[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: data type polymorphism

From: Scott Christley
Subject: Re: data type polymorphism
Date: Mon, 23 Mar 2015 10:43:16 -0500

On Mar 20, 2015, at 7:00 PM, ChanMaxthon <address@hidden> wrote:

> I think you can just unify all calculation into one type, probably float.

Not optimal unfortunately, a main reason is to allow a switch between float and 
double in case the extra precision is needed, if I unify to double then I get a 
memory and performance hit.

> And since you are emulating something using big matrices you may want look 
> into OpenCL as it can help you crunch those numbers on GPU, way faster than 
> your CPU in pretty much all scenarios, and all the extra hardware you need 
> (for a Linux PC) is a moderately-priced gaming graphics card like GeForce GTX 
> 750 Ti.

Interesting that you mention that because that is exactly what I’m doing.  
Though I’m using CUDA versus OpenCL as it seems faster.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]