IEEE - Who?

[K.I.L.E.R]

Well, I searched Google and I came up with nought.

AFAIK IEEE are a company with some dudes who set the FP32 standard 16 years ago.
That's all I know about them.

Can anyone give me a larger/deeper background into them?

Thanks

 

[epicstruggle]

from ieee.org:

The IEEE (Eye-triple-E) is a non-profit, technical professional association of more than 380,000 individual members in 150 countries. The full name is the Institute of Electrical and Electronics Engineers, Inc., although the organization is most popularly known and referred
to by the letters I-E-E-E.

Through its members, the IEEE is a leading authority in technical areas ranging from computer engineering, biomedical technology and telecommunications, to electric power, aerospace and consumer electronics, among others.

Through its technical publishing, conferences and consensus-based standards activities, the IEEE:

*produces 30 percent of the world's published literature in electrical engineering, computers and control technology,
*holds annually more than 300 major conferences and
*has nearly 900 active standards with 700 under development.

 

[g__day]

For Universities the mantra was always "Publish or Perish". At Sydney UNiversity IEEE was always considered the Mecca of scientific magazines to get your technology based thesis or paper published in for a long, long time.

 

[Rookie]

U can consider IEEE as a international standard organizer.

 

[K.I.L.E.R]

Ahh. Thanks for the info guys.

 

[Simon F]

Well, I searched Google and I came up with nought.

Finding this a bit hard to believe, I put IEEE into Google and hit "I'm feeling lucky". It took me directly to http://www.ieee.org etc. Are you sure you looked?


Anyway, for the field of Computer Graphics, IEEE publish the journal "IEEE Computer Graphics and Applications", that often has some useful papers.

 

[K.I.L.E.R]

Well, I searched Google and I came up with nought.

Finding this a bit hard to believe, I put IEEE into Google and hit "I'm feeling lucky". It took me directly to http://www.ieee.org etc. Are you sure you looked?


Anyway, for the field of Computer Graphics, IEEE publish the journal "IEEE Computer Graphics and Applications", that often has some useful papers.

Yeh, I 'cked up my search though by typing in "FP32" after IEEE.

 

[MfA]

Adding "single precision floating point" will probably give more meaningfull hits, if you are interested in the actual standard.

 

[Hellbinder]

Soo.. Why is Tim Sweeney going out of his way to talk about IEEE 32 standards when they have Zilch to do with either one of todays Big API's??

You dont write a game to IEEE-32. You Write a game to D3D or OpenGL or Glide. Which in turn either line up with the standard or dont. That does not change the fact that the Code is written in D3D. No one is writting IEEE-32 code..

If you are going to comment on sayng our engine will be written to X standard. Why not actually talk about the actual standard in API that will likely be targeted. Direct X 10

 

[DemoCoder]

Because everyone who learns programming and how floating point works learns the IEEE standard, and because that standard dominates all data stored as floating point, and because API calls in OpenGL that take floats calls are IEEE single or double precision floats, not FP24's. Tim is just saying that in the far future, people won't have to worry if the 3D card is using a special lower precision format.

The idea is, if I send a 32-bit IEEE number through the API, and I get it in the shader, I should not get a truncated number IF I ASKED FOR 32-bit


Let me guess: you don't like the implication that in the future, if you accept Tim's comments, ATI will have to support FP32 in the pixel shaders. I don't see why you are whining about this, since ATI already supports FP32 in the vertex shaders, and it will only be a small modification to their architecture (given more transistors) to get FP32.


Tim's comment isn't anti ATI. It's merely a reflection of the fact that programmers don't want to learn multiple FP formats, and since the C programming language, the CPU, and the APIs all deal with floats in the IEEE format, then the HW pixel shaders should not drop precision as soon as they are capable of running FP32 at full speed if the programmer requests 32-bit IEEE semantics.

 

[banksie]

Soo.. Why is Tim Sweeney going out of his way to talk about IEEE 32 standards when they have Zilch to do with either one of todays Big API's??

At a guess because he probably wants data portability between the rendering hardware and the rest of his program. Using a standard floating point format allows you to do a lot of funky procedural generation beyond what pixel/vertex shaders alone are able to do - be that because of architecture limitations or simply of processing time available on the GPU.

At the very least it would allow load balancing between the CPU and the GPU to let the engine potentially render more than it could relying purely on the graphics card alone. Or seeing as you can do particle effects using vertex shaders how about feeding the resultant vertex data back to the main engine to allow collision feedback to the main engine? (so a shader generated particle stream could have propulsive force in the game engine say.)

 

[K.I.L.E.R]

Adding "single precision floating point" will probably give more meaningfull hits, if you are interested in the actual standard.

THANKS!

http://www.psc.edu/general/software/packages/ieee/ieee.html

Got it in the first hit.

 

[MfA]

Converting between floating point formats is hardly rocket science, and hardware can do it virtually for free, and feedback always sucks ass ... it is just such a bad idea that adding a conversion step hardly makes it worse.

BTW, floating point is for wusses who dont understand how error accumulates in their algorithms

 

[K.I.L.E.R]

Converting between floating point formats is hardly rocket science, and hardware can do it virtually for free, and feedback always sucks ass ... it is just such a bad idea that adding a conversion step hardly makes it worse.

BTW, floating point is for wusses who dont understand how error accumulates in their algorithms

Yeh, floats blow.
Instead use double floats.