I've Made A Heavy, Custom, Rendering Benchmark. Looking for Testers.

It is essentially just a project folder for Autodesk Maya, but since they have 30 day, free trials available, this benchmark is free to run.

Right now, the scene is not as complex as I'd like, but I'll be working on updating it until I'm happy with how heavily it loads RAM and all of that stuff. Really it just needs more, and larger textures to apply.

The idea is that this is a benchmark for your entire system and will stress all the different computational functions you see in a real world workload. Some single threaded ops, some hard drive access, RAM access and best of all (in my opinion) it takes approximately 1:17:30 to render the frame on my 4.8GHz 2600K...this removes a MASSIVE amount of margin of error present in other benchmarks that are way too short to give a real picture of how well your system performs.

I'd love to get some people testing with similar systems to mine to really get a sense of what impact the RAM has on the final speeds. Cinebench isn't affected by RAM much, but Cinebench also has the ability to more or less sit in L3 without much back and forth through the IMC.

Who's interested?

Dropbox links to the Maya project and scene file below:

Project Folder + Textures

Instructions:

1) Install Maya and launch the app.
2) File -> Set Project, navigate to this project folder. You want to set "lib" as your project. If a message pops up asking if you want to create a default workspace, you picked the wrong folder. Hit cancel and retry.
3) Once the project is set, File -> Open and open the file inside maya/scenes/lighting.
4) In the top row of icons in the UI, towards the far right hand side, there's a director's clapboard icon that, on mouseover, says "Render the current frame (mental ray)". If the render window is 

5) Click the button to the left of that one, it will launch the Render Window. You can resize this to your liking.
5) In the render window, click the clapboard button at the top left and watch the pretty picture get made. Maya will note the render duration upon completion at the bottom of that window.

Please let me know if you give this a shot. Very interested to see results.
 

Quote:

Originally Posted by ugotd8 

I agree with your sentiments on this completely. Nothing tests a system quite like a heavy 3D scene with lots of polys. Unlike you, I have lightwave rather than maya, but there is a standardized SPECapc suite for it HERE. I like to use that test quite for benchmark comparisons to go along with cinebench numbers. Also, the reviews at TH commonly test CPUs with this benchmark as well. For example.

Thanks for that link, haven't seen much from that benchmarking suite before, going to check it out. From appearances alone, I'm not sure how much I'm liking it though...the scenes look overly simple in terms of texture and geometry loads and the quality of the final outputs doesn't look to be quite "feature" quality.

What are the average render times in the tests on the suite and what kind of RAM loads are you seeing?

I just tested an i7 3930K on this project @ 4.2GHz and it came in as being 25% faster than the 2600K @ 4.8GHz. Using straight up math, we can guess that the difference would have been more like 31% faster...but I'm guessing here that the lag seeps in from a combination of some lighter threaded tasks and some other inefficiencies introduced by adding more cores.

In theory, an equally clocked 2600K will return a score 66% of the value that the 3930K posts. In reality, I think we'll see the 2600K doing 71-73% of the speed.

This is the current render it produces:

I'm still refining a couple of things...there was a river generated in RealFlow that was spilling over the rocks, but that produced too many dependencies; people would have to install the RealFlow plugin, on top of Maya, etc. I'm putting the water feature back in, but just simplified...which gives a chance to add a couple large textures to the scene as well as more reflection tracing.

I also wouldn't mind evaluating the render on a calibrated IPS panel once I'm back on my home setup just to make sure the final image looks nice and purrdy...not necessary for a benchmark, but if it's a pleasing image, it's more fun to sit there watching it render.

Let me know if this pic is looking nice on your screens or if it's too dark/desaturated/etc.

Updated the OP as I've finished working on the scene and am happy with the way it is.

The final resolution is now 1024 x 576, or half the resolution of a film frame, but the sampling and detail is higher. The render now takes 1hr on my OC'd 2600K, down from 1hr 17m.

In addition, I have made the scores similar to CineBench. You SHOULD be able to expect more or less a similar score between this test and CB, normalized around the performance of my 2600K @ 4.8GHz with 1600MHz RAM such that it has the same score in each bench.

Here is how the final frame looks now (I did some post color correction for lulz):