We’ve been hearing about AMD’s next-generation Polaris GPUs for a little over six months now. We knew those chips would have enhanced display output capabilites, and we also knew they would be produced using 14-nm FinFET process technology. Now, the first product to use this new architecture is here. The Polaris 10 GPU on AMD’s Radeon RX 480 graphics card isn’t a high-end monster like the GP104 chip powering Nvidia’s first 16-nm FinFET cards, though. Instead, the RX 480 puts a $200-and-up price tag on VR-ready performance.
AMD sees a great deal of opportunity to regain market share in the $100-$300 graphics card price range, and the RX 480 is the company’s first shot at the middle of the enthusiast bell curve. Let’s see how it intends to join this battle.
Baby, you’re a star
Polaris is actually a pair of chips—Polaris 10 and Polaris 11—that are two takes on the same underlying architecture. Both Polaris 10 and Polaris 11 will be making their way into desktops, but Polaris 11 is a smaller chip that will probably find a home in many more gaming laptops than it does desktop PCs.
|Polaris 11||16||64/32||1024||2||128||???||???||14 nm|
|Polaris 10||32||144/72||2304||4||256||5600||232||14 nm|
As you can see from the table above, Polaris 10 occupies a much smaller footprint than Tonga before it. Despite its smaller area, Polaris 10 has a bit more of almost everything on board: more stream processors, more texturing units, two times the L2 cache of older chips at 2MB, and a variety of neat tricks that we’ll discuss momentarily.
The block diagram above should be largely familiar to anybody who’s laid eyes on AMD’s past GCN chips. Polaris 10 has 36 GCN compute units for a total of 2304 stream processors. It has 144 texturing units, four geometry engines, 32 ROPs, and a 256-bit path to GDDR5 memory. From a pure resource standpoint, this part falls somewhere between Tonga and Hawaii in the AMD pantheon.
Polaris 11 is, at a glance, a little less than half of Polaris 10. It has half the ROPs, less than half of the shaders, half the triangles per clock, and a memory bus that’s half as wide. It also has 1MB of L2 cache. It’ll almost certainly be smaller than Polaris 10’s 232-mm² die, and AMD further notes that this chip is its thinnest GPU ever. That may sound like a weird claim to make, but it’s important for achieving the kind of console-class gaming experiences in thin-and-light notebooks that AMD wants to deliver with this architecture. Aside from one desktop card, we don’t really know where Polaris 11 will land or what names it’ll carry yet. The spotlight is on the Radeon RX 480 for today.
Now that we’ve seen the two basic Polaris parts, let’s examine the common improvements that AMD baked into the underlying architecture.
Under the hood, Polaris is a more efficient version of the GCN architecture we’ve known since its inception on the Radeon HD 7970. This fourth generation of GCN incorporates a number of small improvements.
The geometry engines in Polaris now feature a stage called the Primitive Discard Accelerator that can remove zero-area (or “degenerate”) triangles, as well as polys that aren’t being sampled, early in the graphics pipeline. AMD says this feature improves performance in workloads that combine lots of triangles (like highly tesselated scenes) and multi-sampled anti-aliasing. Heavy tesselation has traditionally been a weakness for GCN cards, so it’ll be interesting to see what effect this feature has on Polaris’ performance. Each Polaris geometry engine also gets an index cache for storing what AMD describes as “small instanced geometry.” This cache reduces the need to move data around the chip, reducing internal bandwidth requirements and improving throughput.
Polaris 10’s front end is getting a new pair of programmable units that AMD calls “Hardware Schedulers,” or an HWS for short, alongside its four asynchronous compute engines. These blocks perform a variety of scheduling tasks for asynchronous compute workloads. They can set up real-time and prioritized task queues for audio and VR processing, manage concurrent tasks and process scheduling, and perform load-balancing between compute and graphics workloads. Since the HWSes can perform this work on the chip, they reduce CPU driver overhead. Because they’re programmable, AMD says it can update the capabilities of each HWS with new microcode, too.
