Thorsten Froehlich <tho### [at] trfde> wrote:
> Andycadd wrote:> > Disk buffering ..... that is working in 3.6 currently?>> There is no "disk buffering". What POV-Ray 3.6 and earlier do is write the> file to disk line by line. While POV-Ray 3.7 also writes the file in a> temporary format to disk to support a continue trace in case of a crash, it> further keeps a whole copy in memory as it will now write the final image to> disk at the end of the render, rather than write it incrementally and> (mis)use it as continue trace buffer like 3.6 and earlier did.>> The problem with using any fixed image format for continue trace in 3.6 and> earlier is that image formats do not actually store all the information> necessary to properly continue a trace, in particular with anti-aliasing> enabled. Hence this problem has been addressed in 3.7.>> > How come the windows virtual memory doesn't just pick this up?> > PLEASE excuse my ignorance.>> Because you still only have 2 GB to 3GB (depending on Windows version) of> address space for a 32 bit application.>> Thorsten
Is there an option to not keep a full copy in memory (freeing up that 320MB
mentioned above) with or without a running display ? I know this would be a
speed hit but Rambus memory ain't cheap. Or maybe just keep the previous
row in memory and flush the extra?
From: Thorsten Froehlich
Subject: Re: SMP positive response with minor problem animating PNG files CRASH
Date: 10 Oct 2006 18:11:46
Message: <452c1aa2@news.povray.org>
Andycadd wrote:
> Is there an option to not keep a full copy in memory (freeing up that 320MB> mentioned above) with or without a running display ? I know this would be a> speed hit but Rambus memory ain't cheap. Or maybe just keep the previous> row in memory and flush the extra?
Ben Chambers wrote:
> If it's really a problem with memory addressing, as you seem to be> suggesting, perhaps a more suitable error message could be introduced?
As I said before: "disk buffering has not been implemented yet." So this
issue will go away really "soon"...
Thorsten
Thorsten Froehlich <tho### [at] trfde> wrote:
> Andycadd wrote:> > Is there an option to not keep a full copy in memory (freeing up that 320MB> > mentioned above) with or without a running display ? I know this would be a> > speed hit but Rambus memory ain't cheap. Or maybe just keep the previous> > row in memory and flush the extra?>> Ben Chambers wrote:> > If it's really a problem with memory addressing, as you seem to be> > suggesting, perhaps a more suitable error message could be introduced?>> As I said before: "disk buffering has not been implemented yet." So this> issue will go away really "soon"...>> Thorsten
Sorry bout that I misread to be clear I love the SMP features makes my Dual
2.4 Xeons useful again. I also own an older Quad Xeon w/1MBcache EA proc
does the SMP feature use all 4 procs? They are only 550Mhz ea but big
cache size could help. should I put it back together?