The truth about SLI

1511 0

When 3D graphics specialist Nvidia announced Quadro SLI last year it promised ‘never-before-seen performance’. Twelve months down the line the majority of CAD users are still waiting says Greg Corke.

A recent trend in computing is to double up the available hardware – typically if one component isn’t quick enough. Intel and AMD are both at it with dual core processors, and two hard drives can be used together to maximise read/write speeds. Two are always better than one, right? Well not always.

How does SLI work?
Quadro SLI works in one of two ways: SFR or split-frame rendering splits the screen horizontally, and each card renders a portion of the screen, depending on the load. The second mode is AFR, or alternate-frame rendering, where each card renders a frame in turn.

For the past 12 months we’ve been waiting for Quadro SLI, the professional dual graphics card technology from Nvidia, which has been severely delayed. Using specially designed motherboards with two PCI Express x16 slots, SLI takes two high-performance Quadro FX graphics cards, and joins them together to boost performance under 3D applications. However, the reality is that not all 3D applications will benefit from SLI – and some will even run more slowly.

A background

Nvidia currently has two separate SLI technologies: GeForce SLI and Quadro SLI. GeForce SLI is designed for the consumer market to accelerate frame rates in 3D games. The technology has sold well since its launch last year and has been shown to provide impressive gains in performance under a number of 3D games. Quadro SLI, on the other hand, was developed to boost performance in professional 3D workstation applications. It was originally scheduled for launch in October 2004, but we’re only starting to see these systems appear now. Indeed HP, Nvidia’s official launch partner for Quadro SLI, has only just started shipping SLI in its Opteron-based xw9300 workstation. So, why has there been such a long delay with the technology, when GeForce SLI has been enthralling high-end gamers for over six months?

In short, Nvidia has experienced a lot of trouble getting SLI to work with OpenGL, the programming interface used by virtually all CAD applications to produce 3D graphics. This is a different technology to Microsoft’s Direct X, which is used in the majority of games. Unfortunately the problems with OpenGL are manifold and much bigger than Nvidia first imagined, somewhat naively by its own admission.

The problems with Quadro SLI

Unlike games, which run in full screen mode, OpenGL CAD applications run in windows. These can be resized and re-positioned, and multiple viewports are common in many applications. Configuring SLI for this constantly changing environment has given Nvidia many technological issues to overcome, but the fact that the CPU limits the 3D performance in many CAD applications is giving the company even bigger headaches. Indeed, Nvidia says that many of the workstation applications, especially in the entry and mid-range CAD market are primarily CPU bound, and there’s no getting around that!


Inventor, Autodesk’s core solution for mechanical engineers, is a prime example of one of these applications. Over the years we’ve seen little difference in Inventor when using a ú200 graphics card or one that costs you ú2,000. Therefore, chucking more graphics horsepower at the problem, as Nvidia is doing with SLI, is not likely to make any difference to performance. In fact our tests on an SLI-enabled Opteron workstation from Delcam showed that the 3D performance in Inventor 10 went down by a whopping 30% when running in SLI mode.

Who will SLI benefit?

So which applications will benefit from Quadro SLI? According to Nvidia, in general, applications that are GPU (Graphics Processor Unit) bound, rather than CPU bound, will be able to scale under the dual graphics card technology. These types of applications are primarily in the high-end mechanical CAD sector and Nvidia has reported good scaling under Unigraphics and almost double the performance under MSC.Patran, a (CFD) Computational Fluid Dynamics application. Specialist VR and high-end Flight Simulation applications have also been named as beneficiaries of SLI and Delcam has reported seeing some benefits with Its 3D modelling solution, PowerShape, when working with very large models. But in the grand scheme of things this is just the tip of the iceberg. It’s still early days though, says Nvidia.

It’s taken Nvidia six months to fully understand how SLI can work with OpenGL and now with 80 OpenGL engineers focussed on development the company is confident that by the end of the year they’ll have a pretty convincing story. But there’s still a lot of work to do.

While some applications will work straight out of the box – this was exactly what happened with MSC.Patran, says Nvidia – many ISVs (Independent Software Vendors) will have to work closer with Nvidia to make Quadro SLI perform effectively with their software, possibly including some re-working of their product’s code. The question is how many ISVs will have resources to spare, particularly with many already channelled into porting, or re-writing applications for Windows XP x64 Edition.

SLI, however, is not all about ‘bleeding edge’ performance on a single monitor. SLI Multi View can drive four high-resolution displays out of a single workstation, and the forthcoming Release 80 driver will offer a 32x SLI AA (Anti Aliasing) mode, which is designed to provide much higher image quality without compromising performance at resolutions up to 1,920 x 1,200.

" SLI is worth keeping an eye on to see how it develops, but you’re not guaranteed performance gains "

What do you need?
  • 2 x Identical Nvidia Quadro FX 1400, 3400, 3450, 4400 or 4500 graphics cards
  • SLI bridging chip
  • SLI-capable motherboard with two PCI Express x16 slots(e.g. Tyan Thunder K8WE or Intel D955XBK)


SLI is currently a niche solution in the workstation sector. At the time of press Nvidia could only quote a handful of applications that show any significant benefit from the technology and ISVs have not yet certified their applications for SLI.

In response, Nvidia says that the dual graphics card technology is still a work in progress and expects to see improvements in performance and features with every new driver release. That said it’s still likely that many applications will never benefit from SLI due to their CPU overheads, at least when it comes to speed.

Professional 3D graphics performance is a very complex issue, and the levels you can achieve are largely dependent on which application you use. Some applications will benefit from more powerful professional graphics cards, while others will be just as happy with entry-level boards. Nvidia’s Quadro SLI complicates this even further by adding two high-powered graphics to the equation. If you’re currently running out of 3D power and looking for new ways to boost performance, SLI is worth keeping an eye on to see how the technology develops, but you’re definitely not guaranteed performance gains and the old adage of try before you buy is more important than ever.


Leave a comment