|NVIDIA GeForce GTX 580 Video Card Performance|
|Reviews - Featured Reviews: Video Cards|
|Written by Olin Coles|
|Tuesday, 09 November 2010|
Page 20 of 21
Editor's Opinion: NVIDIA Fermi
My opinion of NVIDIA's Fermi architecture has changed over the past several months since it was first announced, due largely in part to refinements made to their graphics processor. 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 AMD 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. AMD 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, AMD's Cedar, Redwood, and Juniper GPUs are all simply slices of the same processor: Cypress. This is where intelligent consumers will spot the flaw: AMD 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. Now that AMD has unwrapped their Barts (Cypress refresh) GPU, NVIDIA returns to offer all 512 Fermi cores in the GF110 GPU.
Now that NVIDIA has finally cooked Fermi with the perfect blend of tessellation, shaders, and texture units, consumers are able to see what they've been missing. We poked around the inner workings for our NVIDIA GF100 GPU Fermi Graphics Architecture article, but the modular nature of this processor left a lot of uncertainty at the time. With AMD soon to launch their own counter-attack with the Caymen-equipped Radeon HD 6970, it's unclear just how long NVIDIA will keep it's position on the throne of discrete graphics. AMD proved that Barts can do more with less, just like NVIDIA has done with their reworked Fermi architecture, and so now it becomes a matter of finding the acceptable price point for the segment. Of course, this all depends on yield... something both vendors have struggled with as they continue to depend on TSMC for results.
NVIDIA GeForce Fermi Graphics Card Family
With Barts already on the shelf and GTX 580 arriving this week, AMD and NVIDIA are once again even-Steven in their competition for DirectX-11 supremacy. Now all that they need are some highly anticipated video games to showcase tessellation like never before and increase demand for their products as a result. Titles such as Tom Clancy's H.A.W.X.2 certainly showcase a beautifully realistic terrain, but low-end graphical processing requirements don't exactly help to sell the high-performance video cards. It's the games like Lost Planet 2 that will crush performance and push sales for top-end products, but the interest in this genre isn's nearly as strong as FPS titles such as Call of Duty: Black Ops or BattleField: Bad Company 2. Unfortunately for AMD and NVIDIA, most game developers are working to the hardware available inside console gaming systems, and not the more powerful desktop computers.
Even if you're only after raw gaming performance and have no real-world interest in CUDA, there's reason to appreciate the Fermi GF100-series GPU. New experience-enhancing products such as the NVIDIA GeForce 3D Vision double the demands on frame rate output and require more powerful graphics processing power as a direct result. This is where Fermi-based products deliver the performance necessary to enjoy the extended gaming experience. I'm a huge fan of the 3D experience, which is why 3D Vision earned our Editor's Choice Award and I've written a NVIDIA 3D-Vision Multimedia Resource Guide, and at the moment only Fermi-based GeForce video cards deliver the power necessary to drive up to three monitors for 3D-Vision Surround.
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 introduced 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.