Editor's Opinion: NVIDIA Fermi

My opinion of NVIDIA's Fermi architecture has changed over the past several months, as they've developed their graphics processor to fully embody the originally unclear long-term plan. Testing with NVIDIA's GF100 GPU held its own set of challenges, and many times the video cards based on this graphics processor seemed condemned by the inherited legacy of issues. From the flagship GeForce GTX 480 down to the GTX 465, Fermi impressed gamers with strong FPS performance... and that was about it. Thermal output and power consumption were unfashionably high, to which ATI constantly and consistently focused their marketing attacks. Then along comes GF104 on the GeForce GTX 460, a video card that completely changed the game.

NVIDIA's GeForce GTX 460 not only changed the collective opinion regarding their Fermi architecture, it also changed the GPU landscape. ATI held the upper hand by releasing a DirectX-11 video card first, but they've painted themselves into a corner with their Evergreen GPU. Unlike NVIDIA's Fermi architecture, which can shape-shift as desired, ATI's Cedar, Redwood, and Juniper GPUs are all simply slices of the same processor: Cypress. This is where intelligent consumers will spot the flaw: ATI came to the (video) card game and showed their entire hand from the first deal, while NVIDIA had a few spare aces up their sleeves. NVIDIA's GeForce GTX 480 is only 15/16 of the complete GF100 package, and we're just beginning to see what's possible with a 7/8-whole GF104 GPU with GTX 460. It's unknown what NVIDIA has planned for GF102 and GF108... but now we know that GF104 made good on Fermi and GF106 made is supremely affordable.

I don't like to guess at the future, especially since so many other websites exist for this purpose, but something tells me NVIDIA is stock-piling full-blown 512-core GF100 Fermi processors for an upcoming launch. Whether or not we will see GF102 or GF108 GPUs launched this year (2010) is still unclear, but it doesn't appear that AMD has any surprises for the upcoming holiday season. The true potential of NVIDIA's Fermi architecture has still yet to be seen, as in the full 512 CUDA cores. Sure, we've already poked around at the inner workings for our NVIDIA GF100 GPU Fermi Graphics Architecture article, but there's so much more that goes untested. Current Folding@Home clients take advantage of Fermi's architecture and make CPU-based folding obsolete. The difference between work unit performance on the GeForce GTX 400-series surpasses ATI's Radeon HD 5000 series equivalents without much struggle, and CPUs become more like snail-mail compared to e-mail.

NVIDIA GeForce Fermi Graphics Card Family

So now ATI and NVIDIA are even-Steven in the running for DirectX-11, and all that they need are video games to increase demand for their product. This becomes a real problem (for them both) because very few existing games demand any more graphical processing power than games demanded back in 2006. Video cards have certainly gotten bigger and faster, but video games have lacked fresh development. DirectX-10 helped the industry, but every step forward received two steps back because of the dislike for Microsoft's Windows Vista O/S. Introduced with Windows 7 (and also available for Windows Vista with an update), enthusiasts now have DirectX-11 detail and special effects in their video games.

Even if you're only after raw gaming performance and have no real-world interest in CUDA, there's reason to appreciate the GF100 GPU series. New enhancement products, such as the NVIDIA GeForce 3D Vision Gaming Kit, double the demands on frame rate output and hence require more powerful graphics processing. This is where Fermi-based products deliver the performance necessary to enjoy the extended gaming experience. I'm a huge fan of GeForce 3D-Vision, which is why it's earned our Editor's Choice Award, and Fermi delivers the power necessary to drive up to three monitors. The newly dubbed NVIDIA 3D-Vision Surround (stereo) feature requires three 3D-Vision capable LCD, projector, or DLP devices and offers bezel correction support. Alternatively, NVIDIA Surround (non-stereo) supports mixed displays with common resolution/timing.

Some older game titles will also benefit from the Fermi architecture, beyond a simple increase in video frame rates. For example, Far Cry 2 (among others) will receive 32x CSAA functionality native to the game, but future NVIDIA Forceware driver updates could also further add new features into existing co-developed video games. NVIDIA's R260 Forceware release will be among the first to introduce new features enthusiasts have been wanting for quite some time, my favorite is the removal of previous driver files and extensions. Additionally, NVIDIA NEXUS technology brings CPU and GPU code development together in Microsoft Visual Studio 2008 for a shared process timeline. NEXUS also introduces the first hardware-based shader debugger. NVIDIA's GF100-series are the first GPUs to ever offer full C++ support, the programming language of choice among game developers.

Fermi is also the first GPU to support Error Correcting Code (ECC) based protection of data in memory. ECC was requested by GPU computing users to enhance data integrity in high performance computing environments. ECC is a highly desired feature in areas such as medical imaging and large-scale cluster computing. Naturally occurring radiation can cause a bit stored in memory to be altered, resulting in a soft error. ECC technology detects and corrects single-bit soft errors before they affect the system. Fermi's register files, shared memories, L1 caches, L2 cache, and DRAM memory are ECC protected, making it not only the most powerful GPU for HPC applications, but also the most reliable. In addition, Fermi supports industry standards for checking of data during transmission from chip to chip. All NVIDIA GPUs include support for the PCI Express standard for CRC check with retry at the data link layer. Fermi also supports the similar GDDR5 standard for CRC check with retry (aka "EDC") during transmission of data across the memory bus.

<< Previous Page - Next Page >>