NVIDIA SLI: Sometimes They Come Back...

The fastest, the strongest and the most disputed technology returns into the market of computer graphics as NVIDIA rolls-outs its new SLI initiative aimed at professionals and hardcore gamers. Two graphics processors are indisputably more powerful than one, but let us think whether NVIDIA really brings us roses or thorns?

by Anton Shilov
06/30/2004 | 05:49 AM

Introduction

The world of consumer 3D graphics knows a number of examples when companies introduced technologies to allow higher speed because of doubling the number of graphics cards or processors within the system.

 

Historically, professional and visualization graphics solutions boast the technique of multiple graphics chips quite successfully being able to produce roaring performance and image quality. The main issue with the approach of making numerous graphics processors to work in parallel is ability to develop drivers and to deal with exceptional cost of end-products.

While the price and thoroughly tailored drivers for specific applications are not problems for professional products, consumers want a cost-effective and a universal solution, something that always kept the dual graphics offerings away from the consumer market.

 
3dfx Voodoo2 SLI, 1998

3dfx, ATI Technologies, XGI Technology – all tried to popularize dual graphics cards or chips, but all failed because of different reasons, with the only exception of the 3dfx Voodoo2 SLI, which was the king of the hill and a graphics solution of choice for Quantum3D, a visualization company. NVIDIA has just stepped on the rocky road its competitors have failed to finish. However, it looks like the company is not pinning a lot of hopes to popularize the dual graphics card option among gamers, but bets more on professionals, leaving something for those who do not want to save on performance of 3D games.


NVIDIA GeForce 6800 Ultra SLI, 2004

Before NVIDIA’s newly introduced Scalable Link Interface technology gets fully real and there are performance numbers available, let us try to analyze whether the SLI is here to live for long, or is something that is just done to show the muscle.

Scalable Link Interface Explained

NVIDIA’s Scalable Link Interface, dubbed SLI, is a set of hardware and software capabilities that allows two PCI Express x16 graphics cards to work in parallel and deliver higher performance compared to one such graphics card would. NVIDIA is currently pretty tight-lipped about the technology itself, though, there are still some information that is available for general public.


NVIDIA GeForce 6800-series' die with SLI logic

Each GeForce 6-series graphics processor has special circuitry inside that allows it to work in pair with other GPU (or GPUs?). Graphics cards based on the GeForce 6 chips will have a special MIO port to connect one graphics card for another. NVIDIA explains that when two graphics cards work in parallel, each one renders half of the frame. The frame is divided dynamically in a way to share the workload as 50:50 between the graphics processors and, when all calculations are done, synchronize the parts and send them to the screen.

NVIDIA calls the technology of splitting frames between GPUs as “Dynamic Load Balancing” while the whole mechanism is called Symmetric Multi Rendering.


NVIDIA's Symmetric Multi Rendering with Dynamic Load Balancing

The dynamic load balancing is done by the upcoming ForceWare drivers and needs some additional processing power from the CPU. Once the driver splits the frame into two parts, it sends both to different GPUs. Once the GPUs get the data, they start to compute their parts of the frame. Both chips are likely to communicate via the special MIO port, however, not actual graphics data is likely to be send via this port because it can hardly provide enough bandwidth. Instead, the “master” chip sends certain commands to the “slave” chip ordering to send the half of the frame via PCI Express bus, or the “slave” reports that “he’s done”. In theory, MIO port is not that necessary and may be substituted by PCI Express x16 itself, however, it seems that the MIO makes everything just a little bit more efficiently and saves a number of clocks for the PCI Express bus, especially keeping in mind that today’s PCI Express applications seem not to support full-duplex mode allowing to perform read and write operations at the same time. The MIO port is most probable serves for frame-lock purposes as well.


MIO connector, front

At this point it is not clear how NVIDIA plans to “divide” data for pixel and vertex shaders between the parts of the frame as well as carry out other calculations requiring coordinates from different parts of the screen, e.g., Z-test, occlusion culling, etc, with the SLI technology. It is possible that when splitting the frame the software leaves certain areas to be calculated by two GPUs at the same time, but this is not something that is confirmed at this time.


MIO connector, back

PCI Express bus may seem to be a bottleneck in the whole SLI configuration, but NVIDIA claims astonishing efficiency of its Symmetric Multi Rendering technology – around 87% or even 100%, which definitely makes great impression and brings confidence in the technology. Since the number of transactions required between GPUs is nor clear, as well as the sizes of data chunks need to be transacted, it makes sense to wait and see the actual SLI applications.

Requirements for Dual Graphics Cards Configurations

In case everything is so good about the Symmetric Multi Rendering and Scalable Link Interface technologies, what will be required from users to have dual graphics card configuration at home in addition two NVIDIA GeForce 6-series graphics cards?

