64 bit double precision engine - on or off ?

drewfx1


It just does internal mathematical calculations with more precision (64 bits instead of 32) and is different from the 64 bit OS or SW versions, which involve memory addressing instead of calculations.

As a general idea, you can think of it in terms of dividing 10/6 and the result is 1.666666667 instead of 1.67.

But in reality (despite what some people's imagination tells them) 32bit single precision floating point already puts calculation errors far below ever being audible, so whether you turn it on or not doesn't really matter. And it has no effect on whether or not plugins use 32 bits  or 64 bits internally (programmers use what is needed when it is appropriate - 64 bit calculation becomes important for things like recursive processing, but not for simple adding or multiplying).

And the statement that, "Your chances of clipping will be significantly reduced", though technically correct, is particularly humorous in that 32 bits already provides hundreds of dB's of headroom over 0dBFS when in the real world no one ever goes more than a few dB's above 0dB (either by accident or when they're intentionally trying to distort something). But it does show that this is intended to impress people who don't understand math rather than provide any practical benefit. 

arachnaut


A long time ago there was an 80387 Floating Point chip that was added to Pentium systems to do floating point.
This is now called the x87 FPU and there are several of these floating point engines inside our new Intel CPUs.
It doesn't matter if we run the CPU in 64 bit mode or not - the Floating point engines are capable of 32 or 64-bit operations. It's the same chip cell inside.

Since the CPUs can schedule lots of things to run in parallel - integer units, FPU, etc. - the Floating Point width does not matter a whole lot. Check your CPU and if you notice a difference turn it off.

Here are some references about Floating Point math

What Every Computer Scientist Should Know About Floating-Point Arithmetic
http://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.html

Here is a reference for the differences in CPU architecture:

Summary of differences in Intel 32-bit and 64-bit run time models
http://www.arachnaut.net/txt/64bit.html


Intel Manuals:
http://www.intel.com/content/www/us/en/processors/architectures-software-developer-manuals.html



EDIT: added Intel manuals link

spanky


What I don't understand is if the 64 bit floating engine is so good or even better, why even have an option to turn it off or on?

jasonringer
I know this is an older thread, but I thought this info might be useful.
 
I use Guitar Rig 5 pro for all my guitar work, and of course I like to run it as close to real time as possible.
 
With Double Precision on, I was getting crackles occasionally on my lowest latency settings, but when I turned it off, it was perfectly smooth.
 
My system is a not so great AMD quad core build with the Athlon ii X4 631 processor (2.6ghz). and guitar rig seems to not multi thread very well (when monitoring, one cpu would be 50%+ while the others were below 5% most of the time). so for this application, gaining that bit of processing by turning it off was extremely useful, as it may be for anyone who runs vst amp sims or other similar effects.

soens
So IOW the more you mess with something the more accurate it will be?! Interesting. Kinda like multisampling maybe?
Compute it over and over again so much that the inaccuracies take to flight.
... or get bigger?