014 12:56 pm
i see, thank you face very much for the clarification.

strangely when i added ram 16g more, the first time I ran Octane within soft i was getting just over 1 Ms/sec. then now its back to typically around .6 - .7 Ms/sec. but at times fluctuates down to 0.4 so when I render 1280 x 760, it renders all the pixels every 2 seconds. so 1000 seconds for 500 samples, so it should take around 16 mins but the actual render completed after 14min. and my render preview shows that a render after 500 samples is still too grainy. caustic blur setup is all the way up to 1. additionally I tend to run other programs when I am rendering with octane for soft. There seems to be some room for improvement for speed and quality with the way i am setup and the way I am using octane for soft.

Q1 Is there a guide with tips for reducing graininess with different settings? I was hoping to do an animation of an urban scene with a light crowd fx (SI 2014), turning from a day time street view walk through to a night time panorama. the night scene will have multiple IES light sources, i think. at the currently pace with my set up, I think such an animation will take a long time to do.

2x GTX 765m is a bit less powerful than one GTX 780m. it is just under top 25 mobile GPUs today(ranked 26th ) here in this link (http://www.notebookcheck.net/Mobile-Gra ... 844.0.html) Q2 : I wonder if there is any benchmark information for Octane render engine for different GPUs models to check that i am getting what I should expect given that I am running my GPUs on the basis clockspeed?

first all, I am considering to use EVGA Precision X to overlock my 2x GTX 765M just to see the improvement. i am guessing I should get at least 15-20 percent increase in performance. so I wanted to ask if Q3 anyone has any experience with overclocking laptop GPUs for Octane, and has any opinions or concerns.

personally i am actually crossing my fingers for the desktop GPU over thunderbolt connection laptop applications to get pass licensing and development so that I could invest into one or two desktop GTX 780 TI (s) first before upgrading my laptop on board GPUs, since it would be better performance per cost than buying two GTX 780Ms (more than a thousand euros together) which would give me slightly better performance increase of JUST ONE desktop GTX 780 (less than 500 euros).

ive been eyeing Radeon R9 m290X crossfireX setup, which is 2/5th of cost of GTX 780m SLI but AMD cars will not work with Octane so, it seems like i am stuck with the most expensive way to scale up - NVIDIA and mobile GPUs, unless the external GPU over thunderbolt connection for windows somehow WORKS OUT :) like silverstones thunderbolt GPU (http://www.anandtech.com/show/7697/silv ... c-and-more)

in that case, the argument for GPU scale up model being a lot more cost effective than CPU scale up seems, as in my case - a high performance mobile computer, not as compelling as it could be.

50Ms/sec sounds really great. btw :)))
DemiTree
Licensed Customer
Licensed Customer
 
Posts: 6
Joined: Sun Feb 09, 2014 3:23 am

Re: about the benefit of more RAM for overall performance

Postby face » Sun Feb 23, 2014 5:55 pm

face Sun Feb 23, 2014 5:55 pm
50Ms/sec is on the helicopter scene in the "Preview Octane for Softimage 1.02" thread.
Not much fast i think ;)

About your questions, i have found this for your benchmark question:
viewtopic.php?f=9&t=7144&hilit=benchmark
But i don´t know how up to date it is...

You can also try to use the search function for the other questions or, if you find nothing, make a new thread in the "General Discussion".
I think that would have much more readers then this thread here in the Softimage section.
Here are only some peoples active from time to time ;)

face
Win10 Pro, Driver 378.78, Softimage 2015SP2 & Octane 3.05 RC1,
64GB Ram, i7-6950X, GTX1080TI 11GB
http://vimeo.com/user2509578
User avatar
face
Octane Plugin Developer
Octane Plugin Developer
 
Posts: 3204
Joined: Sat Mar 06, 2010 2:10 pm
Location: Germany

Return to Autodesk Softimage XSI


Who is online

Users browsing this forum: No registered users and 5 guests

Fri Mar 29, 2024 3:08 pm [ UTC ]