Nvidia's SLI technology:



SLI, which is short for Scalable Link Interface, was introduced by Nvidia when the first PCIe motherboards hit the retail market. Well, introduced is a big word, as the SLI principles were already known to the big public as Scan-Line Interleave which was develloped by 3Dfx (which was later acquired by Nvidia) to support multiple graphics cards producing a single output. In other words, you place more than one videocard in your computer and all those videocards co-operate to render one output signal, faster than when you'd do the same with one videocard.

Nvidia has done some changes in the design of SLI, going from "you need to identical videocards" to "you need to same type of videocard". They also introduced the Tri-way SLI solutions and even Quad SLI, which places four gpu's in one system, wasn't left out. The main problem, however, is scalability. Nvidia claims to have excellent performance scaling technology, but in real-life applications, adding another videocard doesn't always pay off. Most of the time, you have an increase in FPS of about 35-50%, because when adding an extra videocard, you need a processor that is capable of handling all the information sent by your graphics processing units. However, in the low-end cards segment, this SLI technology story is really interesting, because the CPU doesn't have to be thát good to handle the extra VGA power.

We have released two other articles concerning the SLI technology, if you're interested, please have a read.

Talking about SLI, we tested two Leadtek 9600GT Extreme samples and compared the results with the 9800GTX reference board. Performance wise you can easily judge that SLI can give you quite some extra performance, up to roughly 70% increased FPS! Here and there you'll stumble upon games which have less support for NVIDIA's multi GPU-technology, there you'll have to do it with around 30% extra performance. But on the other hand, even with the 9800GTX having only one GPU and thus probable less configuration problems, the card never took the lead in average frame rate over the dual 9600GT's branded Leadtek. Yet, again, in CM DIRT for example, the GT's had to deal with the phenomena called Micro Stuttering, this does not occur when using single GPU configurations.

~ Leadtek 9600 GT SLI reviewed and compared with 9800 GTX.

A few words on SLI. I cannot deny nVidia's SLI, while exciting at first is now thought of as a marketing gimmick which seduces the Gamer or PC-Enthusiast into spending twice the amount of money for two cards which produce twice the heat, require twice the power and take up twice the real-estate. The choice to release QUAD SLI updates in the future seems 4x as silly. The founders of 3DfX are more than likely rolling over in the proverbial 3Dfx grave, excavated while trying to integrate multi-GPU technology onto a single PCB. SLI reborn as Scalable Link Architecture IMHO probably coined the internal marketing phrase Profiting off the passion. Sadly we see this occurring throughout the PC-Industry. While I do respect Asus without whom there may be no "Enthusiast" since they were the pioneers of OEM, they are in the end a for profit enterprise.

~ Asus Extreme N6600GT Review - SLI on a Budget.

NVIDIA Geforce 9500 GT

The cards we used for this SLI follow-up article are the Leadtek and Galaxy we already tested in our previous 9500 GT article. We had no issues placing them next to each other in an SLI configuration, although both cards have a different bios with different clock frequencies and even have different revisions of the gpu core.

Both of these 9500 GT's perform well enough in my eyes. They're low-end and thus by definition not breaking and performance records. As both cards were released on day 1 of the official release, the prices are a bit high for low-end cards, especially when you compare against their DDR2 brothers and sisters. However, these prices will certainly drop and with the 55nm version coming, I cannot say these are not worth to consider. The last part of that sentence is important ... 'worth to consider' ... there are a lot of different products on the market and most of them are pretty much equal in terms of price/performance ratio. Some are better; some are cheaper and in the end it all depends on what card you feel is best for you. In the low-end segments, there is nothing like THE best card, there are a lot of possibilities. My advice, in other words, is that you should at least consider the 9500 GT when buying a low-end card.

Test setup

9500GT Test Setup
CPU	Intel Core 2 E8500 (from )
Cooling	Stock Aluminum Intel Heatsink
Mainboard	MSI P45 Platinum MSI P7N Sli Platinum
Memory	2 * 1GB PC6400 Micron Technology
Other	Galaxy OC 9500GT Leadtek Winfast PX9500 GT Tagan 520W PSU Western Digital 320Gb SATA HDD Windows XP SP3 Dutch

Methodology: benchmarks

To compare all motherboards, we used the following benchmarks:

3DMark2001SE

3DMark03

3DMark05

3DMark06

Aquamark3

Trackmania Nations

Quake4

Methodology: settings

To run the SLI tests we used the exact same settings as we did when testing the Galaxy and Leadtek cards separately, however this time on an 750i motherboard. Now, we did NOT re-test the single card test conditions on this 750i motherboard and that for a very good reason: when you intend to only use one videocard, you have other and better options than a 750i or even more expensive 780i/790i motherboard, namely the P35 or other derivative products. Therefor it's more valid to run the tests on these two different systems configurations.

Please note that the OS configuration, VGA driver and all other test settings such as memory timings were kept identical.

Futuremarks tests

These synthetic 3D benchmarks from Futuremark allow you to evaluate the expected performance of a system with different generations of games. As each 3DMark uses different features and quality settings, it allows you to get an idea of how your system will perform.



3DMark2001SE

3DMark2001SE has become more and more a system benchmark rather than a graphics card benchmark over the years. Nowadays, you can really see the difference in DDR2 and DDR3 memory setups, high and low FSB setups and so on.



An extra card add some extra performance, about 40.81%, however overclocking the SLI configuration has little to no effect on the performance level.

