Incremental, gradual improvements to graphics performance through driver development isn't exactly a new process. GPU manufacturers now view their products as evolving hardware and software packages, allowing post-release improvements to continue to retain and exploit competitive advantages, or just add value for the consumer. Generally however raw performance improvements have tended to be on the order of a few percentage points, except in a few niche titles or system configurations where the benefits might break into double-figures.
NVIDIA in particular have been known to be past masters these of these updates, pushing GFX card competitiveness long after one might have thought it would obsolete whilst also timing new drivers to the release of the latest blockbuster game. AMD have also dabbled over the years to mixed success, with recent efforts being particularly noteworthy. However today NVIDIA are planning to release something a little extraordinary that could herald a watershed moment for DirectX gaming: the 337.50 beta driver.
The Green Team aren't claiming improvements on the order of 5-10% across a range of carefully selected titles this time around; the 337.50 update is instead set to increase gaming performance by as much as 64% in Single GPU and 71% in Dual-GPU configurations, across a wide range of titles. Such scaling is an order of magnitude more than one would expect from a driver update, and is almost never seen at the driver level unless it's a revision to correct a previous bug or optimise for new hardware. Such is apparently not the case for the 337.50's, which applies to Maxwell, Kepler, and even the now end-of-line Fermi series of GPUs.
How are NVIDIA able to do this? Well, for once they're being uncharacteristically tight-lipped; however one route they're specifically not going down is a propriety API. NVIDIA have been critical of AMD's proprietary Mantle API in the past, although they've certainly embraced proprietary technologies in the past when it's been to their benefit. Their argument against it stretches from building a renderer which partially or fully bypasses the major pathways for that platform - in this case DirectX - which in tern increases game development workload and is generally hardware specific. Somewhat proving their point is that Mantle continues to undergo teething problems and is only really of benefit to three GPUs in AMD entire desktop and mobile product stack.
NVIDIA are now aiming to show that performance tweaks resulting in improvements of the magnitude conferred by Mantle can be implemented at the driver level. The new driver set works within the framework of DirectX 11 and is supported by all of NVIDIA's DirectX 11 GPU families, including the 400- and 500-series 'Fermi' based GPUs, as well as being supported by the vast majority of DX9-11 titles without any extra developer tweaks. In an ideal world therefore this would be an infinitely superior solution.
However there are a few lingering bones of contention which will bear further scrutiny. Unfortunately historic claims of major breaks in GPU processing efficiency have at times relied on shortcutting, either by using approximate techniques or methods which serve to reduce quality (even if it may not be at an easily perceptible level. NVIDIA are adamant that there are absolutely no changes to image quality, merely that optimisations and efficiency gains have been made following months of hard word, but it's important to note that they haven't in any way broken down what these improvements entail. Given the duplicitous nature of the industry in the past expect the press corps to be somewhat cynical in their approach.
Furthermore, NVIDIA are not idle in their comparisons, they're certainly targeting AMD's Mantle with this release. It's possible that they may have been spooked by the way the proprietary API has to a certain extent dominated the news cycle (for good or ill), even over the recent announcement of DirectX 12's 'Close To Metal' approach which is still a year away, but they've chosen to make direct comparisons between the R9 290X /w Mantle and their own GTX 780Ti /w 337.50. Critical to note however us that NVIDIA quote average frame rates, whilst Mantle is designed to impact minimum frame rates due to changes in the rendering pipeline. Given raw disparities between the GPUs therefore it may be a little disingenuous of a comparison, even if it's always cute to beat the competition with their own benchmark.
Even though reservations are expected, and should be welcome to ensure a healthy debate, for the consumer the release of these drivers should be a win. Those with NVIDIA cards will be getting an appreciable performance boost at no cost, likely allowing a tweak upwards of image quality setting or more stably holding at 60FPS, whereas those with AMD hardware may find that a fire has been lit under AMD to realise the benefits of Mantle sooner rather than later.
Today's driver release is likely to be just the first salvo in an ongoing battle between NVIDIA and AMD over APIs. Whilst at one stage they may have been complacent in their belief in both hardware and driver superiority, competition heat up in October and hasn't been allowed to cool since. NVIDIA are banking on their own team of software engineers and collaboration with Microsoft on the hardware and development agnostic DirectX 12 to provide them with a competitive advantage, whereas AMD are seeking to utilise Mantle alongside DirectX and tie in game developers to the process along the way.
Unfortunately for AMD, NVIDIA are forthright in their support for DirectX 12 even through legacy hardware (back as far as the 400-series Fermi generation), whereas AMD are limited to GCN-based hardware. If you believe that DirectX 12 and Mantle are competitors then DirectX 12 already has an insurmountable lead in install base, and it further drives home the fact that NVIDIA may well have a built-in user base advantage for the DirectX 12 generation of games.
For AMD's part DICE and Square Enix are already substantial Mantle partners, and it's unlikely that they would have gone down this path it if was solely as a means to tailor themselves to one hardware vendor. In fact, they may be viewing Mantle as a small stepping stone to DirectX 12, or at the very least an environment most suited to both PCs and consoles rather than excluding the PS3. Either way, the coming year is set to be an interesting time for them.
You will be able to download the 337.50 Beta Drivers via GeForce Experience or the official website later today, with the full WHQL release due later this month. You can read a detailed run-down of the the improvements in their blog post on the subject.