Of course it won't be linear, but at least 70%?
Announcement
Collapse
No announcement yet.
vray scalability to 12/24...or more cores
Collapse
X
-
ok Vlado ,so i can choose something from my production .... or Crayox you can send to me your test scene with expected render time (5-10 min )
cpus i mean cores ofcourse
and yes i changed number of LC passes with numbers of cores i allow to compute vray test scene ...
and my rig : tyan s4985-E , 4x barcelona B3 2,2 ghz , 12 gb of ram mixed 1 and 2 gb modules but still two of them for each socket ....Last edited by Janosik; 21-09-2010, 10:31 AM.
Comment
-
i can explain it by my way : simply somehow vray scaling only up to 8 cores after that number it cant ( ofcourse only on AMD cpus ) intel is fine SOMEHOW ....
so if someone have newer AMD procesors 6 or 12 cores please try to render huge scene to see CPU scalability on amd cores.
btw : vlado maybe my situation is because i have 4 sockets mobo , all other systems here are still 2 sockets mobos nomatter what brand have ....
Comment
-
Well, scalability is kind of difficult, because no matter what it isn't going to be linear. When adding more cores, you are also adding more stress on the RAM/Memory bandwidth. Too me it looks like you have hit a bottleneck in one of the subsystems in your hardware. I am just completely guessing here, but it may be that with that many physical CPUs, you might be running into a problem regarding the memory bandwidth. When all the cores are trying to render and grab information from RAM, you start exceeding the RAMs ability to provide that information, which may end up with very similar results to what your times are showing. I know that this was a problem with some of the early HT Intel processors. Sometimes turning off HyperThreading (HT) there would be a noticeable speed increase.
You mentioned you are running a 4 socket motherboard...what OS are running? Windows Server 2003? I believe that XP, Vista, 7 can only have up to two physical processors.
In some of my earlier tests with AMD, I found that the memory subsystem was not very robust creating many slowdowns compared to a similar Intel machine. Granted that was several years ago and I am NOT very familiar with the Quad CPU machines from AMD. But in all of my hardware evaulation testing, I typically run into a bottleneck somewhere along the line and no matter how much more I try to push, I am pretty much stuck due to that limitation.
This is all speculation on my part, and would require you to do a LOT more testing on your end. You should also start looking into the hardware that you are running and checking out the specs to see where potential problems may arrise.
I don't know if this helps any, but right now, I would highly doubt that VRay is the problem.
Comment
-
comon !!
this is not to do with memory bandwith , this is software issue and question is : autodesk or chaos fault it is ?? win server 2008 64 bit , tyan s4985-E (four sockets mobo) 2,2 ghz barcelonas B3 (without cache maulfunction ) 12 gb or ram two sticks for one socket, so all showing to sotfware problem maybe intel compiler have some problem with more cores and sockets dont know ... who except Vlado can tell righ answear ??
with maxwell demo no problem with scalability ....
Comment
-
Just curious - when it's rendering on more than 8 cores how does the processor graph look? Is each core at 100% or are the extra cores much lower?www.dpict3d.com - "That's a very nice rendering, Dave. I think you've improved a great deal." - HAL9000... At least I have one fan.
Comment
Comment