max rendering

[erilaz]

As far as I know (based on my experience of cards), they have no effect on render time. They only assist viewport speed.

[polymath]

Yea, erilaz is right, if you check the specs for any renderfarm machine, you'll see they usually have built in graphics or something with 8MB. Yet they do fine, so yea :)

[brismyth]

Thanks!

That is what I had hoped. What about a rendering board, like the Pure P1800? Has anyone had any experience with and of them? Is it worth the $3,699 list price? Any recommendations/general benchmark comparisons between that and a heavy duty dual-proc workstation?

[Null_Object]

Not really answering your ? but to add to this thread ram and processor speed directly effect rendering speed. If you have a single ghz pentium with 512 sdram and you have a quad pentium with 512 sdram set at the same speed you can expect to see similar rendering speeds. But up the ante on the processor to 2ghz and keeping the ram the same and you will see an increase in rendering speed and vice versa.

Many people often think that adding one or two more processors to your box will up the speed on rendering time this is a big misconception it will allow you to multi task better while rendering. I find multiple processors useful when I am rendering one scene and I am able to bring up another 3d max window with no lag or bogging at all from the other window rendering.

You have one processor or maybe even two allocating all of its processing power to this one job while the other processor with what ever ram is left over is free to be used for other things.

The gfx card directly effects how fast your prog interface will refresh and how well it handles poly/face count and realtime shading, texturing and lighting.

-Null

[polymath]

I don't notice any difference in the built in renderer with dual processors - however, in Brazil RS, there is a significant increase in speed with duals. This is because of the bucket system (duals = 2 bucks working instead of just one).

If your renderer can use both procs, then heck yea, the render will be quicker. And also, considering the fact we are talking about "rendering-only" systems, you won't necessarily need another processor to play around with.

The pure card is a raytracer basically. I'm almost certain that it does work with standard lights however, I believe it can only speed up the raytracing process. If you have a LOT of glass or other materials that require a lot of raytracing (chrome, gold, who knows), then pure would help you. But i dunno if the price to output is worth it in most cases. It really depends.

[Chris Thomas]

If all you want is a renderbox, then forget what Null_Object
says above. Duals do give a boost in render speed with any multithreaded renderer, and as far as I'm aware that max's scanline + all 3rd party renderers.
If the board you are talking about is the one I'm thinking of, forget it. Its very custom and only accelerates their own max plugin renderer.
The best bang for buck is a Dual Athlon MP 2000+ , on a decent motherboard. Fit it with as much ram as your scenes normaly use, if your local renders use only 300mb it would be a waste equiping the render node with 2gig. If your renders are quite meaty though, supply each render node with as much ram as necassary, i.e. your local renders generally take 1.4 gig, then supply the machines with 1.5gig or 2.
Do some tests with your production renderer, see if it benefits from 2 processors, if it does not, you could then go for a single proc athlon farm, this is very cheap, and still very potent. Its better to have 5 cheap machines rather than 1 expensive Xeon
Also, make sure that the central server that these machines get their jobs and techtures from is adequate and that the network speed is sufficent not to slow doen renders.

[brismyth]

Great info, thanks. We have been finding that we are getting almost a 2-1 decrease in render times when rendering on a 2-proc Xeon versus a similar clocked, single P-4. Also, space does start to matter with our system as we typically install it at a client facility, so form matters somewhat.

[zicher]

Multiple processors DO decrease rendering times in max. The rendering engine is multithread, and the OS will take care of using all the processors available. Unless you change an application affinity.

Not everything can be made multithread, so things like lights preparations (shadow maps) are not, and if your scene is 90% light preparation and 10% rendering you will not notice a big difference.

But on an average scene, where lights preparation accounts for much less, you will see that the rendering time on a dual processor is almost twice as fast as a single processor.

Accelerated graphic cards are useful only for viewport display, and have no influence on rendering time.

[Null_Object]

Through my own exp J I beg to differ about the difference in render speed from going from a single to a dual or quad. I have seen no significant increase in render speed what so ever as I stated above. Now I have seen a difference of render speed from going from a p3 to a p4.

Multi-Threading was mentioned. Multi-threaded processing - allows users with more than one processors to multi task there processors. Like I said in above post. The only advantage of a dual/quad is it allows you to better multi task while you work.

Keyword here is you can never have enough ram plain and simple.


