rFactor 2 – SLI Benchmarking Comparison

We started our journey by testing rFactor 2, a simulator that has commonly been blamed for poor performance for the level of graphics shown. Although Studio 397 has improved the simulator in the last years, it is still a title that costs a lot to run at full graphic fidelity and it will be a good first test for our SLI. Will we be able to finally conquer rFactor 2 in ultra? If you have seen the video above, you will get an idea, but there are many things to test. Before we take action, let’s take a look at the starting situation:

rFactor 2:

• Native SLI support: no
• Graphical API: DX11
• Nvidia’s SLI profile: yes, but inoperative
• Relative CPU usage: average
• Relative GPU usage: very high
• Test conditions: Sebring with 30 GTE at sunset, to gather elongated shadows with abundant reflections and car lights, trying to put as much graphic load as possible.

First test, first issues. rFactor 2 does not have native SLI support, which means that the development team does not recommend its use or work to improve performance. However, a lot of games, simulators or not, work perfectly with SLI in spite of not having support. The problem is that it will be the user who will have to make his own life to configure his equipment and he will not receive any kind of official help, neither from the creator of the game, nor from Nvidia. In this case, we can see that if we load rF2 with the default profile, the simulator only uses one of the two graphic cards.

Luckily, the rFactor 2 community has managed to give, based on what surely have been dozens of hours of testing, the right configuration for SLI, and they have been kind enough to share it with the web. To change the SLI configuration we must download a program called Nvidia Inspector. Motorist users will know it because it’s the only way to apply antialiasing correctly to the simulator since a couple of years ago. Once downloaded and opened, we must load the rFactor 2 profile and make the following changes:

• SLI compatibility bits: 0x02402005
• SLI compatibility bits (DX10 + DX11): 0x024020F1
• Nvidia predefined FXAA usage: disallowed
• Power management mode: prefer maximum performance
• Nvidia predefined number of GPUs to use in SLI rendering mode: …GPU_COUNT_TWO
• Nvidia predefined SLI mode: …MODE_FORCE_AFR2
• Virtual reality pre-rendered frames: 0x00000002

This last setting, although is strange, is what forces the second card to run at full speed as well. It is not clear if it is a problem of the Nvidia driver or of rFactor 2, but Virtual reality pre-rendered frames must be set to 2 in order to use SLI to its full potential. It can also be changed from the Nvidia Control Panel if you prefer. It should look something like this:

With this, we have rFactor 2 ready to run with SLI. So let’s get the testing started.

1080p 

Well, we’re off to a good start. From the cockpit to 1080p, not only do we not gain performance, but we lose it. And the gains from the TV camera aren’t enough to launch rockets either. What’s going on here? It happens that our 7700K is the limiting factor. At the ultra graphics level, the i7 is not capable of producing more than 150 FPS on average, and it doesn’t even fully utilize a 1080Ti. Adding another one, far from improving performance, makes it worse by having an extra step of communication between the two graphics cards, which are only being used at 40% each. When we switch to the TV view, which is capable of using a 100% 1080Ti, we see a slight increase in FPS but again we hit the limits of the i7, even overclocked at 5GHz, which is unable to use two 1080Ti with all its potential. Let’s see what happens in a more normal scenario:

As we can see, we are facing the same situation. From the cockpit, the performance is worse, and from the TV camera we get a significant improvement this time, but still below what was invested. It should be noted that disabling Hyperthreading and turning the i7 into an i5 has no effect on performance, so we can conclude that rF2 does not know how to use more than 4 cores and therefore will not benefit from a Ryzen 7 or 9, or one of the new i7 or i9. This is the most you can expect from a processor in rF2 today.

Conclusions 1080p:

• 240Hz is unattainable with ultra graphics.
• SLI is not at all worthwhile. In fact, performance is lost.
• A GTX 1080 or equivalent is more than enough for 1080p 144Hz.
• The only SLI that might be worth it would be a 1070 SLI to go from 60Hz to 120Hz, and only with Gsync or Fresync.

