will this double precision be added as an option may be to blender 2.6?
so many peoples have complain about the internal precision and errors generated when doing lot’s of calculations!
i guess here that if blender use double precision it will slow it down a little and may use lessmemory !
like instead of using a max of 2.5 GB it may use only 1/2 of it may be!
any idea what the differences would be between windows 32 bits and 64 bits system?
thanks
happy 2.6
This guy uses blender 2.49 for creating machinery parts so maybe it is useful if you read his guide here:
http://homepage.ntlworld.com/r.burke2/precision_modelling1.html
look for the “Blender units and precision” explanation.
There’s Free-Cad.
http://sourceforge.net/apps/mediawiki/free-cad/index.php?title=Main_Page
It goes the other way around – double precision uses a lot more memory. Each double is 8 bytes rather than 4 bytes for a single-precision float. So when you have three floats per vertex, you can see how that’d start to add up real quick 
Note that this is mostly unrelated to the 32-bit vs. 64-bit CPU/OS question. That affects pointers (addresses in memory) which allows you to use a larger address space, but doesn’t affect floating point stuff.
memory limits is not the same on 32 bits then on 64 bits!
so if data takes more space using double precison = full python resolution then it takes more memory space
meaning there are less data that can be used by blender to fill up the maximum space
but it would give more precision for calculations which for certain scripts would be good!
and not certain if this would impact on the speed of calculations!
i mean doing double precision may not change anything in terms of speed because python is already using double precision by default
but is internal blender using python or C functions to do the calculations ?
thanks
The speed of calculations with doubles is not much different from floats (at least on x86, don’t know about other processors.)
The real difference is the size of the data, and here memory limits are important; using a 64-bit system doesn’t magically give you more memory, it just allows you to address a large amount of it. If your scene data gets big enough that parts of it are getting swapped in and out to disk as you work, you will not be a happy user.
Hmm. It seems I was wrong. So we could have about 44 million vertex positions per GB at double precision? The rest of the data (faces/edges/vertex colors, weight groups etc.) would stay the same. That’s not too bad I think if you only consider main system ram. Could be a strain on memory-starved GPUs though, and if I remember right double floats compute 8x slower on Cuda capability 1.3 and are artificially capped on all consumer cards starting with 2.0.
Considering all that, and the limited appeal of double precision, I think I’m against switching.
this should not be a default for blender !
only an option when doing some scripting and requiring higher precision for some calculations
it would be a nice options for peoples needing this high precision
which now a day is not that high with Todays’ PC getting faster and the new Intel generation of PC 22 Nmeter coming out this spring
plus the new GPU super high speed
hopefully this will become and advantage for blender soon !
just hope this can be implemented soon
happy 2.6
This isn’t the kind of change you would make optional, it touches any code which moves vertices about, including a lot of math funcs and drawing code.
while possible to typedef this and have DNA support the typedef - this would be a built time option, having as a runtime option would be really messy! (a lot of duplicate functionality), dont think anyone would seriously attempt that, and doubt it could be accepted into trunk.
so you basically saying that it wont be done for the time being!
or if it is then it would use double precision only no possibility to come back to single precision!
one note here
how come sometimes depending on the scale i think we can see in the N panel some locations and dim
numbers that have a lot more digits then single precision probaby double precision
is this an error or normal ?
thanks
happy 2.6
I just love the whole mindset behind Blender, like I said I use a cad/modelling system that I have paid for but it has huge issues which the paying customers have to find.
This all goes to my position that no software is released in a ready for customer state, they all have bugs which we have to work around whilst trying to do our day jobs.
Interestingly I gave Cinema 3d a go because it advertised an engineering interface, what I found in the engineering mode was a few more materials and a slightly different layout.
I admit I am no software genius but to me it seems only a short step to give Blender a cad environment.
A while back somebody posted a thread on this forum asking whether people could think in 3d, I wish I could think in 2d, I live in a 3d world so have always thought in 3d.
Hi, perfect timing - recently i have tried to modify Blender (2.49b) to work with doubles.
I was stopped by this issue: forked “double”-Blender can get common “float” blend files and transfer them to “doubles” on-the-fly, work with “doubles” and store back as “double” blend files. Though such files will lose their precision as soon as they are opened and saved by regular “float”-Blender version.
I have no idea how to prevent this without lose the compatibility.
Maybe it is acceptable though?
Compromise: to stay with standard “float”-format for blend files/data storage, but to use modified “double”-edit-procedures for all CAD tasks internally.
[quote=“RickyBlender,post:28,topic:529587"”]
only an option when doing some scripting and requiring higher precision for some calculations
it would be a nice options for peoples needing this high precision
[/quote]
Err, why exactly would you need such precision when it comes to scripting and calculating?
How many times have you modeled something and wished you had higher precision (that is unless you were intentionally trying to model at an ultra-high zoom level)?
How many times have people been wanting precision that floats can provide when scripting, how often would you need a precision like 0.000000000001 instead of 0.001? Now I have worked with creating scenes to render out in Luxrender and scripting logic for games in the BGE, but very really have I ever needed a precision of more than 4 places past the decimal point. Where would it even make a difference that you can see on your screen?
As for the CAD question, I will agree that Blender wasn’t designed to be a CAD program, but perhaps there could be an addon that adds better snapping and measuring tools and uses the grease pencil to draw on the statistics and the schematics.
well Blender is a modeling application for polygons and subdivision surfaces not really useable for CAD at all in the professional sense but you can integrate it into a modern workflow when you consider current changes in how software is being used.
if you look for something workable maybe give Rhino and T-Splines as a combo a test run.
You can use Blender subdivision modeling with T-Splines and bring the polygon model into a perfectly fitting NURBS model.
SharkFX is pretty cool but only a surfacing tool - I learned Cobalt which is pretty much the same software.
If you need a more a solid tool SolidWorks with T-Splines is very powerful more on the high end side function but also cost wise.