-Null

[jimcoldt]

I found from running my own tests that the dual processor setup does indeed speed up renders.
My setup:

a typical max scene with 5395 polygons, 3 lights (one with shadows), some texture maps using opacity (which slow down renders).

My system: Dual Athlon 1.67ghz each
1gig registered DDR ram
Windows XP pro
Did one render with multithreading disabled (to be sure, disabled in max and disabled from the application through Task Manager).
Render time= approx. 12 seconds

Did the same render with multithreading enabled.
Render time= approx. 7 seconds

You can do the math regarding how this adds up over many frames....

By the way, Brian, I meant to mention in the email I sent you that I rendered the scene at 720x480 (NTSC video), as you would be doing.
Here is the image: (quality reduced in photoshop)

[zicher]

If you add a processor and things don't speed up, then you have a problem on your machine. Check every setting, multiprocessor kernel, registry. There are settings that might affect how the machine actually use the processors.

A second processor makes the rendering time almost half (except for the cases that I mentioned earlier), a quad processor it's a bit more than 3 times faster. I've been using dual and quads for years now, and I can guarantee you render faster.

Multithreading and multitasking are two different things. Don't confuse them. You can multithread and multitask with one single processor, but of course, things gets better when you have more than one.

Rob
discreet

[Null_Object]

Multi-threading is used solely in 3d terms for machines with more than one processor on board. I did not mistake multi-threading for multi-tasking only stated that it helps out in multi-tasking.

3d Max was one of the first 3d progs to implement multi-threading even before Softimage. This was partly due to the fact that it was geared for a Windows NT platform which at the time had the ability to support up to 32 processors.

The system and the way 3D studio was designed allows you to run more than one session at a time while running other apps.

Now being a native to Lightwave I understand how helpful Multi-threading can be as in Multi-tasking. It allows me to designate how many processors I want to be used solely towards my render job and how many I don't want used.

Zicher have you ever had a render job that used 100% of your processors while you tried to surf the net or texture in pshop?

I'm going to stand my ground on this one. As I said before I don't see any significant increases in render speed going from a single to duel or quad.

-Null

[MFreywald]

The difference in opinion here comes from the fact that Max lets the NT kernel distribute the threads unlike say Brazil or VRay which actually does the distributing itself through the use of buckets. Dual processors DO speed up the render times. But not drasticly unfortunatly if you use the default scanline. The biggest misconception is MS's use of the word multithreading. NT doesn't actually multithread. in the sense of what might/prob comes to most peoples minds as running two instructions of the same program/thread/process at the same time. What is actually happening and is still true to the word. Is NT distributes threads between the two. Depending on, as zicher noted, the way you have your NT configured. This is why it seems to speed up multitasking so much more than rendering (assuming your using the default scanline renderer of course). You can verify this by opening the task manager, going to the process tab and right clicking on a process. You can manually dictate which processor that thread/process is running on or let NT decide for you.

[gaggle]

Originally posted by jimcoldt
multithreading disabled. Render time= approx. 12 seconds
multithreading enabled. Render time= approx. 7 seconds
Originally posted by Null_Object
I don't see any significant increases in render speed going from a single to duel
:curious: ------------------> :annoyed:

[Chris Thomas]

Null, I'm sorry, but I've been using single and dual proc machines for years now. On render farms ranging from 15 to 90 machines. Some farms based on singles, others on duals. Over a year ago now I bought and built my own dual athlon, based on concrete experience of the benefits. Yes they multitask better, and yes they do accelerate correctly multithreaded software. I'm afraid your pissing into the wind on this one, otherwise why do you think duals are so popular for the CG workplace?

here's another real world example

my old athlon 700 single proc (768mb ram) frame = 240 seconds
my "new" dual athlon 1200 (768mb ram) frame = 70 seconds

now the clock speed is less than twice as fast as the 700, but it was a new athlon architecture, that combined with 2 procs gave a renderspeed of approx one quarter of the time.

I could be wrong.....
ILM cant be wrong
Pixar cant be wrong
The Mill cant be wrong

or maybe they are.....

[Null_Object]

Guess I shall just have to piss in a different direction.

-Null

[jarch]

Here's my dual PIII performance. Nothing else in the background except a few required processes.

With Max 5 running, no activity etc.
http://jarch.freeyellow.com/Processbeforesm.jpg


During the render of the gallery scene in the lighting tuts
http://jarch.freeyellow.com/Processduringsm.jpg

As you can see, both processors are at 100%...

The same results with a few other applications such as Photoshop.

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.