Date: Thursday , August 01, 2013
Today, AMD is releasing a new Beta driver to the public, this will be Catalyst 13.8 Beta. Normally we don't dedicate an entire article to a Beta driver release, unless there is some big new feature or news to report. This is such a case. Inside this driver is a new technology called Frame Pacing. This is AMD's answer to smoothing out and creating a more consistent and fluid gameplay experience with CrossFire. For a long time NVIDIA has benefited from technology in its drivers to smooth out SLI, AMD is finally catching up. However, read-on, there are some caveats.
Let's backtrack a little. For as long as we have been evaluating multi-GPU configurations we've been relating to you the gameplay experience delivered. This means a lot of things. Primarily, this means not only what performance is like, but more importantly the actual feel and consistency while gaming on said multi-GPU configuration. If you look back, we have years and years of editorial content and commentary in each multi-GPU review telling you how we experienced gaming on hardware.
You will find a common thread as you read back over all of our evaluations. The common thread has always been that NVIDIA's SLI has been smoother and maintained a better consistency compared to AMD's CrossFire. Sometimes you could see this in the framerate on our graphs. However, most of this fluidity is related to the frame times. While we lacked the tools at the time to show you this smoothness and consistency difference, we didn't really need any tools other than actually sitting down, and playing games, to see the difference.
In fact, is you ask NVIDIA as to why it built its FCAT frame time montioring software and gave it to reviewers, it is because of what the engineers there read on HardOCP and they wanted to investigate further.
This is the heart of what we do, we play the games. Through this method of evaluation, which is exactly what the gamer does while playing, we've experienced consistently NVIDIA's SLI feeling smoother at lower framerates compared to AMD's CrossFire, which needed higher framerates to feel smooth. What this meant is that games felt fluid and smooth, with no stuttering, and a consistent framerate, at lower tolerable framerates, such as 30-50-FPS on NVIDIA SLI. However, with AMD's CrossFire we often had to maintain higher average framerates, to get the same smoothness and fluidity, often times in the 50-60FPS+ range.
The reason for the inconsistency we experience in AMD CrossFire is because inherently, AMD's CrossFire was not outputting the same frame time durations. There were some frames that rendered longer, and some that rendered shorter, or some frames that didn't even render completely. In recent times, this frame time inconsistency has come into the limelight in a big way as has been reported by other websites. We wonder what took them so long to see what we were seeing? For years we've been reporting this gameplay difference by simply playing games as they were intended on both hardware configurations. Score another point for real world gaming hardware reviews. The bottom line is that we did not, and do not need FCAT to tell us if our gaming experience is not the best it could be. That said, we understand how many of our readers have been looking for objective data on this particular point.
At any rate, this greater scope and focus on the inconsistencies in AMD's CrossFire has lead to some tools finally showing promise in showing us exactly what we've been seeing for so long. NVIDIA developed a program called FCAT as mentioned above, which is able to output your gameplay via a capture card to a separate system and then use tools to analyze frames and the time between frames. Other websites have grabbed hold of this tool as the Holy Grail to demonstrate frame time issues.
Recently, we have built our own FCAT system and have had personal assistance and demonstration from NVIDIA in how to use it. Unfortunately, it's not something that is easily transformed into meaningful data in the way we can publish information, and is very involved to use in the first place. In our case specific monitors were needed as there were incompatibility issues with our displays we normally use. That said, we are investigating how best to use FCAT to show you, our readers, information that is useful to you in a meaningful way. Keep in mind that we have been telling our readers about smoothness issues literally for years, be it in an subjective sense rather than the FCAT objective way. We have argued the need for FCAT at all and the fact that looking at FCAT data can also be misleading when it comes to evaluating actually gameplay impacts.
We have plans to first investigate NVIDIA's SLI and AMD's CrossFire in its own article based on real-world gameplay experience. Then, we will follow-up with a data driven article, using FCAT, for those that find that useful. However, this AMD Beta driver is not final yet, and is being released in phases. The final phase won't be out for quite a while, potentially a month. We want to summarize the differences between NVIDIA SLI and AMD CrossFire in a separate article and do it right, providing real meaningful gameplay and data to get to a final conclusion between those. Time is needed to accomplish this. We personally feel that simply playing the game, and relating the gameplay experience, is the best way to tell you what the real deal is when it comes to gaming.
Also, FCAT right now has a problem with the Catalyst 13.8 Beta driver, reported by AMD, which causes it to provide falsely dropped frames right now. Therefore, FCAT is unreliable with this driver. To quote AMD:
"You might have come across scenarios while testing this driver, where FCAT reports 15 dropped frames, in some applications, fairly consistently. These are not dropped frames. The way FCAT works with frames and color bars, is whenever its 16 color bar sequence is interrupted, it interprets that as dropped frames . We’re working on resolving a glitch in our driver that is causing FCAT to incorrectly report dropped frames."
Beyond FCAT, what we can do is what we've always done, play games the way gamers do on hardware and report our experiences. Today, this is what we are going to do as well and check out if the Frame Pacing Technology causes any performance loss. We will be focusing solely on AMD CrossFire today with this new driver.
As the title suggests, the big new technology contained with Catalyst 13.8 Beta is something called "Frame Pacing." This is AMD's new technology to eliminate "micro-stutter," inconsistency, and create a more fluid gaming experience.
What is Frame Pacing? AMD describes it like this. Frame Pacing is a driver driven technology that works by measuring the time it takes to render each frame. If it starts to see disparity in the time it is taking to render frames, then the driver can insert in the next frame an artificial delay in the following frames for as long as it takes to even out the frame time. Therefore, it has to detect an issue with a frame time first, before it kicks in and smoothes out the subsequent frames. I personally wonder if this will cause a stutter. Gameyplay will tell.
Frame Pacing will be an option you can enable and disable in the driver. We like that we have full control over the feature to use it or not. This allows us to see if it really is improving anything. The option will be enabled by default.
AMD claims that most games AMD has tested show no raw performance difference with Frame Pacing enabled. However, there are a small percentage of potential GPU intensive games that could have a 3-5% raw framerate performance decrease with Frame Pacing enabled. To us, this is acceptable. You most likely won't notice a 5% performance drop, and if it helps by creating a more fluid and consistent gameplay experience, then it will be a win.
AMD is releasing this technology in phases. The driver today, Catalyst 13.8 Beta is Phase 1. In Phase 1 Frame Pacing only works at resolutions up to 2560x1600 on a single display. That's right, Frame Pacing will not work on CrossFire in Eyefinity resolutions in Phase 1. In Phase 1 only games that are DX10 and DX11 are supported, no DX9 games will work with Frame Pacing in Phase 1.
Phase 2 of the driver will come later this month. In Phase 2 CrossFire in Eyefinity resolutions will be supported. There will also be support for DX9 and OGL games in Phase 2. So if you game in Eyefinity, or across multiple displays, or higher resolutions than 2560x1600 then Phase 2 driver is what you will have to wait for. Also, if you want support in DX9 games.