Last updated on September 11th, 2010 at 04:20 pm
All things being equal, how do you make a product run faster? Easy. You make it simpler. Right now, there is a rumbling in Asia that nVidia will drop CUDA support. KitGuru could hardly believe its ears. nVidia drops CUDA. Now THAT is dramatic. But is it correct?
We all know that increased complexity means more things to go wrong. Even so, chips of all varieties continue to pack in more and more technology each year. Intel is fortunate enough to have massive manufacturing capability and the tick-tock program that means it’s able to prove a process, before it needs to start up a new design.
With graphics, this is a lot more complicated. While TSMC is, undoubtedly one of the most advanced manufacturing operations on the planet, it is still hard to make the latest GPUs work. Launched on 23rd September 2009, ATI is still struggling to satisfy global demand for its Radeon HD 5000 series 8 months later.
Fermi has been a challenge for nVidia. Without doubt, the GTX480 is the fastest GPU ever made, but volumes are still low. Also, as a company, they are under pressure to deliver the all-important mainstream products that will allow for volume sales in a global recession. The GTX470 is not the first product in that segment, but the GTX465 could be. That begs the question, will nVidia deliver Fermi without CUDA support?
For those who don’t know, CUDA is nVidia’s parallel computing architecture (launched February 2007) that enables dramatic increases in computing performance by harnessing the power of the GPU. When we say ‘dramatic’, we’re talking about an increase in video transcoding speeds of around 18x over traditional CPU methods. In briefings and conferences, it has formed an important part of nVidia’s offer.
So, the question remains, how to make the GTX465 sexy and available?
If the rumours are to be believed, then nVidia has one possible response that answers both of these questions at the same time. Make the GTX465 faster and easier to build by removing support for CUDA? Disable it in the driver?
nVidia has made a big deal of CUDA in the past, saying that “computing is evolving from ‘central processing’ on the CPU to ‘co processing’ on the CPU and GPU”. In layman’s terms many consumer video applications are already accelerated by CUDA. Cyberlink, creator of the ubiquitous PowerDVD product, has been integrating CUDA technology since the end of 2008.
While discussing his 3 key strategies for the future at the recent FiRe conference, nVidia’s CEO Jen Hsun Huang put GPGPU (with technologies like CUDA) first on his list.
Another challenge to speed is heat. If nVidia stays with the same PCB as the GTX470, but with a simpler chip, then heat will form less of a problem and we could see some monster clocks.
Whether they would be monster clocks ‘straight from the box’, or somehow supplied through an OC tool, we’re not sure.
One thing is certain, nVidia needs to add spice to Fermi if it wants the dish to be memorable.
If nVidia does drop CUDA to simplify and speed up the GTX465, even on a driver level, then this will be a dramatic statement about the link between processor complexity and pure performance.
Last question. How can nVidia logically drop CUDA support without it looking like a complete embarrassment ?
If true, then KitGuru thinks nVidia’s explanation will be simple and along the lines of, “GTX465 fully supports the OpenCL standard and is a response to customers who have asked for nVidia’s powerful Fermi technology to be made available to the significant audience of users who believe that open source is the future.”
So, there you have it. nVidia can launch a sexy Fermi card into the market, with serious clocks and the possibility for overclocks, without support for CUDA – and it will still look like it was ‘the way it’s meant to be played’.
KitGuru says: We’ll have to wait and see if this is true, but one thing is certain: Our Labs Manager is salivating for some GTX465 action!
Discuss on our forums or leave a quick comment below
EDIT: We have spoken to nVidia and while we have no official quote from them just yet, we will update you if they have anything to add.