AMD announces Radeon Software 17.7.2 for Windows users

AMD hasn’t really made a splash about a new version of the Radeon Software that they ship for consumers in a while, and it seems like they’ve been hoarding it all for pushing new features in one go. Available today for download, Radeon Software version 17.7.2 doesn’t seem to be a big hit, but it’s the end result of a lot of work behind the scenes to tidy things up and introduce features that people have been asking for. And when I say “a lot”, there’s really a lot of things to go through today, so let’s get cracking.

In a first for AMD, they’re releasing the Radeon Software 17.7.2 drivers for both Windows and Linux (listed as version 17.30 for AMDGPU-PRO) on the same day. AMD did not detail how many of these features are making their way into Linux, or if they’re even porting the driver interface over to match the version on Windows. A lot of this is down to how small the team is (less than 20 developers total), and the bulk of the team’s energy is being used on driver development for the Linux kernel. With time, this will change, but for now Windows provides the better experience and a useable GUI.

So let’s start with the changes. Radeon Settings now integrates Radeon Additional Settings into the control panel, which means that the colour options and advanced pixel accuracy features are no longer hidden away in a separate window. This makes it easier to set up UHD 4K monitors that support the full colour space with the full 4:4:4 pixel format, which helps reduce colour banding on monitors that support this mode for ultra high definition content.

One of the slides has AMD poking fun at NVIDIA’s GeForce Experience app, which is installed with the GeForce drivers and gives you an interface to control game settings and set up game recording or streaming using Shadowplay. GeForce Experience requires a user sign-up though, and they have a host of social networks linked in so that they can share this information with third parties. AMD’s marketing now boasts that Radeon ReLive, their competing game recording and streaming feature, doesn’t require an account sign-up. Heh.

ReLive also gets some new features and updates. It now supports recording footage at 100Mb/s instead of 50Mb/s, which reduces the need for compression in the final video output. This is GPU-dependent, so not every AMD GPU can support it. The Radeon RX 400 and RX 500 series cards are probably the best options for this purpose. For streamers, you can now overlay your face on the output with transparency. Neat.

Despite the other features that AMD is shoving into ReLive, they’ve managed to optimise it even further, reducing the hit on framerate from an average of 6.3fps down to 4.2fps in their test scenario running Call of Duty: Infinite Warfare. Every game engine is going to behave a little differently when you’re recording gameplay, and unless you’re using something like OBS, which relies on CPU horsepower instead, making use of the GPU to run the game and render video will incur a performance hit. This is likely AMD’s best scenario in their testing. Also added, but not shown on slides here, are options to make changes to the audio setup for streaming, as well as new overlay and notification options that pop up when you’ve started a recording.

Radeon WattMan also gets an update, adding in more power state controls for the GPU. It now supports memory underclocking, which can help a lot with power consumption tweaks, and will be a useful feature for small-scale cryptocurrency miners who have two or three other cards in their system that are used for mining and not gaming. This is limited to Radeon RX 500 series GPUs, so owners of older cards may have to turn to third-party solutions like MSI Afterburner instead. Also now available are power state controls for Radeon RX 500 and RX 400 series cards.

Both of these features were highly requested, according to AMD, and they are both tremendously useful for notebook users as well. We’ll see some new notebooks based on Radeon RX 500 this year, and maybe I’ll be able to personally check a few of these out myself.

Next up is Radeon Chill. This is a feature that automatically limits the framerate of your game based on user input. So for slower-running titles like strategy games or puzzle games, your average framerate will trend downwards, saving on power and heat, while faster-paced games like twitch shooters, racing games, or MOBAs will run at higher framerates, but still not at the maximum framerate that the GPU could push. Because Radeon Chill also supports variable buffer flips, the idea is that the display is updated according to your input in addition to the most recent data about where your mouse cursor is, or what inputs are coming from the keyboard. The end result should be a more responsive gaming experience with less input lag.

