Watch Out for CrossFire!
Ever since NVIDIA successfully demonstrated their Scalable Link Interface (SLI) technology, ATI had been touting of a similar technology in the works and would even surpass the standard set by its competitor. Formerly referred by ATI as Multi VPU technology (MVPU), the big red kept shielding themselves for countless months my mentioning this moniker whenever there were queries from the press of SLI-like solutions. Half a year after NVIDIA's inception of SLI technology, ATI and its close vendor partners showcased at Computex 2005 (held in June '05) for the first time a full working solution of the CrossFire platform, the official name of MVPU technology. A select few privileged media such as www.hardwarezone.com even managed to perform our own tests to gauge for ourselves firsthand of CrossFire's potential (as published here
). Our initial run in with CrossFire was a positive one, though it still left some doubts such as the flexibility it boasted.
Months have passed with numerous changes on official launch dates and finally, we are quite certain that ATI's CrossFire technology will come to pass soon enough as several tech media such as ourselves have obtained a reference test kit to fully assess the technology and the platform. It's also a very fitting timeframe for CrossFire to debut with the imminent launch of ATI's next generation X1800 series graphics cards.
Cross Who, Fire What?
In case you are a little lost at all these new acronyms and what they are for, let us bring you up to speed. To better understand, we'll be drawing close analogies to how the CPU market has been progressing. If you haven't noticed yet, CPUs haven't really been scaling up the MHz wall for a while now and that's due to the state in which the materials and process technologies are progressing at the moment. Instead, CPUs have been scaling in terms of feature specifications such as increased secondary level cache size, 64-bit memory addressing, 64-bit processing, multimedia specific registers and other enhancements that would bring about better performance when applications are designed or updated to recognize them. Likewise, graphics processing units have long adhered to this path in bringing about more features rather than just brute frequency increments. Besides, GPUs have less real estate to tinker with cooling solutions as opposed to CPUs.
Recently mainstream CPUs have been featuring dual cores two processing brains in one chip. Again, the industry believed that they could scale overall system performance better and at a reasonable cost than if they were to continue building more and more complex single core processors. The graphics industry too sought a method to instill this idea to the graphics card. Build dual core GPUs? Not exactly since GPUs are already far more complex and transistor hungry than CPUs, hence taming such a dual core GPU (if there ever was one), would have been a nightmare given the typical graphics card form factor and mounting characteristics. In addition, GPUs are already featuring multiple pipelines as compared to most CPUs with only a single pipeline. Therefore, today's GPUs are somewhat already multi-core in nature.
Anyway, the enabling factor that came just in time was the PCI Express interface. The speedy, lean PCI Express Graphics interface had 16 times more bandwidth than the already bandwidth plentiful standard PCI Express interface. Since a graphics card alone doesn't hog this giant pipeline, NVIDIA saw an opportunity to wring out a new usage model to split the single large PCIe graphics channel to be shared between two slots and still offer plenty enough bandwidth without sacrificing performance. Thus began NVIDIA's Scalable Link Interface (SLI) technology that allowed anyone who bought an SLI ready motherboard, to toss in two identical SLI certified graphics cards to double the graphics crunching power. Of course, the graphics cards must posses some multi-GPU logic to enable this and those that possess this ability are identified by their SLI certification logo on the box.
ATI's CrossFire is exactly the same concept as NVIDIA's SLI. Albeit nearly three-quarter of a year late, ATI is finally offering something similar, but they promised more frills such as better performance scaling algorithms, enhanced image quality rendering modes, improved multi-GPU compatibility and most significantly, the flexibility to mix-and-match different ATI graphics cards. Arguably, these features do make it more difficult to engineer and more time consuming to validate. This is perhaps the very reason why CrossFire took so long to mature even though we've all seen it in action months ago.
By now you would have had a recap of the multi-GPU scene from both NVIDIA and ATI turfs, so without further ado, we detail you more specifics of the CrossFire technology, how it works, its requirements, how to set one up and finally what you can expect out of it.