Hi All,
Check out this post by Aleksey on the Voxengo Forum:
http://www.voxengo.com/forum/ar/1522/?offset=10
From my investigation into CUDA his conclusions so far look correct. It is more of a \"Co-Processor\" than a Multi-Core CPU. It doesn't have enough of the support hardware to function as a true CPU. However, as a replacement for the UAD-1 IT IS ABSOULUTELY PERFECT!!!!
The UAD-1 is not running your computer and DAW it is just processing the specific UA plug-ins. This is the perfect scenario for CUDA. I mean DAMN, Aleksey was able to run 384 CHANNELS, each with 25 Bands of EQ, at FREAKIN 192kHz!!! And he has the SLOWER CARD (8800GTS vs. 8800GTX). And that's not even SLI! That is only ONE card.
What's even better about CUDA/NVidia is that the VIDEO Rendering part IS ON A TOTALLY SEPERATE CHIP! If you look at a 8800 Video card, you will see a smaller chip located closer to the DVI Connectors. That is the actual VIDEO Display chip, the main chip in the center of the card is the G80 GPU. You can OEM the card WITHOUT the Video Display chip and you have basically a UAD-1 type card.
Remember, this is NOT the same as BionicFX which was using an ATI Video card with the typical setup of Dedicated Pixel, Vertex, or Geometry shaders. All of your code had to be CONVERTED to OpenGL to be processed on one of the Shaders and then converted back in order to process anything. With NVidia/CUDA, the G80 is the total opposite. The G80 has 128 cores that are GENERAL PURPOSE STREAM PROCESSORS. Their processing function isn't predetermined. They can be programmed to process any type of IEEE-754 Floating Point instruction without conversion. NVidia has even released an API, SDK and compiler (C or otherwise) for developers. Can it get any easier?
Imagine, a processor with 128 individual cores running at 1.35GHz (the stream processors run faster than the rest of the chip @575MHz), with 768MB of GDDR4 Ram, a 384-bit Memory Interface, with a dedicated Thread Processor managing THOUSANDS of threads at once, just to run UA Plug-ins. And the BEST part of all of this? You can BUY an NVIDIA 8800GTX for $600.00!!!!!!!!!!!!!!!! Even if UA decided to sell if for $1,000.00 it would be a bargain for us and they would make a profit.
What about processing power? Well two of these cards in an SLI configuration are capable of a more than a TERAFLOP (that's one TRILLION INSTRUCTIONS PER SECOND). Anyone else ready to jump on this bandwagon?
Check out this post by Aleksey on the Voxengo Forum:
http://www.voxengo.com/forum/ar/1522/?offset=10
From my investigation into CUDA his conclusions so far look correct. It is more of a \"Co-Processor\" than a Multi-Core CPU. It doesn't have enough of the support hardware to function as a true CPU. However, as a replacement for the UAD-1 IT IS ABSOULUTELY PERFECT!!!!
The UAD-1 is not running your computer and DAW it is just processing the specific UA plug-ins. This is the perfect scenario for CUDA. I mean DAMN, Aleksey was able to run 384 CHANNELS, each with 25 Bands of EQ, at FREAKIN 192kHz!!! And he has the SLOWER CARD (8800GTS vs. 8800GTX). And that's not even SLI! That is only ONE card.
What's even better about CUDA/NVidia is that the VIDEO Rendering part IS ON A TOTALLY SEPERATE CHIP! If you look at a 8800 Video card, you will see a smaller chip located closer to the DVI Connectors. That is the actual VIDEO Display chip, the main chip in the center of the card is the G80 GPU. You can OEM the card WITHOUT the Video Display chip and you have basically a UAD-1 type card.
Remember, this is NOT the same as BionicFX which was using an ATI Video card with the typical setup of Dedicated Pixel, Vertex, or Geometry shaders. All of your code had to be CONVERTED to OpenGL to be processed on one of the Shaders and then converted back in order to process anything. With NVidia/CUDA, the G80 is the total opposite. The G80 has 128 cores that are GENERAL PURPOSE STREAM PROCESSORS. Their processing function isn't predetermined. They can be programmed to process any type of IEEE-754 Floating Point instruction without conversion. NVidia has even released an API, SDK and compiler (C or otherwise) for developers. Can it get any easier?
Imagine, a processor with 128 individual cores running at 1.35GHz (the stream processors run faster than the rest of the chip @575MHz), with 768MB of GDDR4 Ram, a 384-bit Memory Interface, with a dedicated Thread Processor managing THOUSANDS of threads at once, just to run UA Plug-ins. And the BEST part of all of this? You can BUY an NVIDIA 8800GTX for $600.00!!!!!!!!!!!!!!!! Even if UA decided to sell if for $1,000.00 it would be a bargain for us and they would make a profit.
What about processing power? Well two of these cards in an SLI configuration are capable of a more than a TERAFLOP (that's one TRILLION INSTRUCTIONS PER SECOND). Anyone else ready to jump on this bandwagon?