2GB or 1GB?

[tricky]

Does having 2GB Ram really make a massive difference to rendering speeds? I will be building 3 or 4 rendermachines, so I really want the fastest processor/mainboard/RAM that I can afford. However, 2x1GB sticks is significantly more expensive than 2x512MB sticks. I'm just wondering what the performance hit would be like (roughly).

All our workstations are 2GB BTW.

[nvanherpt]

a major factor will be how big the scenes are you intend to render. speed won't count for much if the pc's fall over due to insufficient memory.

[DaForce]

well it depends on what you render. If you could render all your scenes fine within 1gb of ram, then adding another gb will do nothing. If you couldnt render your scenes proper because it was always swapping over to the disk then adding another gb of ram will make a big difference.

So more RAM isnt faster, unless your renders need it. And even then its not making things faster its just preventing it from getting slower. (if that makes sense) I know that sounds odd but its the way it works. No point putting 4gb in a machine thats only going to be used for webbrowsing and the occasional game for example.

If you plan to render heavy scenes then more RAM will help for rendering those big scenes because it will allow more to be cached into the ram and hopefully none to the disk which is what slows things down or crashes max altogether.

So in short, if you render small scenes that dont require much RAM then no, adding more RAM will make zero difference.
If you render large scenes then yes it will most definitly make a difference.

I would go for 2gb atleast, because once you have the OS overhead and max opened up etc...etc.. alot of the first 1gb gets eaten up quite quickly.

[tricky]

Thanks for the tips. We do work with pretty big scenes, so 2GB would probably be a better option.

The Dynamic Memory option in Vray - does this effectively let you dial in the amount of RAM you have in order to prvent renders from crashing?

[DaForce]

I believe thats how it works. But to be honest i havnt used it all that much.

I think i will let the chaos guys describe it.

Static geometry is precompiled into an acceleration structure at the beginning of the rendering and remains there until the end of the frame. Note that static raycasters are not limited in any way and will consume as much memory as necessary.

Dynamic geometry is loaded and unloaded on the fly depending on which part of the scene is being rendered. The total memory taken up by the dynamic raycasters can be controlled within certain limits.

Dynamic memory limit - the total RAM limit for the dynamic raycasters. Note that this limit is divided between the number of rendering threads. If you specify 400 MB total limit, for example, and you have a dual processor machine with multithreading enabled, then each rendering thread will use 200 MB for its dynamic raycaster. If this limit is too low, and geometry needs to be loaded and unloaded very often, this may turn out to be slower than rendering in single-threaded mode.

Note that some objects (displacement-mapped objects, VRayProxy and VRayFur objects, for example) always generate dynamic geometry.

[tricky]

I've read that potion of the manual about 10 times - its still a bit confusing. I'm probably just a bit thick!

To me, it sounds like if I have 2GB RAM in my machine, I should be able to set the Dynamic Memory to, say, 1.5GB and, assuming the 3GB switch is enabled, my scene is guranteed to render (allbeit slower)

I doubt, however, that this is the case.

[DaForce]

you are right.. it only loads the certain parts of the object into memory that it needs. Not everything at once.

And you can set the maximum memory usage, but like it says if you set it too low it will take much long because of the extra work involved with loading and unloading.

( to clarify you are right about the way it works...not the being think part.. haha )

[tricky]

lol (though the way I'm feeling...)

[vlado]

Note that this limit is divided between the number of rendering threads. If you specify 400 MB total limit, for example, and you have a dual processor machine with multithreading enabled, then each rendering thread will use 200 MB for its dynamic raycaster. If this limit is too low, and geometry needs to be loaded and unloaded very often, this may turn out to be slower than rendering in single-threaded mode.

I suppose this statement needs updating in the help index; this was true for the 1.45.xx builds, however in 1.46.xx and later, all threads share the same memory pool.

Best regards,
Vlado

[DaForce]

ahh ok, i was kinda thinking that it seems abit odd.
Sharing the same pool is a better idea

[pshupe68]

OK so - If I have 2 GB Ram, I have enabled 3 gig switch and I am still getting crashes, which I am assuming is memory related, I should use Dynamic and set it to say around 1.5 Gig? That way it may be slower but at least it won't crash, which is a big deal when rendering an animation over a long weekend, in which crashing may take place with the first 2 or 3 hrs of rendering.

Would putting in another 1 or 2 Gig of Ram help? Is the 3 Gig switch the final number - If I had 4 Gig of Ram could I set it to 4 Gig??

Thanks Peter.

[percydaman]

i would think you could go up to 3gb of memory and still effectively use the /3gb switch. it means you have access to 3gb of memory per "program"

so if you have 2gb of mem you are using up to 1gb of virtual memory, which is a nono if you can avoid it. So if you can afford another gig, throw it in... and no the limit is 3gb, so putting in 4gb would be a waste at the moment.

[pshupe68]

Thanks percy - I think I will be doing that - adding another Gig.

Is that limit due to the OS? Will WinXP 64 solve that issue? I find I am using more and more memory as I get further into Vray. The images are definitely getting much better but render times and crashing is getting a little out of control. I guess that is to be expected.

I might look at another computer - I have Dual Xeon 2.8 with 2 Gig now and it takes about an hour to do test renders of a small interior - I could utilize another computer with switch box for mouse, keyboard, monitor. Go back and forth while the other does test renders.

Any suggestions on that thought - optimum setup? Dual AMD?

Thanks again.

Regard Peter.

[percydaman]

yes winxp64 will help. Itll really take all the memory you can throw at it. Until max goes 64bit though, not sure how much youll really be able to use. I have a 4gig system at home running winxp64bit and have been able to use the 4 gigs.

I think everyone here would recommend the amd X2 dual core processor...its rocks.

[DaForce]

Yeah the X2 processor is definitly the CPU of choice at the moment. But seeing as you already have a dual 2.8ghz xeon then the performance gain wont be huge. It will be abit but not massive. Check some of the benchmark threads in general and off topic ( i think) to see the difference in render times.

The biggest improvement would be if you got a dual Opteron 275 setup as then you would have dual dual-core so 4 cpu threads. That is "the" fastest render machine out at the moment.... well for the general public that is.

(sorry if this aint making much sense, its 3am here and im pooped... off to bed I go)