1440p

We increase the resolution and… well, this is something else! From the cockpit we see that we are still limited by the processor, but we keep perfect FPS for 120/144Hz. When we switch to the TV view we see the SLI glow for the first time. 60% to 90% magnification! This resolution in that kind of view puts the focus on graphics power, and here we get, for the same price, a much higher performance than a 2080Ti would give. Let’s see what happens when the cars disperse:

Good news, again. The improvements are starting to be substantial. We can maintain the 144Hz from the cockpit, which with Gsync is a wonder, and from the TV camera we get a 60% increase to the average FPS. There are some drops, in places where the processor can’t take it anymore, but the average is above the 144FPS target in both situations. Again, we improved by far to a 2080Ti with the same investment.

Conclusions 1440p:

SLI is worth maintaining a strong 120/144Hz
1080Ti SLI substantially better than 2080Ti
A single 1080Ti would perfectly maintain 90Hz at all times
1080 down would limit us to 60Hz

4K DSR

Let’s go with 4K, where things get interesting. A second card here gives us about 60% extra performance. At this resolution the gains with SLI occur in all situations, since the limitation is the graphic power and our 7700K can work comfortably. In 4K we again substantially exceed what a 2080Ti would offer and we can always stay above 60Hz. What’s more, from the cockpit, even a 120Hz 4K screen would make sense. Let’s see when things calm down:

As you might expect, we have even better results. Gains of up to 70%, with a spectacular 175 FPS on average from the cockpit. While doing the tests, I saw over 200 FPS in some parts of the circuit from that view. With this configuration a 4K 120Hz screen would be a very good option to consider, if you have Gsync or Freesync. We could drive at about 175FPS on average and then watch the replays with all the ultra without ever going below 60FPS. Sync would make the experience totally seamless.

But maybe the most interesting thing is the triple 1080p theme. If we take into account that the amount of pixels is more or less in the middle between 1440p and 4K, we can expect around 150 FPS on average with everything in ultra. With a 1080Ti we could maintain 60 FPS without problems, but with SLI we could have a three screen configuration at 144Hz with Freesync or Gsync without sacrificing anything, with an impressive latency and fluidity.

A triple 1440p configuration at 60Hz also makes sense. It would mean a 33% increase in pixels, so from the cockpit we would theoretically have a performance of around 70 to 90 FPS. The repeats, however, would be difficult to run in ultra, with an estimated performance of about 40 FPS. On the other hand, it seems that the time of the triple 4K has not yet arrived and it would be a configuration that would force us to lower the graphics a lot to make it work.

4K conclusions

• SLI is very worthwhile
• 1080Ti SLI much better than 2080Ti
• 4K at 120Hz possible with SLI
• Triple 1080p at 144Hz possible with SLI
• Triple 1440p at 60Hz possible with SLI
• No SLI limited to 4K 60Hz and triple 1080p at 60Hz.

Final conclusions

SLI in rFactor 2 works very well once the profile is set up with Nvidia Inspector, especially considering that it has no official support. You can check it yourself in the video that heads this article. We are not going to consider the results at 1080p because, first, they are not the fault of a technology malfunction, but of the processor and the low resolution, and second, it would be a complete foolishness to buy two graphics cards of this caliber to play at 1080p. The results we have seen here would be repeated in any other game, simulator or not, at such low resolutions. Let’s look at some final data:

• 50% extra cost to add SLI (400 ? 1080Ti second hand)
• 2080Ti costs 62.5% more than 1080Ti, and adds about 35% in yield.
• SLI 65.6% average gain at 1440p TV (limited graphics power scenario)
• SLI 64.7% average gain at 4K (all scenarios)

Two 1080Ti in SLI, on rFactor 2, are significantly better than a 2080Ti, and at the current second-hand price, much cheaper. Apart from the pure data, with Gsync activated, the experience is completely smooth and satisfactory. We do not appreciate sudden frame drops, nor the dreaded “stutter”, nor any adverse effects. It just works, and it does it very well. That’s why we will declare this first result as very satisfactory and money well spent.

See you at the next stop: Assetto Corsa.

See you on the track!