RX 480 is one of the mid-segment graphics cards of the RX Series. RX 480 High Performance is still attractive at its low price. RX 480 Review.
Performance for everyone. With Polaris, AMD is attacking the top-selling entry-level to a mid-range segment of dedicated graphics cards. In the test, we look at the top model Radeon RX 480 in the reference design and clarify what the improved architecture and the significantly smaller 14 nm production process bring.
According to AMD, 84% of gamers buy graphics cards in the $ 100-300 range. It is precisely this segment that AMD is now devoting itself to with its first graphics cards based on 3D transistors (14 nm FinFET at Global Foundries). The first graphics card is the top model of the Polaris series called Radeon RX 480 and is based on the Polaris 10 chip. This is code-named Ellesmere (Canadian island). It offers 2,304 shaders and should therefore represent the full configuration of the chip. Thanks to the 150-watt TDP, it gets by with a single 6-pin power plug and would theoretically also be suitable for notebooks (as a competitor to the GTX 980).
Polaris is AMD’s first architecture to be manufactured using the 14 nm FinFET process. It is the 4th generation of the GCN (Graphics Core Next) architecture and offers the following improvements according to AMD:
- Improved geometry processing
- Improved shader efficiency (up to 15% per CU for RX 480 vs. R9 290)
- LiquidVR API support with “Variable Resolution Rendering”
- The improved memory controller and Delta Color Compression (DCC) with full 2/4/8: 1 support
- Doubled level 2 cache
- Shader instruction pre-fetch and improved buffering
- Improved async compute
- AMD TrueAudio Next (calculated by shader)
- H.265 Main 10 Decode and 4K60 HEVC Encode
- HDMI 2.0b, DisplayPort 1.3 HBR (5K60 with a single cable) / 1.4 HDR with HDCP 2.2
- HDR display support
These improvements should lead to up to 2.8x better performance per watt (RX 470 vs R9 270X). You can find more details on our detail page about the Polaris architecture.
Our test system for the following benchmarks consists of the following components:
Intel Core i7-4790K (not overclocked), 16 GB DDR3, Asus ZU97-Deluxe, Asus PB287Q 4K monitor, Intel SSD 530 (240 GB, system), OCZ Trion 100 SSD (480 GB, benchmarks), Windows 10 64 bit with all Updates in mid-June. As a driver, we used the pre-driver 16.6.2 which was made available for the press before the launch. The 16.7.2 should increase the performance slightly on average (+ 3% with Computerbase, for example).
In the synthetic benchmarks, we rely on the 3DMark series from Futuremark and the products from Unigine to test the graphics performance. However, you should be careful with these values, since all manufacturers optimize their graphics cards and drivers, especially for the 3DMark programs.
The new Radeon RX 480 is on average ahead of the GeForce GTX 970 (-6%) and Radeon R9 290X (-5%) in our benchmarks, but mostly lags behind the GTX 980 (notebook) (6-11% faster) and R9 390X (4% faster). In the old 3DMark 11 graphics test, the RX 480 can even put itself just ahead of the GTX 980), but the performance in Unigine Heaven 4.0 is rather weak.
As expected, the new Geforce GTX 1080 (Pascal) plays in a different league and can clear an impressive 86% on average.
The Radeon RX 480 has 36 compute units that can be used for general calculations. For this purpose, AMD propagates the open OpenCL protocol, but DirectCompute is also supported. In the test with LuxMark (OpenCL) and ComputeMark (DirectCompute), the graphics card shows a very good performance, which is a bit faster than in the games. The GTX 980 and R9 390X are slightly slower on average here and the GTX 1080 can also break away less than in the games.
Our brief attempts with the new “WattMan” overclocking tool were only characterized by modest success. With a 5% higher core clock, the card ran stably with up to 1330 MHz under Witcher 3 and achieved about 4% higher performance (Full HD, high details). However, we have not validated this setting with lengthy performance tests for stability. With a + 10% increase in the core clock, the system hung up when the setting in WattMan was adopted.
By and large, the RX 480 ran relatively reliably over several days in our test parkour. However, there were about 3 reboots during a game and some games were unexpectedly ended while they were running. Assassin’s Creed Syndicate especially didn’t like our hardware very much. In addition to a reboot, the game ended reproducibly. According to AMD, the problem is being worked on. Fifa 16 as a second example did not want to start with Fraps (can only be activated during the game). Otherwise, the remaining 32 games ran without problems in the test. Overall, it a slightly worse performance than the GTX 1080 in our test, which only had difficulties in Hitman but did not provoke reboots. Overall, however, a good performance for a new architecture before the official sale.
The new video engine can definitely score in the test. A 4K H.264 100 MBit video could be played back with 84 fps and 3% CPU load in the DXVAChecker Playback Benchmark. The GTX 1080 managed this in comparison with 138 fps and 4% CPU load.
The new H.265 decoder managed a 4K 10Bit video with 72 fps and 2% CPU load.
The RX 480 cannot come up with a PlayReady 3.0 (SL3000) certification like the GTX 1080 already has. However, according to AMD, this should be possible in the future.
Thanks to the 14 nm process, AMD has made a significant leap in power consumption. While the idle values of our system were not as convincing as those of the GeForce GTX 1080, a slight improvement over the XFX Radeon Fury Pro was measurable. With the connected 4K monitor, our Voltcraft VC-940 showed 77 watts for the RX 480 (after a long period of rest on the desktop). The GTX 1080 reached 68 watts in the same system. According to AMD, a bug in the driver is to blame for the relatively high consumption. With the launch driver, the card should require significantly less power (about 1/3 less and thus no longer significantly more than 1080).
The consumption is clearly significantly lower under load than with the GTX 1080. We measured 242 watts when FurMark was used for a longer period and 247 watts with The Witcher 3 at 58 fps. GPU-Z reports between 100 and 120 watts for the GPU alone. The GTX 1080 achieved 294 watts in both tests and 109 fps in The Witcher 3. The XFX Radeon Fury Pro achieved over 340 watts in the same system.
The AMD Radeon RX 480 scores well in terms of the measured volume. There is practically no measurable difference to the system noise with the integrated GPU when idling. The volume increased to a maximum of 46.3 dB at 2176 rpm of the fan under load. The GTX 1080 reached up to 49.4 dB under load. The transistor whistling is also much less pronounced in our RX 480 test sample than in the GTX 1080. Even in the 3DMark Ice Storm with 2000 fps, a faint whistling is only audible from 20 cm. It should not be audible when the case is closed.
The back of the RX 480 heated up significantly in the test. We measure a maximum of 72 ° C in our test system. The top and sides stay relatively cool with a maximum of 50 ° C. However, the GTX 1080 in the reference design reached similarly high temperatures.
Inside the graphics card, the sensors report a maximum of 81 ° C under heavy load (The Witcher 3, for example) – absolutely uncritical.
As an alternative to the GTX 1080, AMD is promoting two RX 480s in Crossfire mode. These should achieve a comparable performance cheaply. AMD showed Ashes of Singularity at their press conference, which achieved a comparable frame rate. In the absence of a second test sample, we could not verify this claim. Basically, we are skeptical about Crossfire and SLI. Too often there are problems due to missing profiles (no performance improvement) or micro-stuttering. The situation could improve a bit with DirectX 12 games, but we wouldn’t rely on it too much.