Articles

SLI: Worth it?

Written: 13th May 2008
Section: Hardware


My bursary finally came in not so long ago, and with the lacklustre performance I'd been getting with some of my newer games, a cheap upgrade was on the cards just to give it that extra edge.

The question was where to start though. I know that for TF2 my CPU was the bottleneck because I briefly had a dual monitor setup and the frame rate was pretty much unchanged. The original plan was just that, replace the CPU. But you know how these things go - you start with the CPU, then think the RAM is too slow, the motherboard chipset sucks, the current PSU won't handle the new componenets... and before you know it you've shelled out over 300 quid on an SLI system. Not the 100 it was meant to be... but never mind.

Before the second card was fitted, I booted up with just the one 7900GTX and noticed a massive performance increase in the games that were giving me hassle before. Call of Duty 4 was running well over 100fps average at 1280 x 720 but my intention was to push it up to the native 1680 x 1050. The frame rate fell quite low and I ended up turning everything off - even anti-aliasing. Now was the time to unleash the SLI capabilities and finally see if it was good as it's hyped up to be. For a start, it wasn't simple. My original 7900 was bought off eBay, as new, in October 07 but had been giving me some grief with regards to 'cutting out' after watching full-screen video in windows. It came back to haunt me when running some benchmarks in SLI and the desktop became unusable. The quick fix was to switch the cards and put the used one I bought off ebay in the master slot. This didn't solve the problem fully, but I could still navigate around the desktop. Besides, it was only on one particular shader test that it triggered the problem and I've not seen it since - nor the full screen video error.

Benchmark performance proved SLI to do exactly what it says on the tin with the frame rate doubling in every test. Not bad for sixty quid I figured. Some time back I read an article about how synthetic benchmarking and real world performance can sometimes be totally different and I really hoped this wasn't one of them. You'll imagine my frustration when I noticed no immediate improvement when playing games. Using Call of Duty 4 as an example, the frame rate did indeed excel where it had already been quite high before but still hit as low as 40fps in certain spots. Not even sure if the SLI was working, I enabled the visual indicator which shows how it's performing. In Alternate Frame Rendering (AFR) it shows a bar down the side of the screen with an indicator starting from the middle. The higher it goes up and lower it goes down the screen, the more the cards are able to run at their full potential. This basically means that if the bar was full and the frame rate was low, the the cards would be having trouble rendering the images. However it turns out that the slow spots were actually bottlenecked by the CPU as the bar was mostly empty. I was pretty pissed off by this point and retired for the night.

In its current state, it was AFR which is basically one card does one frame, the other does the next. This generated nice results in the benchmarks but I noticed that the increased presence of screen tearing was awful, only made worse by the constant low to high fluctuations in frame rate. There's also a slight lag on what you're viewing because you're always a frame behind what you're actually viewing. They say it's not that noticeable but it did make the game feel quite sluggish to control. The alternative method to AFR is Split Frame Rendering (SFR). While more CPU intensive, the frame rate wasn't wildly skipping about as before and was at a more stable level - plus the input lag and most of the tearing was gone. The game played a lot better using this mode, despite its lower than average frame rate I was getting with AFR.

So all in all, a moderate success with CoD4. TF2 did pretty well too, where I maxed out the res, pushed the detail settings up and still got a higher frame rate than before the upgrade. In AFR mode the frame rate crawled at 5fps in high action scenes. I'm not sure if the cards couldn't render the frames in time and simply choked, but switching to SFR completely fixed that. SFR is definately looking the winner here and it's now my default rendering method. I was looking forward to trying AudioSurf on the highest setting and was currently getting 50 fps on the medium setting. With SLI I saw no noticeable improvement. Two or three frames at most and it still slowed to 44fps in parts on the medium setting. Why such a negligible increase, I don't know. It may be to do with vsync being forced on but then SFR has no problem with that.

Is SLI worth it? There are two situations where it is: You're a hardcore gamer with a high end system and can afford to shell out on another top of the range card; You've a card that was considered high end in its day and is now reasonably cheap, but still stands up well to the new competition. Anything else, forget it. It's either too expensive or you'd be better off buying a single newer card than running two older ones. At the end of the day it was only 60, but in any other case I'd be bitterly disappointed.

Javascript disabled!

Choose a theme


System Status #1

Javascript must
be enabled


System Status #2

Javascript must
be enabled


Monthly vote

Computers...

Annoy me to hell
Are part of life
Are my forte
Are for facebook


© Daniel Richardson 2007 - 2017