General Processing Units and Technology

The rumor mill has been busy over the last few days. Not only speculating about the Xbox‘ and Playstation’s next installments (and sometimes simply reurgitating blatant fakes), but also about Nvidia’s monster GPU GK110 that’s supposedly being productized in february 2013 as some form of Geforce.

The 7.1 billion transistor chip was introduced at last years‘ GPU Technology Conference on Nvidias home turf. It was being touted as the next best thing next to sliced bread and first productized as Tesla K20 and Tesla K20X accelerator cards at ISSC in November 2012. Conveniently, it was by then already being put to use in the world’s fastest supercomputer, ORNL’s Titan.

Speaking of Titan, the rumor mill also has it, that the card that’s gonna be launched in the next few weeks (if correct) also carries this denominator: Geforce Titan – whether this goes with a number (I’d like „Titan 1“) or not or if it’s true at all, is still shrouded in the crystal balls of the fortune tellers. What is equally unknown outside Nvidia is the exact configuration which Titan 1 will be carrying and in this article I am going to explore a few variables and give reason as to why I think Geforce Titan will (or should) be configured this way or another.
...

As Creative Technology, Singapore-based parent company of ZiiLABS announced on November 19th, they expect to close a deal with chipdesigner and -maker Intel Corporation in the current quarter ending December 2012. The subject of the deal will be various technologies and patents from ZiiLABS as well as engineering ressources from their UK-based headquarter. Intel will pay 50 million US-Dollar altogether and acquire non-disclosed engineering ressources of the UK sudsidiary of Creative Technology.
...

Intel has a less than stellar record when it comes to driver support for anything that is not an inf-update for their mainboard chipsets. After being established by industry leaders like Apple, Nvidia and AMD, Intel could not longer ignore the industry standard OpenCL - the open compute language. More or less reluctantly, they put out driver support for their CPUs, mostly functional, but quite underwhelming when it came to performance.

One popular example for how enthusiastic Intel was about performance optimized OpenCL drivers could readily be seen by running Luxmark for example. This raytracing benchmark utilizes OpenCL-enabled processors, no matter what kind of beast they are. When you're using a recent Radeon graphics card in your system, chances are that you already have an OpenCL-1.2-enabled driver for your central processor installed - the one that comes with AMDs Catalyst driver package.
...

Modern processors, be they of CPU, GPU, APU or SoC flavour, employ very sophisticated power management techniques in order to better leverage the full potential of their respective functional units in combination with the applied thermal solution. What sounds quite basic is indeed a multi-dimensional topic and can easily mess with any performance analysis that is not taking into account the specifics in each processor. In this article I am giving an outline what contributes to the power management in modern processors.

A while ago - in IT-time quite an eternity - processors had a certain clock speed which they maintained during all operations. They had a fixed TDP that defined the necessary cooling solutions' capabilities to the fraction of a watt. Of course, power was quite low compared to more recent devices in the desktop space. Mobile devices, being much more dependant on a low power consumption, employed simple techniques like a reduced state for clock speed and core voltage as early as the turn of the millenium. With more recent devices, power has become a much more important issue in more than one respect - even in the desktop space. ...

AMD's recent launch of it's latest and greatest, the Radeon HD 7970, despite taking over the single-gpu performance crown, left quite a few people wondering. Wondering, if that was all; wondering, if there might be performance left on the table somewhere in the drivers; wondering, if the focus on compute did cost to much gaming performance. This was especially true when comparing the number of transistors to it's predecessor Cayman, which already boasted an impressive 2.64 billion little transistors. Tahiti's number was far more impressive though and with 4312 million transistors broke the 4 billion mark. Now, a possible explanation has surfaced in a photograph of what looks like a product catalogue from AMD partner Sapphire.
Update Jan 3rd, 2012 21:45: According to a story over at Brightsidenofnews.com, Chris Hook, currently „Head of WW Social Media and Ecosystem Communications” at AMD flat out denied, saying

There are no hidden cores…

...

AMDs recently launched Radeon HD 7970 brought the first installment of „Graphics Core Next“ in the shape of a 4.312 billion transistor processor code named Tahiti. With Graphics Core Next (or GCN for short) AMD has taken a radically different approach to compute than with their former Very Long Instruction Word (VLIW) based micro-architectures. The emphasize not only in the last sentence is on compute because that's the area were the new architecture is set to flex it's ALU muscles first and foremost - for graphics only, AMD remained adamant, VLIW was (and is) a very efficient way of cramping highly potent circuits on a small amount of die space.

Gaming performance of the Radeon HD 7970 is somewhere between 25 tp 50 percent up from HD 6970 and somewhere from 10 to 30 percent above the former single GPU champ Geforce GTX 580 - no matter if 1,5 or 3 GiB. With that settled, let's take a look at Tahiti in it's XT version and dig a little deeper.
...