Radeon Chill’s interface now has two sliders to configure what range you want the framerate to stay within, and it now supports the DirectX 12 and Vulkan rendering engines. There’s now sizeable reductions in power consumption on the GPU alone, with DOOM dropping 72 Watts of power with a Radeon RX 580, or Battlefield 1 seeing a 44W decrease. These aren’t huge numbers, but the gains in power efficiency help AMD punch above their targets for power consumption.

Radeon Chill now supports 22 additional games natively, and will be enabled by default on these titles moving forward. It’s yet another thing I need to disable before doing benchmarks in reviews, but it’s a tangible saving in power costs for the consumer, so keeping it enabled isn’t a problem. The difference between a native implementation as opposed to the default implementation is that AMD works with the developers of popular titles to have better support for Radeon Chill at the game engine level, helping to further decrease power consumption. It works for just about any game, but there are always more savings to be had. AMD currently doesn’t support games sold through the Windows Store for Radeon Chill, but that will probably be a feature worked on for next time.

Finally, Radeon Chill works for graphics cards attached to your PC through an external Thunderbolt enclosure, as well as multi-GPU Crossfire configurations. Saving on the GPU load for Crossfire is a big change, and it’ll make the experience more reliable as well if the GPUs aren’t thermally throttling themselves. In addition, Radeon Chill now supports laptops, so there’s some battery life savings in store for mobile users. AMD’s claimed savings are for League of Legends, which is one of the lightest workloads one can throw at a notebook with discrete graphics.

In a similar vein, there are also updates to Frame Rate Target Control (FRTC). FRTC is just a standard frame limiter, but it now includes support for DirectX 12 games as well as multi-GPU Crossfire setups. FRTC and Radeon Chill do share some common software features like lower buffer flip queues to display the latest, most recent frame, but it isn’t meant to complement Radeon Chill. Rather, you can pick which setting you prefer based on your use case and preference. If you want Battlefield 1 to max out at 120fps instead of 100fps with Radeon Chill, this option allows you to configure it like that.

FRTC might even see further power savings when you limit the framerate to 30fps, which is why NVIDIA’s drivers implement a 30fps limit automatically for gamers playing on a laptop while drawing power from the battery. You can use AMD FreeSync with either option, but you have to choose between FRTC or Radeon Chill.

Twenty-seven slides into AMD’s presentation, we reach the second most popular feature request from users – per-display colour controls. In older versions of AMD Catalyst, colour controls were available on a per-display basis, but the move to Radeon Crimson Software in 2014 did away with that, replacing the feature with a global colour profile. This works just fine for multi-display setups that have identical monitors, but for people like me with mixed panels, it’s a pain in the butt to configure colours properly. Now you can do it from one interface in the drivers.

Next up is Enhanced Sync, a new feature for Radeon Software. This is similar to NVIDIA’s Fast Sync solution in addition to Adaptive V-Sync, and Enhanced Sync mimics both of NVIDIA’s features and puts them into a single driver option. What this is, is a replacement for the common options of V-Sync and triple buffering for gamers who don’t have a variable refresh display. When Enhanced Sync is turned on, games can run at their maximum framerate (120fps on a 60Hz monitor, for example), but frames rendered outside of the monitor’s refresh rate are dropped from the display chain. In effect, you get a FreeSync-like experience for games that run fast on your system. The frame displayed on your monitor is always the most recent one that the GPU has just rendered, and there’s no perceptible input lag because it also has the last action you made in the game included in the completed frame. The only difference is that there’s a frame cap of 60fps on your display.

In comparison to V-Sync, which matches the game’s frame rate to the display’s, there’s much less input lag, and the game updates at a much faster rate. If you’re a CS:GO player, you’ll find some use for this because it also eliminates tearing.

Enhanced Sync also eliminates the stutters that you see at times when you have V-Sync on and the game’s framerate drops below a rate that the monitor can display, which will drop the displayed framerate to 30fps until it can reliably match the monitor refresh rate again. This has all the usual problems attached to it from V-Sync, including increased input lag, repeated frames, and a slow update rate. Instead, now you’ll only see a single tear as the frame drop occurs, and tearing remains enabled until the frame rate climbs back up again. There’s no input lag, and the frame rate isn’t automatically lowered to match what the monitor expects as an input.