Compute Unit Reservation in action. Source: AMD
One example use of the HWS duo involves audio processing for VR. In order to be sure that a given audio task will complete within a certain time frame, a developer can use a new feature called Compute Unit Reservation to request a specific number of on-chip resources that will be dedicated to a specific task queue. The HWSes ensure that the proper resources are allocated for the job—”spatial management,” in AMD parlance. These blocks can also perform what AMD calls “temporal management.” An example of such a task is managing the Quick Response Queue that the company specifically made for handling VR-related workloads like asynchronous time warp for the Oculus Rift.
The stream processors in each GCN compute unit are getting some new tricks in Polaris, too. If many wavefronts (AMD’s name for groups of threads) of the same workload are set to be processed, a new feature called instruction prefetch lets executing wavefronts fetch instructions for subsequent ones. The company says this approach makes its instruction caching more efficient. Polaris CUs also get a larger per-wave instruction buffer, a feature that’s claimed to increase single-threaded performance within the CU. Polaris can group client L2 cache requests, too, so it can fetch data from that cache more efficiently.
In addition to a larger L2 cache that allows more data to remain on the chip, Polaris has improved delta color compression (or DCC) capabilities that allow it to compress color data at 2:1, 4:1, or 8:1 ratios. These compression methods should allow the chip to enjoy greater effective memory bandwidth and higher efficiency. Polaris’ memory controller supports 8 GT/s GDDR5 DRAMs for up to 256 GB/s of memory bandwidth. Instead of moving to a new memory technology, AMD says it’s getting more life out of GDDR5 mostly thanks to Polaris’ improved DCC capabilities. Between DCC, the expanded L2 cache, and improved cache access methods, AMD claims it can reduce the power required by Polaris’ memory interface by up to 58%, too.
Getting smart about power usage
Along with the move to the 14-nm FinFET process itself, AMD is deploying several new monitoring technologies on Polaris that are meant to help each chip perform at its best. The company is making this move in response to a basic problem of power delivery to the chip itself: variations in input voltage as large as 10% to 15% require an increase in the average voltage sent to the chip to compensate. AMD says this safety margin wastes a lot of power, so it’s responding with a new technology called adaptive voltage and frequency scaling, or AVFS.
When the company designed Polaris, it borrowed a few pages from its CPU design team. Each Polaris GPU now has embedded frequency sensors on its die that work in concert with its temperature and power sensors. If a chip can run at lower voltages to achieve a given frequency on the DVFS curve, for example, this tech will let it do so and allow it to save power at the same time. The chip can also quickly adjust its frequency in response to voltage droops instead of running within a safety margin at all times, extracting 5%-10% more performance on average.
That on-chip monitoring technology also allows the GPU to analyze the input voltage it’s receiving from its host system at boot time and compensate for any differences between that input and the power characteristics of the test equipment on which the chip was initially binned. The chip can then use this information to adjust its voltage regulators to deliver the same operating environment it saw on the test bench, improving efficiency.
Finally, AVFS lets Polaris chips compensate for the effects of transistor aging and the aging of other components in the system. Polaris’ AVFS modules have aging-sensitive circuitry inside that let the chip compensate for any degradation in performance as it gets older. That same boot-time monitoring technology can determine when other components in a system (presumably, the power supply) are no longer as young and spry as they once were, too. By self-calibrating and adapting to this aging, AMD says the chip will offer “more robust operation” over time while delivering better performance out of the box.
Getting down to the real nitty-gritty, AMD is improving the design of the multi-bit flip-flop circuits it uses in Polaris. The company says there are about 21 million of these circuits on Polaris 10, and they account for 15% of the chip’s TDP. By moving to a quad multi-bit flip-flop, AMD says it reduced Polaris’ TDP by 4%-5%.
All that’s well and good, but we know you really want to see the RX 480 card itself. We won’t make you wait any longer.
The Radeon RX 480 and friends
If you’ve watched TR over the past few days, you’ve already seen the AMD reference design for the RX 480. Today, however, we can spill all the beans about clock speeds and field-strip the card down to its PCB. Let’s get to it.
|Radeon RX 480||1120 MHz||1266 MHz||2304||4GB or 8GB GDDR5||1x 6-pin||150W||$199.99 (4GB)
The Polaris 10 GPU on the RX 480 will run at 1120MHz base and 1266MHz boost speeds. We expect AMD’s board partners will push those reference numbers up a bit as part of their usual process of tweaking and tuning. At those stock clocks, this card has a rated board power of 150W.
