Rendering and Hyperthread

[lando75]

I was under the impression from what someone told me that Maya does not make use of Hyperthreading, however, when i disabled hyperthreading in the bios my rendertimes have gone up, like 4 seconds on average. However, when on in the Bios, and turning on all available processors in batch render does not improve render times. So i assume that only applies to actual physical processors.
At any rate, from what i've observed, hyperthreading does improve rendering performance in maya. Can anyone shed more light on this situation?

[gmask]

Hyperthreading should improve the performance of any multithreaded application however from my tests the maya renderer does not allways take full advantage of the avaible CPU's but that may not be because of HT.

[lando75]

well, i guess i was just curious about this subject because of the test center on this site. My rig is a p4 3ghz with 2gb of pc3200 and my benches are drastically slower than the other two p4 3ghz systems in the test center. I'm assuming that the test is cpu intensive so I assumed perhaps that I myself was not utilizing something within maya to improve render times. my system posts a 59 or 1:00 for the test on this site and when compared to 40 sec render times by the same processor in another system, it makes me wonder what I can improve in my own. I mean that is a dramatic improvement of 30%!!! which could also improve my own workflow. Just wondering if anyone else with a similar rig has numbers to share or anyone who could shed some light on this.

Thanks

Oh, and i even overclocked my cpu just to see if the faster results were due to an overclock and I could only get like 56 with a 3.3ghz 220fsb overclock at 1:1. Even if i pushed it up to 3.6 i don't think my rig could reach the 40 mark.

[gmask]

Well the main thing I see that is different is that the systems youa re talking about in the database are 800mhz FSB and yours is 220.

However one thing you can try is to add -tw 256 and -th 256 to the commandlien and that shall force maya to use more memory when rendering. I thinkt he default is 64 and the range is 16-256

[lando75]

wel, i guess i may not have been clear. Actually my cpu is a P4 3.0 ghz 800mhz FSB w/ Hyperthreading. Intel has a quad pumped FSB which allows for 4 data transfers per clock cycle. Thus 200x4=800mhz FSB and therefore my cpu is the same type of processor as the other p4 3.0ghz.
I said 220 FSB because I did a 10% overclock just to see whether their times were results of overclocking. I overclocked witha 1:1 (FSB:RAM) synchroneous overclock w/ very low latency. Thus the FSB is 220x4=880mhz and the ideal memory bandwith is 7.04GB/s due to the dual channel (two 64 bit controllers on my motherboard) Therefore, to my understanding, if u overclocked my cpu another 10% to 3.6 and if the timings went proportionately i'd be sitting at a 52sec bench. It just bugs me as to what I might not be doing right to get the same results. Also, I don't think one buys Alienware to overclock and w/ over 25% better bench than my rig I must say i am amazed/shocked/and even maybe a little jealous =).
But, honestly I just want to optomize my rig so my workflow will improve.

Sorry, if I rambled a bit.

Oh, by the way are those commands permanent or do i have to manually input those values, each time i'm working. Ah, I guess I'll go try and see what happens myself.

Thanks for the info tho

[gmask]

Hey.. did you know that the results posted there are mostly dual processor systems?

My dual 2.6 system scores :45 but as a single it scores 1:14

Still the fastest entry there is a bit suspect cause it's a dual 2Ghz system.. I dunno but it soudns liek your system is what it should be. Then again my system scores better than most of the systems that shoudl be faster in that list.

BTW the tw and th commands can only be used on the commandline and are re-entered each time.. mosatly help with larger images but I find a minor speed increase when I use it in general

[lando75]

heh, i am a bit unfamiliar with MEL perhaps, but do u mean to type exactly this in the command line
-tw 256 [negative tw (space) 256] by itslef and then entering the second one or both together. Im not sure exactly what the command is? =P

[gmask]

I don't know why they don't have one flag for this but you have a flag (-tw) and then the numeric value (256) you need to declare both.

render.exe -th 256 -tw 256 myrender.ma

[lando75]

yea i know many are duals and for the most part the benches seem orderly, but the dual 2.0 p4, the p4 alienware and the 2.6 overclocked @3ghz rounding out the top? strange.

hmmm, call me stupid but i keep getting syntax error or procedure render.exe not found. Could u type out exactly what is to be typed in the command line as if u were explaining to a 1st grader? =) I think i need MEL for dummies

[gmask]

I'm assuming that the path environment variable for windows is set to point to th emaya directory?

What happens if you type render.exe into the windows commandline?

Here's a verbose commond.

E:Maya4.0binrender.exe -n 0 -tw 256 -th 256 E:rendertest.ma

btw this isn't really MEL it's simply commandline arguments

[joojaa]

Just so that you dont miss what gmask is saying he isnt talking of using mel hes talking of using a command line to start the render. You allways do that by the way!

Dont use the maya menucommand batch render. Its bit on the half stupid side of things.

[user_deleted]

Hi

Can you please explain it to me what does this command exactly do ?

Because my machine which was rendering the scene in 59 seconds is rendering in 48 using this command ????

Thanks

[mestela]

Its a direct interface to the maya renderer. When you run a batch render within maya, maya itself runs 'render.exe', but keeps an eye on it, sending output to the maya console. It also has to share resources with maya itself, slowing things down slightly.

By calling it externally yourself, your saving memory that'd otherwise be used by maya, and it also allows you access to a bunch of options otherwise hidden or difficult to reach. You get a full list by typing in 'render' by itself.

Incendentally, it can be different names depending on your platform. Under windows you can use 'render' or 'mayabatch' (is that right?), under linux 'Render' (note the capital r!), and I think its the same under osx and irix.

-matt

[joojaa]

Mayabatch isnt same as render... its a convinience command for running maya without the ui for script execution and such.

[user_deleted]

Hi

Probabely I didn't clear my self !

I meant the switches ???

-tw 256 -th 256

What do this two mean ???

Thanks

[gmask]

-tw tilewidth
-th tileheight

The switches force the renderer to work on a larger section (tile) of the image at a time and thus work faster.. also if you are running out of memory you can make it smaller which will take longer but will avoid memory errors.

Also using the -mm ### flag will force maya to use more ram instead of disk caching and as long as you have lots of ram this will improve performance.. however currently on windows Maya can only use about 1.5GB of ram.

### is the amount of ram to use in megabytes

[lando75]

Is there anyway to speed up the preview render in the ui?

[lando75]

oh and its weird but the -mm flag doesn't improve my render time at all for some reason. I tried -mm 2046 even though u said windows only will use 1.5 and -mm 1536. I have 2gb or ram. Hopefully it's not because i put the wrong numbers in =)

[joojaa]

Not much no. Besides i do my prewiews on command line too... much stabler (its not like it cant be initiated form within maya). And i cna reuse depthmaps.

Theres no use putting it past 1.4 - 1.5 (tha actual limit is somewhere betewwn theese values). Theres NO way any one proc can get allocated more than that, because the pointerspace of your processor ends. (linux has better memory management in that sense, as it can use slightly more mem than windows).

Besides if maya needs 400 meggs for the scene then theres not much you can do to use more (if thats all), ok so if you raytrace then you can increase the quadtree size.

This is why we want 64 bit porcs along with 64 bit OSes!