Unified architecture and built-in RAM within the graphics core - NVIDIA Reveals its technological roadmap for the next few years and offers quite a few surprises and ambitions
Annual GTC Conference NVIDIA Did not disappoint this year either and provided us with a very intriguing glimpse into the future of the company - in the form of updated roadmaps both for the world of graphic cores and for the world of systems on a chip (the products of theTegra of the company).
So what will the chipmaker's product line look like in two or three years? Let's start with the video cards.
If you believe it To previous reports It seems that 2013 will not bring with it large lines in the world of graphic cores. Kepler 's architecture NVIDIA Will continue to compete with the competitor's GCN AMD, While in the background we will continue to hear about the drop in the volume of sales in the world,PC And prophecies about the near end of the computing world as we know it.
|The future according to NVIDIA|
2014, however, will bring with it a new and up-to-date architecture NVIDIA Which is nicknamed Maxwell We first heard about her several years ago. In the best tradition, the chip developer declares that Maxwell will bring with it a doubling of the processing capabilities (double the amount of floating point calculations exactly double the watt), but that's not all - Maxwell will also present architecture memory Unified.
the meaning? A uniform memory space for both the graphics core and processing cores that will enable them to function together in an efficient, cost-effective and smarter way. How this will helpNVIDIA Not really offering products that combine processing cores and graphics cores right now? Well - the world of Tegra chips is one obvious answer and the processing cores of the mysterious Denver project are also a possible answer, when it is not impossible that a combination of cores made NVIDIA On the chips of other manufacturers is a logical option for the future.
Either way - this is a sensible evolution for NVIDIA and a line-up with AMD, which has been talking for some time about HSA architecture (Heterogeneous System Architecture) that will offer similar capabilities in its future APU processors.
The big innovation of the updated road map is not about Maxwell, but about what will come after it - please get to know Volta, the architecture of the future of NVIDIA.
Similar to the move from Kepler to Maxwell, the move from Maxwell to Volta will also be characterized by doubling performance to watts, but the true story and "winning feature" of future architecture will be something else - chips Memories RAM 3-D to be integrated into the graphics core.
|Not only the next generation: NVIDIA also used the conferenceThe Titan cards New ones
In order to demonstrate breakthrough technologies to create a human face digitally
We will first explain that a combination of RAM on the graphics core chip and not on the printed circuit board (as is customary in graphics cards now) makes it possible to reach bandwidths and significantly better stay times which in turn contribute a lot to better performance - inNVIDIA It is estimated that the Volta architecture can offer speeds memory On the order of terabytes per second, compared to 300 gigabytes per second on the most advanced graphics cards currently.
A combination of large and expensive RAMs on top chip The graphics core is an expensive and not at all simple process - and this is where the XNUMXD chip stack technology comes into play, which will allow the memory chips to be compressed on top of each other in a way that saves valuable space on the semiconductor without compromising speed and access (thanks to technologies such as TSV - Through Silicon Via that enable תקשורת Directly through the "floors" chips).
|In case or not, NVIDIA Used a graphic element that we had seen before Technology
Micron's Hybrid Memory Cube - A hint of future cooperation between the companies within the framework of Volta architecture?
A three-dimensional chip stack is a technology that is still in its infancy but is expected to gain significant momentum in the next few years,NVIDIA Avoid giving an estimate, even the most general one, regarding the date when we will see the Volta architecture coming to market - the chip developer is totally dependent on the development of the relevant production technologies of TSMC And its competitors, and therefore we will not be surprised to discover that the integration of the memories on the graphics core will eventually be postponed to later architecture.
In the end, this is an ambitious and very interesting idea on the part of NVIDIA, but it seems that without turning the necessary technologies into it To the new and dominant industry standard It will be very difficult to implement in the foreseeable future.
What do you think about the future vision of NVIDIA To the graphics core market? Exciting or boring? Tell us what you think about the reactions!