Initially power-hungry users will have to buy workstation components or special workstation or gaming machines to enjoy the power of two GeForce 6-series graphics cards. Desktop chipsets available this Summer and Fall from Intel, VIA Technologies and Silicon Integrated Systems are unlikely to support two PCI Express x16 ports for graphics cards. The only core-logic that has support for PCI Express x16 + PCI Express x8 ports is Intel’s E7525 also known as Tumwater designed for workstations. The core-logic itself costs $100, which leads to assume that mainboards based on the chipset will be priced in the range of $350 to $500 or even higher.


Intel E7525-based mainboard

In addition to a workstation mainboard for Intel’s newly announced Xeon processors users will have to acquire a Xeon chip (or two) with 800MHz processor system bus, a pair of DDR2 400MHz memory modules and, most probably, a special computer case along with a special and very powerful power supply unit to feed the graphics cards as well as the rest of the system.

All of such components will bring the pricing range of a system with two graphics cards to about $4000 - $5000 or so. Such price-point is typical for workstations, but not for desktops, though, quite some PC makers expressed intention to produce such systems. In the workstation segment dual-Quadro graphics cards are likely to be a very desirable and not that expensive option.


Two NVIDIA Quadro graphics cards in SLI/SMR configuration

Acquiring a Xeon system for gaming is not something loads of end users would do, as such computers are more tailored for work rather than for entertainment. Furthermore, even getting such system with one graphics card aiming to get another one afterwards (the pattern a lot of enthusiasts used with the “dual” Voodoo2 SLI) requires significant investment into the platform and CPU (while a pair of Voodoo2 boards could be installed into any PC), which may slowdown sales of gaming machines with two graphics cards.

A more user-friendly option from the price point of view will be brought by NVIDIA itself in the fourth quarter of the year. The company will release its NVIDIA nForce4 chipset that will sport two slots for PCI Express graphics cards and will be compatible with the desktop infrastructure, meaning significant cost reduction compared to the Intel Xeon platforms. In case NVIDIA succeeds, end-users will be able to get desktop computer at desktop price-point with one GeForce 6-series graphics card and add another card in future. The only thing end-users should be worried about is powerful and quality PSU.

NVIDIA’s Luciano Alibrandi Dots All "i"

In order to clear everything about the NVIDIA’s Scalable Link Interface out, we decided to make a short interview with NVIDIA’s Director of Product PR, EMEA, Luciano Alibrandi since the most reliable source for information is always official.


Notice the golden connector on the top of the graphics card - that is MIO port.

X-bit labs: For whom do you position the technology? For extreme gamers, or professionals, maybe some other types of users? 

Luciano Alibrandi: Both is the correct answer. All professional applications can take advantage of SLI also so Quadro customers will also see significant performance gains.

The SLI technology will provide a huge boost in performance and will allow the hard core gamers to use the highest resolutions and the best filtering all with FSAA turned to the max, thus experiencing a brand new gaming experience. For the professionals we have announced yesterday that the SLI technology will be possible on the new Quadro solutions based on the PCI Express bus. Where applications are significantly bottlenecked by graphics hardware, SLI technology is able to nearly double performance. For applications that are less bottlenecked by graphics hardware, SLI technology still delivers significant performance gains.

X-bit labs: Do you expect dual-card configurations to occupy significant market share?

Luciano Alibrandi: The SLI solution is target at the hard core gamers and high end professional graphics, market size in these areas is relatively small

 X-bit labs: How the technology actually works? Does it divide a screen into multiple into multiple parts and then renders, or it works like the original Voodoo2 SLI?

SLI stands for Scalable Link Interface.  It is a high performance technology that allows users to combine and scale graphics performance by having multiple NVIDIA GPUs in a single system. SLI works by intelligently scaling geometry and fill rate performance for two (or more) identical boards.

There difference in the way the technology differs from the Voodoo 2 SLI.

NVIDIA SLI technology is PCI-Express based, uses a completely digital frame combining method that has no impact on image quality, can scale geometry performance, and supports a variety of scalability algorithms to best match the scalability method with application demands

X-bit labs: Is it possible to implement more than two graphics cards using the technology?

Luciano Alibrandi: We are introducing dual GPU scaling only at this time.

X-bit labs: Do all GeForce 6 series GPUs sport the feature, or only GeForce 6800-series?

Luciano Alibrandi: All GeForce 6800 PCI Express cards or Quadro FX 3400 boards support SLI technology. Boards must be the exact same model number and by from the same vendor – for example, two GeForce 6800 Ultras from Vendor XYZ.

X-bit labs: Does the technology require any support from drivers and applications? Will professional applications work correctly and flawlessly with two graphics cards right out of the box? Do you have to certify possible dual-card professional solutions with software makers?

Luciano Alibrandi: SLI technology is application transparent and is enabled through the graphics drivers.  All DirectX and OpenGL applications can take advantage of SLI.  When two graphics cards are configured with SLI technology, applications will automatically be able to take advantage of performance improvements with no modifications.

X-bit labs: Does the technology rely more on hardware or software (e.g. drivers)?

