Curious about a few things.

[ToiletDuck]

Ok there are a few of things that I am wondering... A) how does cache affect render times? For example would a dual xeon 800mhz 2mb cache machine out do the dual athlon 2000+ 256kb cache machine? B) Does lightwave support 64bit rendering? If cache is a factor a dual itanium setup with 4mb cache each running at 800mhz would be nice. C) what is with the threads when rendering. I have a dual 2000+ system and I was wondering what the best option to set the threads on would be. Are there some cases where using 2 is better than the max? If you could explain that part to me I'd be greatful.
Duck

PS. If lightwave doesn't support 64 bit encoding why not? It is the future of computing is it not?

[mattclary]

That's a lot of questions, some of which have no simple answer.

A) More cache always helps, but will an extra 256kb outweigh ~1200mhz speed difference (for practical purposes)? Probably not.
B) No
C) To take advantage of multi CPU rigs, the app must be multi-threaded, which LightWave is. Under render options, you need to set threads to 2 or higher. Play with it to see which setting works best. (Hmmm, might have been quicker to do that than waiting for a response here... :) )

PS) Why? Because few people own 64bit processors. 64bit processing is not the future for ALL apps, but yes, I predict LightWave will (probably soon) be 64bit (or have a 64bit version available) due to the Opteron being available now.

[ToiletDuck]

I heard that it was 64bit up until version 6 or shortly after 6. You can buy 64bit itanium processors with 2mb-4mb of cache for cheap on ebay... less than $100. So on a dual system what are you guys seeing as the best amount of threads to set it on to render?

[mattclary]

Well, I'd say you heard wrong. Yeah, you can get 64bit Itaniums, but cheap? Find that hard to believe, unless it's very recent. People might be selling them cheap because they are moving to Opteron. Either way, 64bit hasn't been mainstream. As to the best amount of threads, that is not a cut and dried answer. For a dual processor machine, 2 threads is a minimum, and I'd say 4 is a maximum that might give any benefit. The most efficient use of thread count isn't something that can or should influence a buying decision. More CPUs are better, so get 4 CPUs rather than 2 if you can, but with either a 2 or 4 CPU system the thread count/performance gain is something you will need to experiment with. On my P4 3.0c is seen by Win XP as two CPUs due to hyperthreading, I have determined 2 threads is most efficient for me.

[ToiletDuck]

If you look around on ebay i always see 700 and 800mhz itaniums for under $100. I know they aren't that big on mhz, then again I was looking at cray's systems just to see what they are made of and some of them have like a 650mhz proceessor.

[Elmar Moelzer]

looking at cray's systems just to see what they are made of and some of them have like a 650mhz proceessor
Yeah but like a few hundrets of them computing in paralel, hehe ;)
CU
Elmar

[Matt]

maybe I've got it wrong, but I thought LW had a 128bit floating point pipeline render engine?

[ToiletDuck]

Ok but I'm still wondering. Would a dual Xeon then.... with 2mb cache at 800mhz out render a dual athlon 2000+ machine with 256kb cache. Yea or nea?

[mattclary]

Go to http://www.blanos.com and look at benchmarks for systems with 2 processors. When I look a Radiosity Reflective Things, the first Athlon comes in at number 18.

[Lightwolf]

Tough question...
The size of the cache hardly matters when rendering 3D, the speed of the cache does though.
Different parts of Lightwave are optimized for different processors (or, run better on different processors). Athlons seem to be quicker with straght forward rendering and Hypervoxels, while P4s / Xeons seem to excel with GI and raytracing.
Matt:
as far as the rendering pipeline is concerned: 128 is 32 x 4, i.e. 32 bits per channel (one for the alpha), which is the same as a single precision floating point number.

Cheers,
Mike

[ToiletDuck]

So your saying that I could take something like a very highly overclockable celeron that is cheap as dirt and it would be good for rendering?

Also what do you mean by the speed of the cache? As in the FSB?

[Lightwolf]

Originally posted by ToiletDuck
So your saying that I could take something like a very highly overclockable celeron that is cheap as dirt and it would be good for rendering?
Basically yes, if it has the same core. I'm not that up to date first hand on newer processors (I didn't update my machines for a long time), but when we built our first renderfarm we went for Celerons 466 instead of PIIIs 500, because they were almost just as fast and much cheaper. This is pure rendering speed though.

Also what do you mean by the speed of the cache? As in the FSB?
I doesn't make much of a difference now, since most L2 caches are on die and run at the full processor speed. The cache usually runs at a speed that is related to the processor speed, either 1:1 or 1:2.
Cheers,
Mike

[js33]

Originally posted by ToiletDuck
Ok but I'm still wondering. Would a dual Xeon then.... with 2mb cache at 800mhz out render a dual athlon 2000+ machine with 256kb cache. Yea or nea?

Well it's kinda silly to compare an 800Mhz CPU to a ~2Ghz one isn't it? Also Xeons run at over 3Ghz now so if you want a fast machine you should buy the fastest processors.
P4s are cheaper but you can't as yet run them as dual machines.
But for the price of one dual Xeon you could probably buy two complete P4 machines.

I'm waiting for fully parallel 16 processor P4s before I upgrade again. :D

Cheers,
JS