Why don't I read the articles? - page 5

New comment
 
Reshetov:
Who knows?
There are enough theorists around here as it is.
 
TheXpert:
There are enough theorists here as it is.

К. Tsiolkovsky never went into space, but without his calculations, many others would not have been able to go there for a long time.

С. Korolev has never flown in rockets of his own design either.

 

I figured the only one who knows more about this topic than anyone else isMathemat.

If you're interested, you can write about it together, discussing all the details through the profile beforehand, and then it's up to you.

 
TheXpert: Imho, the level of pitch you set the level of interlocutor. Is there a sense to lower?

It makes sense, it doesn't.

Reshetov: I have OpenCL only on the CPU. I don't see any productivity gains, and spending money on hardware and time on tethering it makes no sense. Whenever I feel the need, i.e. when an application task comes up and it cannot be solved without OpenCL, I will definitely get down to it. But for the sake of speed of drawing fractals, to measure with someone or something - it's not my path.

We are not even talking about fractals, it was only an illustration of the meta-quotes themselves.

I had a CPU-only one myself in the beginning. There are testimonies in the OpenCL thread. MetaDriver's examples inspired me, as even on dual core Pentium G840 I had up to 80x acceleration on his examples. So I decided to buy an old, but rather powerful card - Radeon HD 4870 - for cheap. The results were worth it. However, the system unit regularly howls, as it is too weak, it hardly pulls it :)

All you need to do is to choose the right hardware. Decent modern cards are quite enough. And then it's up to your own brains.

GKS: If you are interested,

If this is a reference to me, then yes, it is interesting - but not so much as to give up everything and do only that. I will write to you when I have enough time and thoughts.

 

Mathemat:

It's enough to choose the right hardware. Today's decent cards are enough. And the rest is up to your own brains.

It is enough to choose or refine the right algorithm for the task, so that you don't have to choose the hardware later.
 
Reshetov: It is enough to choose or refine the right algorithm for the task, so that you don't have to choose the hardware later.
The good thing about good hardware is that it can still count fast if you want it to. Even a 4870 that only supports OpenCL 1.0 is not much worse than a good modern 6930. And even at double precision...
 

Если это обращение ко мне, то да, интересно - но не настолько, чтобы бросить все и заняться только этим. Я Вам напишу, когда мысли оформятся и время появится достаточное.

Yes it is for you. I'll start preparing for this event. In the meantime, I'll be thinking about how to make it all happen.
 
Mathemat:
The good thing about good hardware is that it can still count fast if you want it to. Even a 4870 that only supports OpenCL 1.0 is not much worse than a modern 6930. And even at double precision...
What's the point? Where are the tasks for them? Just don't suggest fractals or MetaDriver tests. I warned you, it's not my business to measure speed for the sake of speed.
 
GKS:
Yes it is for you. I'll start preparing for this event. In the meantime, I'll be thinking about how to provide it all.
Block diagrams. Nothing better, for those who don't know programming languages, has been invented yet.
 
Reshetov:
What's the use? Where are the tasks for them? Don't suggest fractals or MetaDriver's tests. I warned that measuring speed for the sake of speed is not my domain.

Don't worry, Yura, I'm not interested in fractals myself.

There is one voluminous task, but I have not yet thought too much about turning it into a quick calculation in OpenCL. It is not an easy task, it will not be easy to solve.

123456789101112...18
New comment