Now this sounds awesome – and it really is – but it’s only available for Radeon RX 400 and RX 500 series graphics cards. You know, the kinds that are currently out of stock because cryptocurrency miners are snapping them up constantly. AMD’s older GPUs from one generation ago aren’t getting this feature, and their current flagship GPU family, the Radeon R9 Fury, R9 Nano, and R9 Fury X, are ineligible for this feature update, even though they’re perfectly capable of doing it. Maybe one day AMD will fix that, but for now the feature focus is on Polaris and the upcoming Vega architecture.

Performance updates

Lumping all the performance improvements over the past year, AMD has pulled off some great feats. For the Windows platform, AMD managed to extract an extra 12% performance on average with just a handful of games shown here, which is about as much as you get from overclocking the GPU. On the Linux platform, AMD’s open-source drivers continue to improve, almost doubling framerate for Deus Ex: Mankind Divided, as well as pushing out an average improvement of 25% for native OpenGL games on Linux. It isn’t the biggest platform for AMD, and they continue to understaff it because it isn’t tremendously important for them gaming-wise, but it’s nice to see some improvement nonetheless.

More involved, behind-the-scenes work has also yielded improvement, although these are much less obvious. AMD’s shader cache feature will save shaders that have been constructed by the game engine using the GPU, and that work doesn’t have to be done again. It saves a small amount of time in game level loads, but it’s more advantageous for weaker systems like cheap PCs and consoles. AMD’s drivers also spend less time computing things in the display process, so there’s a few milliseconds worth of time savings in the drivers that improves game responsiveness.

It seems like community-based alpha and beta testing is the “in-thing” to be doing now. Apple’s doing it, Microsoft’s doing it, and there’s an entire section of the game industry that makes money off it. Well, AMD now has their own program for beta testers called “Radeon Software Vanguard”. If you sign up with Vanguard, you’ll be able to push submissions for bugs, suggestions, and reporting glitches to AMD directly instead of going through the report form or the Radeon forums. As a Vanguard, you’d get access to early drivers, and you’ll be expected to submit quality improvement reports to AMD. Now you can have more stuff break on you more often!

The rest of the slides now focus on developer features and improvements, so you can click out of here if you’re done with the consumer-focused announcements.

VR improvements are coming in sparsely for AMD, but there’re still in the works. For now, they’re announcing that there’s improved video playback for 360 videos that you view using a VR headset. Videos now run at the full 4K resolution, support 90fps playback, and also support multi-channel and multi-direction audio. That last part is probably an extension to AMD’s TrueAudio support functions, but TrueAudio was designed for VR environments in the first place anyway. It’s a little weird. There’s also plugin support for accelerated video playback support for Unity and the Unreal Engine, so it’s an easy feature for developers to support.

Because developers also need to figure out game performance in a standardised way, AMD is finally picking up and updating their OCAT software which allows them to hook into the Windows 10 PresentMon API to record frame times and general game performance just like FRAPS. I use OCAT myself for GPU testing and looking at frame variance, and the older version was very out of date. Hopefully moving forward AMD pays more attention to this software. And it can do overlays now! I’m not sure how this function works, so I’ll have to play around with it later this week.

Trust me, if you’re tired out by now looking at these slides, the actual presentation AMD had was over an hour long! There’s a lot of stuff to chew on with this update.

For developers and hardware reviewers, AMD is finally enabling access to a free GPU profiler as part of their GPUOpen initiative. The Radeon GPU profiler will allow users to create hardware traces of events to see in-depth into the performance of their systems, and it will give developers a deeper understanding of how their application works and responds to the hardware. Profiling tools are usually pretty expensive, at least on the Windows platform, and as a bonus they’re making this open-source software free on Linux as well. Special access needs to be granted to get this tool though, so you have to directly apply to AMD to get it.

And that’s it for the Radeon Software Update. There’s an update in the wings for Radeon Pro, which I’ll have up when the NDA drops, but for now eager developers can get their hands on Radeon Software 17.7.2 for Windows 10 from this page, with AMDGPU-Pro version 17.30 for Linux systems releasing a bit later this month.