3DMark03

3DMark03 is very GPU dependant, which comes down to: the more cores, the higher the score. SLI and CF performance increase is spectacular in this benchmark.



As expected an insane increase in performance in the 3DMark03 benchmark, as it's very GPU dependant: 69.93% faster than a single card solution.

3DMark05

This product of the 3DMark series is again more CPU dependant than its predecessor, meaning that an overclocked CPU will have a big influence on the end score.



Quite a nice boost in 3DMark05 as well, performance went up by 59.03%.

3DMark06

3DMark06 is known to stress the GPU to the maximum and provides SM2.0 and SM3.0 test to compare products.



In the latest DX9 benchmark of Futuremark, the performance goes up by a stunning 67.94%.

Aquamark3

While this 3D benchmark is aging quickly, it still serves as a good evaluation tool for system performance, as it stresses the VGA, CPU and memory subsystem, any advantage had from a faster motherboard or memory setting will be apparent.



28.20% increase in Aquamark3, which is not really that shocking, knowing that Aquamark likes a faster CPU over additional video cards.

Trackmania Nations



A free game in the truest sense of the word, TrackMania Nations Forever lets you drive at mind-blowing speeds on fun and spectacular tracks in solo and multiplayer modes. Nations Forever will offer a new "Forever" version of the Stadium environment, a solid solo mode and 65 brand new, progressively difficult tracks. TrackMania Nations Forever will unite an even larger number of players than the original Nations thanks to its engaging multiplayer modes, innovative online functions and revolutionary interactivity between players.

We'll be testing the differences in 3D rendering in two set-ups:

1280x1024 - Medium quality & 1680x1050 - Very high quality



This is what is really important: gaming performance. In lower resolutions, the SLI configuration hits performance levels that are 75,18% higher than single card configurations. In higher resolutions, the performance difference is pretty much the same, being 74,09%. All in all a very decent increase in performance!

Quake 4

Quake 4 was released in 2006, based on the popular Doom 3 engine from ID Software. Even with the latest video cards, this game is a hard nut to crack if running in highest possible settings. Luckily, if you're able to play at the highest settings is so astonishing you never want to downgrade your card again.



We made a custom timedemo which takes the most out of the cards and ran the timedemo a couple of time, the results were averaged.

We'll be testing the differences in 3D rendering in two set-ups:

1152x864 - High detail & 1280x1024 - Ultra



Quake 4 is another level of game in comparison to Trackmania, especially considering the stress on the different system components. In lower resolutions, an additional card gives you 33.40% higher performance, in the higher resolution test environment 88,33%! Especially that last test result is quite remarkable.

Power consumption:

An extra card brings along next to the performance increase an increase in power consumption. We tested our SLI configuration in both idle and load test environments and noted down the power draw of our complete setup.

As you will see, the increase in power consumption is quite noticeable as one 9500GT at full load will draw as much power as two 9500 GTs SLI at idle.



In idle mode, the increase is 41.86%, in load mode, 21.62%, which is considering the performance increase not that bad.

Evaluation:

First of all, let's have a look at the total performance increase charts. We calculated the performance increase in all different tests and averaged them to find the average performance increase over a reference 9500 GT single card solution.



Almost 60% increase, which is definitely not bad for an SLI configuration. In most cases, adding an extra card does not have this kind of improvement. Probably due to the weakness of the card, as at reference speed it's not that fast. Compared to an overclocked card, the SLI configuration is only about 30% faster. Notice that changing to the 9600 GT will make your gaming experience almost twice as good, because of a close to 100% performance increase.

At first sight, you'd all be cheering for the SLI configuration, but ... there is a little problem; to be more specific a little money problem. Below you find a price rating chart for which we consulted Geizhals.eu (€) and Froogle.com ($) and averaged the prices of 9500 GT and 9600 GT. The 9500 GT SLI rating is just two times the single 9500GT price, as you might have guessed.



A single Geforce 9600 GT is less costly than two 9500 GT cards. The choice is not hard to make.

Conclusive thoughts:

In the end it is very difficult to recommend a configuration which is clearly outperformed by a card that is, if we follow Nvidia's logic, considered to be a bit faster, a configuration which is clearly using more power and is, to end with, more expensive. No, if you have to choose, don't go for the 9500 GT SLI configuration, but for the 9600GT instead.

However, the most interesting question is the upgrade cost if you want better performance than a reference clocked 9500 GT. My first answer would be: consider overclocking it! As you can see, the Galaxy sample comes factory overclocked and is about 20% faster than a reference card. If overclocking is not your thing, or you want even more performance, then there are two solutions:

Add one extra 9500 GT, costs about €63/$96. Don't consider this option if you don't have an SLI-ready motherboard. Otherwise the extra expenses will be dominated by the money you pay for a new motherboard.

Change to the 9600 GT, costs about €110/$170, which roughly is 75% more expensive than adding an extra 9500GT. However, if you sell you 9500 GT for 65% of the price, which is not much if the card is in good condition, you will be paying €69/$108. And that is close to the same you pay for an additional 9500 GT card.

Change to ATI HD Radeon 4670 which has proven itself to be almost on par with the 9600 GT performance wise, but with a below €100 price tag!

To finish, I would like to thank Selene from Leadtek and Igor from Galaxy for allowing us to re-test their 9500 GT based products, MSI for the P7N Sli Platinum, Vvikoo for being able to re-use the 9600GT and Tones.be, Belgium’s Biggest Hardware Shop for providing the Intel E8500 CPU.

'Till the next time, stay tuned!