# of buckets during render?

[markf]

I have noticed on one computer with a core 2 duo processer Vray renders with two buckets. On anther machine of mine with dual Xeon 2.66 procs. Vray renders with four buckets.

 

I'm not sure why this is. Does the # of buckets used for render make any difference in rendering times? Is there somewhere in Vray that I could change the # of buckets used at render?

 

Thanks in advance for any relpy.

[alias_marks]

Good question, I have noticed the same thing and am curious what the deal is too.

 

*subscribed

[alias_marks]

whoops, double post (note to self - don't stop browser once submit button is pushed)

[Christopher Nichols]

If Hyperthreading is turned on, each CPU will have two threads. So a dual Xeon will have 4 buckets. It is not actually that much slower (if at all) if you turn off hyperthreading and only use 2 buckets. Hyperthreading is sort of a dead technology since they came up with dual cores... a much more efficient way way of adding threads... adding cpus...

[markf]

Thank you for explaning that Christopher.

[blaz]

i read on another forum, and somebody gave a script to bring a according to number you want. it works for me in vray. you can try in your render engine. here is the script, type them in your script listener..i hope u are using 3dsmax..

 

"renderers.current.system_numThreads=4"

 

"4" is the number of bucket you want, if you want six bucket then you just type "6"

 

i hope it helps

[markf]

Thank you for your reply.

 

Why would one want to change the number of buckets?

[tedesco]

I don´t get why would I have 6 buckets if I only have two possible processors calculating it?

The hyper threading tecnology let´s you use two operations at a time, so each processor can calculate two buckets at the time...2 processors 4 buckets. If you say u want 8 buckets and you only have 4 available, maybe (I´m not sure about it) it will give two to a single processor, so it will take two times more (so there´s no reason to do that). I don´t know how the processor will handle the extra buckets, but if there´s no possible resorce to calculate them, what I can be sure is that you will not gain from doing it.

I´m using distributed rendering on 8 HT machines, realy wonderful, having 16 buckets at a time in your image.

This is an interesting issue, I haven´t made a test to see what´s best, HT on or off.

Any coments on the subject?

[AJLynn]

That is correct - you want the same number of buckets as you have "processors". I put that in quotes because a processor here can mean a core of a multicore CPU (a Core2 Duo has 2 processors) or a logical CPU caused by hyperthreading (a P4 HT has 2 processors with HT turned on, and a dual-core Pentium 4 Extreme, with dual HT cores, has 4 processors).

 

There is no good reason to have more buckets than processors - doing this would increase memory use without decreasing render time.

 

HT is a kludge to compensate for the incredible inefficiency of the P4 (Netburst) microarchitecture. Because Netburst has a very lage number of pipeline steps (34, IIRC, compared to 11-14 for most other CPUs) it is possible to start one instruction while another is still in the pipeline. This is nowhere near 100% efficient, so the actual speed increase varies by what you're doing but 10-15% is a reasonable average. The Core microarchitecture dispenses with the long pipeline and its associated artificial clock speed numbers and high power consumption, and HT is no longer worthwhile so they don't include it.