DescentBB

Yes, I am on a news kick today. Deal with it.

September 2, 2004

BionicFX announced a technology for music production that turns NVIDIA video cards into audio effects processors. Audio Video Exchange (AVEX) converts digital audio into graphics data, and then performs effect calculations using the 3D architecture of the GPU. The latest video cards from NVIDIA are capable of more than 40 gigaflops of processing power compared to less than 6 gigaflops on Intel and AMD CPUs.

http://news.harmony-central.com/Newp/20 ... -AVEX.html

Wonder if they're going to do the same with ATi.

geez... another 5 years and these guys are going to be building thier own systems

Dont they call that SoC? "System on a Chip"

I wonder if you could do that with any kind of data... you could build something that acted almost like a dual processor.

Duper wrote:geez... another 5 years and these guys are going to be building thier own systems

I doubt it. Graphics processing is fast because there is a small genre of calculations that it needs to perform. That and calculations can be approximated for speed, where a generic CPU must support things like memory management, interrupts, etc exactly. If they can get C++ code to run on a NVidia chip faster than on an Intel, then maybe they'd be in business.

Its CISC vs RISC. RISC is good for specialized tasks like Vid cards, sound processing, networking devices, etc, while CISC stuff can do anything and is basically Jack-Of-All-Trades, Master of None.

The whole reason your nVidia GPU can crank out 100FPS in some *.FPS game is because it was designed, specifically, to do so. CISC chips such as Athlons and Pentiums are designed to run pretty much any x86 code and therefore cannot afford to be specialized.

Topher wrote:

Duper wrote:geez... another 5 years and these guys are going to be building thier own systems

I doubt it. Graphics processing is fast because there is a small genre of calculations that it needs to perform. That and calculations can be approximated for speed, where a generic CPU must support things like memory management, interrupts, etc exactly. If they can get C++ code to run on a NVidia chip faster than on an Intel, then maybe they'd be in business.

Actually, I was only being facetious.

As an electronic musician and owner of an online music store, I am drooling.

Maybe someone will start building MB's with 2 AGP slots. Then I can start buying 2 $400 gfx cards per box.

That's what PCIx is for You can do exactly that with some of the new PCI express boards.

Avder wrote:Its CISC vs RISC. RISC is good for specialized tasks like Vid cards, sound processing, networking devices, etc, while CISC stuff can do anything and is basically Jack-Of-All-Trades, Master of None.

The whole reason your nVidia GPU can crank out 100FPS in some *.FPS game is because it was designed, specifically, to do so. CISC chips such as Athlons and Pentiums are designed to run pretty much any x86 code and therefore cannot afford to be specialized.

Wrong. All modern desktop CPUs are RISC chips, on Wintel platforms they include decoders that convert from x86 CISC instructions to their RISC equivilants.

Wheather a Chip uses CISC or RISC has no bering on the degree of specialization.

video chipsets lend themselves very well to vector and matrix maths. So they are ideal code-breaking tools.

I'm often wondering why my VPU sits there doing essentially nothing for 99% of the time.

gah, FINALLY!!!

i've been waiting for this to catch on every since i heard the geforce2 (yes way back then) line of chips were *programmable.

there were some early attempts at playing around with this, i saw some people doing simple things with it, simple math calculations and whatnot.

i can hardly wait until someone writes a physics engine that can be simmered down into graphics manipulation techniques, and fed into a programmable videocard (well... then it'd be called a physics-card wouldn't it).

*there's a better lingo term for this.

fliptw wrote:Wrong. All modern desktop CPUs are RISC chips, on Wintel platforms they include decoders that convert from x86 CISC instructions to their RISC equivilants.

Wheather a Chip uses CISC or RISC has no bering on the degree of specialization.

Excuse me, but all AMD and Intel processors are CISC. What do you think MMX, MMX2, SSE, SSE2 are if not additional, specialised instruction sets?

You are referring to the possibility that all these instructions are micro-coded in terms of a reduced instruction set, but that isn't the case. The best proof is that some of the multimedia instructions require special registers (e.g. 96 or 128bit wide) which are definitely implemented in hardware in the processor.

The only desktop processors that are RISC are those found in Apple computers.

You are correct, of course, when you say that RISC or CISC does not determine the field of application of a processor. However, it does indirectly. CISC is slower than RISC (especially so if the instructions are microcoded on a lower level). So RISC is the natural choice when power is important and the field of application is narrow.