Hybrid SLI.. The definitive comparison

Well.. it seems alot of people don't seem to understand the workings under SLI and have simply decided that the Hybrid SLI found in their laptop is pretty much useless.. While there is some truth to this it isn't completely true.

As those of you with SLI on your desktops will know, SLI is largely based on 2 things. Game support, and driver support.

By game support I mean some games will darn right refuse to work properly with SLI enabled. Some games may also work with SLI enabled, but show no clear benefits with SLI turned on. And linking in, nVidia have to make profiles for each game in their drivers. If you're up for tweaking it is possible to add your own profiles using nHancer, but then again, how many of you would be willing to do that?

So what I plan on doing, is benchmarking several games to give people an insight into the workings of SLI, what games support it, what games don't, what games need tweaking.. and hopefully based on this you can decide whether it's worth it or not for you in particular.

At the moment I've only compiled benchmarks for one game.. hopefully in the coming weeks/months I'll be able to add plenty more. From time to time I also add gaming videos to my youtube to demonstrate how the game plays (I record using a camera) be sure to check those out too at www.youtube.com/pillowshooter

Anyways, on to the meaty stuff.

Crysis - Low DX9 1280x768





As you can see, at low the FPS makes a tasty improvement from running on one GPU. The average moves up from 27 to a slick 40. That's a 50% improvement. The minimum makes a slight improvement and the maximum is also a slick 50% higher. During the actual benchmark the difference is pretty noticable. Hopefully I'll have a video up to show you the difference.



Crysis - Medium DX9 1280x768





Bearing in mind, medium settings means I set every setting to medium and above low means everything to low. The numbers in this case show smaller improvements, but improvements none-the-less. The average is up by about 30%, the minimum takes a similar increase, whilst the maximum improves by over 50%. In all honesty though, the game wasn't playable like this with either solution. With a bit of tweaking you could probably squeeze a few more fps out (I'm benching with the pre-release demo so the full version patched up should probably also yield some performance enhancements).

However it's clear to say that Hybrid SLI has increased performance by reasonable amounts. It'll allow you to up the details above the standard solution without dipping below it in FPS. Of course, it means you need to play around and test but if you do decide to, then you get to reap the benefits.


Counterstrike Source - Max DX9 2xAA 1280x800




All settings set to their highest values except Vsync (Off) and Antialiasing (2xAA).

I feel bad putting this one in because most people know that the source engine is probably the most friendly engine out there. It looks great, scales well on older hardware and supports dual graphics cards like a champ. Due to a driver issue when using SLI the framerate is very low at the start of the benchmark. Using the standard Dell drivers stops the issue, but unfortunately I can't use the default dell drivers for other reasons. Having said that, the average FPS jumps by about 50% from near to 50, to 75ish. The maximum is a whole 75% higher, and as you can see the minimum is pretty poor after the bad start.

(I'm currently using the 179.67 drivers. I plan to test out the more newer 18x.xx series drivers soon which should yield more benefits to the SLI benchmarks.

Driver Improvements

Well I've recently update my drivers to Dox's Hybrid Drivers (185.68) and I think the graphs speak for themselves. (Crysis will be rebenched later)




More benchmarks to follow

 

Nice graphs .

 

Very interesting! I'm going to put a link in my site. Would you mind to run our standard benchmarks (at least for Crysis) to compare your laptop against others?

 

My graphs scaled a little wrong earlier, I've fixed that now and you can see that they are much better aligned.

I will try to get that benchmark for crysis for you by tonight.

Edit: Added CSS bench

 

hate being a sticklier... but label your graph's axis..

took me a while to realize that the second graph's y-axis is FPS, and that the first graph's y-axis is FPS and x-axis is probably time (hey, its 6am in the morning after pulling an allnighter to study calculus..)

 

You should read them when you're not tired then

Imo it's pretty obvious what the graphs are showing, and you should know that on time/other graphs the time is on the X axis, and whatever else is on the Y axis.


Also added 2 new graphs to demonstrate the improvements drivers make.

 

Hey, nice work with the benchmarks.

My experience with the dell xps 13, Power save vs Boos mode is that in games like warhammer online and command and conqure 3 I found that the framerates were much more stable in power save mode but on average lower.

In boost mode the frame rates were much higher on average but dived to sub 10 fps when new rendering was needed. ie when you look around in and FPS or scroll over the terrain on a RTS.

To the effect that in power save mode the games ran smooth but in boost mode they stuttered.

So i ask myself, what is the point in boost mode?

These observations were made both with and without dox drivers. The next step is trying custom SLI profiles for these games.

Can anyone comment if they are having similar issues?

 

It's true. Some games don't really like SLI that much and don't play well. Whenever I play FSX or Age of Mythology I keep SLI off because in FSX it causes micro stutters and in Age of Mythology it just isn't needed.

But in CSS and Company of Heroes I've found SLI improves performance nicely without these stutters.

Unfortunately I don't own THAT many games, so testing is kind of hard. Often game demos don't have proper profiles and that means the performance of the demo is often lower than that of the actual game.. and obviously as I've pointed out a number of times SLI is really dependent on the drivers.

I do remember reading an article though that said micro-stuttering is a side effect of using multiple graphics cards, not just a problem in this system. Here it is: http://www.overclockers.com/index.p...:microstutter&catid=60:videocards&Itemid=4266

You may want to have a read.

 

Is there any way to run the 9200m on its own without the 9400m?

I'm thinking in terms of freeing up the RAM used by the 9400m as shared memory.

 

Nope. The 9400m only uses up 256mb of RAM anyway.

 

There's something odd going on with my video memory, more than 256mb has gone "missing". I'm running 32-bit vista at the moment but it should be able to use more than it sees. I have 4 gigs of RAM and the nividia cp is showing 895mb of shared memory in use.

895mb is fair chunk of memory so I'm trying to investigate a way to get it back.

 

The 895mb of RAM is the maximum amount of RAM that will be assigned to the card only IF it needs it. Else it will just use 256mb default. I currently have 3837mb of RAM available on 64-bit Vista and my nvidia control panel shows similar stuff to yours.

The only way you're going to reclaim that ram is by using a 64-bit OS.

 

I heard it suffers from micro-stuttering due to the difference of frame times of the two cards. Is it really much of a problem?

 

It hasn't been too noticeable for me in real life application as performance is pretty much always better than the 9400M G alone by a decent amount.

 

I've been having an opposite experience. 9400M works just fine, hybird SLI stutters so much it's unplayable.

 

I haven't noticed it too much either.

I wouldn't say it is unnoticable though, but in some games I'd have no idea that it's there (Company of Heroes for example is perfect with SLI - tidy performance boost so I can have higher settings and it's silky smooth)

However some games do suffer from microstutters, such as Flight Simulator X.. then again that game is more CPU bound tbh.