Luciano Alibrandi: It is a combination of both. We have dedicated scalability logic in each GPU and a digital interface between GPUs to help enable this.
In addition we have a full software suite that enables dynamic load balancing and advanced rendering algorithms that provide the best image quality.

X-bit labs: Have you run into any issues with the technology? E.g., performance anomalies, artefacts, etc? Are they likely to emerge?

Luciano Alibrandi: Not at all. We run extensive testing on our solutions before delivering these to the market.

NVIDIA’s Luciano Alibrandi Dots All "i", continued

X-bit labs: How much power will two powerful GeForce 6800 graphics cards require?

Luciano Alibrandi:
Complete systems requirements will be rolled out during partner launch of the product.

X-bit labs: Any performance numbers or speed gains you expect from 2 GeForce 6800 Ultra boards, or GT boards, or, non-Ultra solutions?

Luciano Alibrandi: We will let reviewers to see the results... The amount of performance improvement will depend on the application. In some circumstances it is upwards of 2x increase in performance.  In general, the heavier the load, the more benefit delivered, so applications running at higher resolutions with higher image quality settings will benefit most.


Two NVIDIA GeForce 6800 Ultra graphics cards in SLI/SMR configuration

X-bit labs: Do you expect any graphics cards makers to offer bundles with two graphics cards certified to operate in pair?

Luciano Alibrandi: Our partners are free to decide the best way to promote their products and to promote our technologies. I do not see why not.

X-bit labs: In case there would be no makers to offer graphics cards bundles, where users should get the MIO connector?

Luciano Alibrandi: People will need to buy 2 identical cards and get the SLI connector from one of our partners.

X-bit labs: When do you plan to enter the technology into the commercial market?

Luciano Alibrandi:
SLI will be available in Fall 2004.  SLI is only for PCI Express, so system builders will supply SLI in their systems first, as they create an install base for PCI Express systems. Retail details will be rolled out in August.

X-bit labs: Do you expect PC makers to offer computers with two GeForce 6800-series boards installed?

Luciano Alibrandi:
Systems based on NVIDIA SLI multi-GPU technology are expected to become available in the second half of 2004 from the world’s leading PC and workstation manufacturers including Alienware, Atelco Computer, Boxx Technologies, Falcon Northwest, Mouse Computer Japan, Network Technical, Paradigit, Scan Computers, TSUKUMO, UNITCOM, Velocity Micro, ThirdWave Corporation, VoodooPC.

X-bit labs: How much the system with two graphics cards is going to cost?

Luciano Alibrandi: This will be decided by the system builders.

Happiness for Quantum3D, Disaster for 3Dlabs?

While not a lot of gamers are likely to afford computers with two graphics cards, professional users are likely to gladly make use of the dual-Quadro options. In case the Symmetric Multi Rendering and Scalable Link Interface technologies prove to be absolutely problem-less, professional users are likely adopt the dual card options massively.


Quamtum3D Aalchemy rack with 8 AA5 realtime 3D graphics subsystems.
Each Aalchemy AA5 subsystem is powered by up to 16 3dfx VSA-100 chips.

While NVIDIA does not say it directly, the name “Symmetric Multi Rendering” implies that more than two graphics chips can work in parallel. In case this is true, the GeForce 6-series is likely to be a successor of 3dfx VSA-100 that is still being used in high-end visualization systems by Quantum3D. The best side of the VSA-100 is scalability – tens of such chips can work in parallel, but given that the processors are totally outdated, Quantum3D may want something more advanced. For the first time in four years the company may finally get the chip it really needs with one of the GeForce 6-series products (of course, we should keep in mind power consumption of such components).


3Dlabs' Wildcat Realizm 800 graphics card

Some other professional graphics companies, especially 3Dlabs, should probably worry about NVIDIA’s dual-Quadro option. The computing and geometry power of the Quadro are known around the industry and two such graphics cards may beat high-end offerings from companies like 3Dlabs or, at least, put some pressure on their sales.

Final Words

3dfx Voodoo2 SLI was among the most expensive graphics solution for gamers in history. But for its days the option was the fastest. Now the SLI abbreviation returns to bring the ultimate performance crown to NVIDIA.


Unreal 3 requires more power than conventional GPUs can provide.
Maybe NVIDIA GeForce 6800 Ultra in SLI delivers enough power?

Preliminary examination of the Symmetric Multi Rendering and Scalable Link Interface technologies show that from technology standpoint both are pretty promising. However, the final cost of applications featuring two graphics cards is likely to slowdown sales of such systems.

The advantages of dual graphics cards configurations are pretty clear:

However, there is a number of disadvantages too:

To sum up, two graphics cards is a terrific option for professionals who need performance right here and right now as well as gamers who want to have highest performance possible at any price. To put it for gamers: if you are not satisfied with the performance of NVIDIA GeForce 6800 Ultra or ATI RADEON X800 XT, get yourself two GeForce 6800 Ultra and put them into SLI configuration – performance of your graphics will be undisputed.