Custom PC - UK (2020-06)

HOW SLI

WORKS

Scalable Link Interface
The very first iteration of SLI actually dates
back to 1998 when 3dfx implanted a
technology that used two GPUs to render a
3D scene. Here, while the acronym was the
same, the technology was called Scan Line
Interleave. You linked up two matching 3dfx
Voodoo 2 cards with a ribbon cable, and they
would split each frame into lines, with each
card handling alternate lines. In some games
you could even see the split between the
lines as the two cards worked on them.
This didn’t just get you higher frame rates

• it also enabled you to hit higher resolutions.
  Voodoo2 cards were generally limited to
  800 x 600 in games, but you could push
  your system to 1,024 x 768 with an SLI setup.
  Sadly, the technology proved too tricky and
  expensive to obtain widespread use.
  The last gasp for this form of SLI was seen
  on the Voodoo5 range, with the Voodoo5

5500 featuring two chips on the same card,

working in tandem in SLI. There were even

plans for a four-chip SLI card, the monstrous

Voodoo5 6000, but it didn’t get released

before 3dfx folded, with much of its assets

bought by Nvidia.

Six years later, Nvidia repurposed the now

defunct brand name, refashioning it to stand

for Scalable Link Interface. This new version

of SLI achieved essentially the same job,

splitting up the workload of rendering a 3D

scene into as many chunks as there were

GPUs (some graphics cards, such as the

GeForce 7950 GX2, also implemented SLI on

a single card) to provide double, triple or even

quadruple the theoretical processing power.

At its core, the technology is quite simple,

with the graphics driver deciding which

chunk of the scene should be handled by

which card and the card then going off and

doing the number crunching. The results

are then added together to produce the final

stream of frames that we see.

As has long been established,

performance with multiple GPUs doesn’t

scale linearly, but with the right games, a quad

SLI setup was the ultimate gaming rig for a

while, until Nvidia abandoned support for any

SLI setup with more than two graphics cards.

SLI modes

At the heart of the way SLI works is the choice

between two different rendering modes:

Alternate Frame Rendering (AFR) and Split

Frame Rendering (SFR). As their names

suggest, AFR simply sends each frame to

alternate GPUs; for SFR, the driver splits the

scene into multiple parts, with each one sent

to a GPU and combined again at the end.

As you might expect, AFR is the far simpler

version, and requires less overhead in terms

of communication between the two GPUs, so

it’s preferred by developers and Nvidia alike.

However, it also has the higher potential for

introducing lag, as the system may need to

wait a split second for a frame to fully finish

being rendered. SFR requires much more

communication and some of the work is

duplicated, although it can result in better

performance when implemented properly.

This also hints at the reason why SLI setups

with three or four cards produce diminishing

returns. With AFR, you potentially have one

GPU sitting idle for some time while it’s waiting

for all three other GPUs to finish rendering their

withtheGeForce 6800 series

3dfx’s Voodoo5 5500 card
featured two chips, which
communicated using SLI, when
it stood for Scan Line Interleave

AfterA buyingmuchof3dfx’sx s

assets,NvidiarelaunchhedSLI

asScalableLinkInteerface,

with the GeForcee 6800 series