AMD Close to metal Technology

[Srek]

Welcome back assembler code :)

[CHRiTTeR]

Dont understand much of it, but I do like the stuff I do understand (the eightfold performance increase). :scream: :eek:

Even 3x would be awesome

[Womball]

I almost learned ASM well for my calculator. Really cool but tons of work to do something simple. But really really fast.

[EnlightenedPixel]

So... harder to program, but better results?

[EndOfSewers]

If it's like assembly code, it will be very efficient, as well as nightmarish to write, debug, and maintain. High-level languages were developed for a reason.

[bloodbunnys]

they just need to rename the forum from cg news to off topic.

how ever this is good news maybe we will see it in 5 years and with apps

[Per-Anders]

Are they seriously suggesting AMD chips are better for consumers and will reap massive speed gains... not by the chips being any faster, but by coders simply writing more optimal low level code?

So you could equally put an Intel, or IBM or Motorolla logo on the release saying exactly the same things.

[Fredl]

they just need to rename the forum from cg news to off topic.

It's bizarre that you find this "off topic".

I think it is very interesting, and very relevant to 3D.

[Szos]

So AMD wants to throw away years and years worth of high-level standards, that are in place on purpose to abstract and isolate the hardware?!?

I do need to ask though, what exactly constitues a "Stream" processor? Is that just another extention to the regular x86 architecture - not unlike MMX, or is this something completely different? I recall reading something else about Stream, but it was so full of marketing jargon is was unreadable.

[bloodbunnys]

It's bizarre that you find this "off topic".

I think it is very interesting, and very relevant to 3D.

you and the ort forum leader must be brothers or something always posting stuff that has nothing to do with cg. Amd processors and 3d tv's have nothing to do with cg news there is a tech forum on cgtalk.com

back to this topic, all this speed is great but the software needs to work with it +5 years for that lol. I think people will work in HD and not SD now. so they will be back to being slow again. meh i want to see what 15 years from now will bring

[arquebus]

from Wikipedia:
Stream processing is a relatively new, yet quite successful paradigm to allow parallel processing at never-before-seen efficiency with minimal effort. Compared to existing architectures, stream processors are able to provide up to 20X the performance at the same power dissipation and die size. Given a set of input and output data (streams), the paradigm is essentially based on defining a series of compute-intensive operations (kernel functions) to be applied for each element in the stream. While it seems it would be possible to have multiple kernels in a theoretical world, the uniform streaming paradigm is the only one which has had success. The uniform streaming paradigm uses one kernel function at a time, applied to all the elements of the stream. Kernel functions usually work on streams in a pipelined fashion, where local on-chip memory is reused for input/output streams to minimize external memory bandwidth. Uniform streaming, essentially SIMD, is able to simplify interconnects and get large increases in performance and a simplified programming model allowing development in high-level C and still get optimal performance from the hardware. Another important benefit of Stream processing is since the abstraction of streams and kernels expose data dependencies, it is possible for the compiler tools to fully automate and optimize on-chip management, such as sizing the streams and allocate them, while hardware can use scoreboarding to launch DMAs at runtime (as depedencies are known). In other words, there is no need for hardware caches or manual mangement of DMA, tasks that can consume the majority of a project's time when using conventional DSPs, for instance. The higher efficiency of the on-chip memory ("software managed cache") reduces die area or frees up area for more ALUs.


http://en.wikipedia.org/wiki/Stream_processor

[bloodbunnys]

I like it when they say UPTO 20x

[ffear]

you and the ort forum leader must be brothers or something always posting stuff that has nothing to do with cg. Amd processors and 3d tv's have nothing to do with cg news there is a tech forum on cgtalk.com

You seem to have limited your thinking of 3d DCC to video games and neat little movies. There is a whole industry out there that has been using 3d TV’s for years day to day. We can't escape the fact the DCC takes these devices.



This news in particular (when implemented) can affect someone’s entire workflow as an artist.

[AnharMiah]

Yes, its nice progress, but I'll wait till Quantum Computers[1] are available (may be some time) ;)


[1] http://en.wikipedia.org/wiki/Quantum_computer

[wizzackr]

If we can use the massively parallel architecture of GPUs (or stream processors, if you like to call them that) in order to speed up rendertimes I'm all for it. There seem to be many new ventures into more specialized processors being used for various tasks, recently - AMDs torrenza technology, HTX becoming non-proprietory, now the CloseToMetal papers, the recent announcement of StreamPorcessors being avaiable for development, MentalImages joint-venture with ArtVPS's raytracing co-processors etc.

Speaking of which - is nVidias CUDA technology anything similar (http://www.nvidia.com/object/cuda.html)? drakken pointed me to it the other day... seems similar to me, but then again - I have no clue ;)

[monovich]

I'll bet AMD wrote that wiki entry.

[arvid]

Unfounded negative comments for the win!

yaaaay :banghead:

[cuog]

To those who said this would just mean the use of more ASM type code, I think that you are headed int he wrong direction. From what the initial post stated it seems to me that AMD created a new API that does all of that already, so that programmers can implement the faster low-level code already written in their standard high-level Language.

It also doesn't surprise me that AMD is the one doing this for the 3D platform since their purchase of ATi it has looked like they have been trying to push for better 3D graphics from both the software and hardware side of the equation.

[ThomasLC]

if it's like the old ibm/motorola "altivec" (multiple data, one process), then maybe in 8 years a few games and some photoshop plugins will be 20X faster... wow ;)

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.