XFX Radeon R9 380 Double Dissipation 4GB review

AMD’s graphics cards have long provided impressive performance for the money, but with most of its cards being based on three-year-old architecture, many gamers in search of the most powerful cards have turned to Nvidia’s GeForce GTX 980 and its newer and more energy-efficient Maxwell architecture.

AMD fans have been waiting a long time for a brand-new GPU design, but unfortunately the Radeon R9 380 isn’t the all-new silicon we were hoping for; that’s reserved for the top-end Fury and Fury X, two seriously powerful cards designed for gaming at 4K resolutions. The R9 380, like the rest of the 300-series range, is more of a mid-range card, built around the same Graphics Core Next architecture that first arrived in 2012.

The GPU

The Tonga GPU core seen here is effectively a rebadged and slightly higher-clocked version of the same chip used by last year’s Radeon R9 285 – which itself was a modified version of the graphics chip used in the Radeon HD 7950 with Boost GPU released in 2012. With 1,792 stream processors, 112 texture units and 32 ROPs, the only thing differentiating the 380 from the 285 is its slightly faster 970MHz boost clock speed. AMD doesn’t tend to reveal core, as opposed to boost, clock speeds, and third-party manufacturers are free to overclock their own AMD-based cards to higher speeds, often using custom coolers to help them do so.

It might have a higher boost clock, but the R9 380 draws an identical amount of power to the R9 285. A claimed 190W average under load is a definite improvement over previous AMD cards, which were notorious for running hot and needing plenty of juice, but it still falls short of the incredibly efficient Nvidia competition. The GeForce GTX 960, the R9 380’s closest rival, only draws a claimed 120W. You’ll need two six-pin PCI-Express connectors to power the card.

The R9 380 is available with both 4GB and 2GB of VRAM. The 2GB version is around £20 cheaper, but we think this is a false economy, particularly as modern games such as Batman: Arkham Knight need at least 3GB. More VRAM will also make a big difference when you start gaming at resolutions higher than 1080p. Interestingly AMD has made the switch from the 384-bit memory bus seen in the HD 7950 with Boost to a narrower 256-bit bus here. However, with the memory clocked at a higher 5.5Gbit/s, the actual memory bandwidth has increased overall – meaning the new card should still outperform the old models.

The card

AMD usually supplies a reference design for each of its graphics cards, but for the R9 380 launch it seems the company was content to let its board partners create their own custom designs. The XFX Double Dissipation edition we looked at for this review has a twin-fan, four-heatpipe cooling system that connects directly to the GPU core and vents hot air out into the case.  

It all works well, keeping the GPU below 40 degrees when at idle without making a sound. It might not stop the fans completely when you’re on the Windows Desktop, as Nvidia’s GeForce GTX 960 manages, but the XFX card’s fans spin so slowly under light load that it was impossible to hear them over the other fans in our reference system. You can certainly hear the fans spin up when in games, but we would hardly call them loud.

XFX has overclocked the GPU boost clock to 990MHz and pushed the memory clock to 5,700MHz – 20MHz and 200MHz increases respectively over the reference R9 380. These are fairly modest bumps, but also give a clear indication that AMD is approaching the limit of its GPU architecture – we struggled to eke out much more speed when overclocking our review sample manually. 

The card needs two six-pin PCI Express power connectors, and has two dual-link DVI outputs, a single HDMI and a DisplayPort 1.2 video output on the rear. Frustratingly, all AMD’s 300-series graphics cards, and indeed the Fury and Fury X, make do with HDMI 1.4a video outputs. This makes them less than ideal for hooking up to a 4K TV, as you’ll be limited to a 30fps refresh rate; we prefer 60fps for properly smooth gaming, and Windows Desktop applications are jerky at less than 60fps.

Performance

It’s performance that really matters, though, and in this respect the R9 380 didn’t disappoint at 1,920×1,080. In Dirt Showdown we saw a silky smooth 86.9fps, along with 49.9fps in Tomb Raider with the very demanding super sampling anti-aliasing (SSAA) turned on.  Swapping SSAA for the less resource-hungry FXAA resulted in a superb 102.8fps. Even the demanding Metro: Last Light Redux was playable at maximum detail, with an average 30.4fps. Again, disabling SSAA boosted the frame rate considerably to 55.4fps.

The R9 380 trades blows with Nvidia’s GeForce GTX 960 at 1,920×1,080, where the extra 2GB of VRAM doesn’t make much of a difference to frame rates, but stepping up to 2,560×1,440 reveals more of a performance gap. 67.3fps in Tomb Raider, with maximum detail and FXAA enabled, is an excellent result from the R9 380, beating the GTX 960 by as much as 20%. There was a much smaller gap in Metro, however, with both cards averaging 34fps.

It was only when we increased the resolution to 3,840×2,160 that the R9 380 truly began to struggle. You’ll have no trouble with older games, as Dirt Showdown still produced a playable 37.6fps, and even Tomb Raider ran at a borderline 30.9fps, but Metro dropped to a choppy 15.6fps.

Conclusion

It may not be entirely new, but there’s no doubting the R9 380’s performance in games. Depending on the title it will outperform the similarly priced Nvidia GeForce GTX 960, cope comfortably with any game at 1,920×1,080 and even manage smooth frame rates at 2,560×1,440.

Unfortunately, while the faster clock speed and extra memory give it a slight edge over the R9 285, there’s actually very little to differentiate the R9 380 from its predecessor now that AMD’s driver optimisations have made their way to the older cards. If you’ll be sticking to 1,920×1,080 or 2,560×1,440 gaming for the foreseeable future, with no plans to make the jump to 4K, the R9 380 will be more than sufficient for playing new games, but if you only recently upgraded to a 200-series graphics card there’s no reason to make the jump to AMD’s new model.