The RX 480 will ship in two versions: one with 4GB of GDDR5 RAM, the other with 8GB. The 4GB card will have a $200 suggested price, while doubling the RAM will set buyers back an extra $40. AMD is giving its board partners freedom to adjust GDDR5 speeds, but it’s set a 7 GT/s floor on the speeds those vendors can use. The reference card we’re testing is an 8GB version, and it has 8 GT/s GDDR5 on board.
Around back, the RX 480 has three DisplayPort 1.3 ports and an HDMI 2.0b port with HDCP 2.2 support. Those DisplayPorts are DP 1.4-HDR-ready, too. Folks who want to plug a DVI monitor into their RX 480 without an adapter are out of luck.
Flipping the card over reveals a stubby PCB with an extra vent above the blower fan. This vent hole is a nice touch for folks planning to install multiple RX 480s in their systems in CrossFire.
Stripping the cooling shroud from the card reveals a small aluminum heatsink for the GPU itself, plus another heatsink for the card’s power-delivery circuitry. You can also see the single six-pin PCI Express power connector the RX 480 needs to do its thing.
Unscrewing the card’s entirely standard fastener complement lets us strip the RX 480 right down to its bones. As you can see from this view, the aluminum heatsink uses a copper disc in its base, presumably to improve heat transfer between the GPU and the rest of the heatsink.
Moving closer to the PCB gives us a better look at the Polaris 10 chip itself, as well as its eight GDDR5 memory chips. With that, you’ve basically seen all there is to see of the RX 480.
While we’re only reviewing the RX 480 today, AMD should be releasing two other Polaris cards in the near future. The RX 470 will use the same Polaris 10 GPU as the RX 480, but it’ll lose four compute units, bringing the shader count down to 2,048. It’ll also be available with 4GB of GDDR5 RAM only. AMD isn’t releasing full details of this card today. From the outside, this card looks exactly the same as the RX 480.
The RX 460 is a tiny card built around the Polaris 11 GPU. It has 14 of that chip’s 16 GCN compute units enabled, and it won’t need an external power connector. We also don’t have full details of that card yet. It is a fair bet that both of these cards should slot in under the $199 price point established by the 4GB RX 480, though. We expect to learn more about these products over the course of the summer.
Now that we’ve fixed a course on Polaris, let’s see how the RX 480 performs.
Our testing methods
As always, we did our best to deliver clean benchmarking results. Our test system was configured as follows:
|Motherboard||Asus X99 Deluxe|
|Memory size||16GB (4 DIMMs)|
|Memory type||Corsair Vengeance LPX
DDR4 SDRAM at 3200 MT/s
|Chipset drivers||Intel Management Engine 184.108.40.2065
Intel Rapid Storage Technology V 220.127.116.111
|Audio||Integrated X99/Realtek ALC1150
Realtek 18.104.22.16825 drivers
|Hard drive||Kingston HyperX 480GB SATA 6Gbps|
|Power supply||Fractal Design Integra 750W|
|OS||Windows 10 Pro|
|Driver revision||GPU base
|AMD Radeon RX 480||Radeon Software 16.6.2 beta||1120||1266||2000||4096|
|Sapphire Radeon R9 380X||Radeon Software 16.6.2 beta||–||1050||1375||4096|
|Gigabyte GeForce GTX 960 4GB||GeForce 368.39||1228||1329||1753||4096|
|Gigabyte GeForce GTX 970||GeForce 368.39||1076||1216||1750||4096|
|Gigabyte Windforce GTX 980||GeForce 368.39||1228||1329||1753||4096|
Our thanks to Intel, Corsair, Asus, Kingston, and Fractal Design for helping us to outfit our test rigs, and also to AMD and the Nvidia board partners who sent us graphics cards for testing.
For our “Inside the Second” benchmarking techniques, we use the Fraps software utility to collect frame-time information for each frame rendered during our benchmark runs. We sometimes use a more advanced tool called FCAT to capture exactly when frames arrive at the display, but our testing has shown that it’s not usually necessary to use this tool in order to generate good results for single-GPU setups. We filter our Fraps data using a three-frame moving average to account for the three-frame submission queue in Direct3D. If you see a frame-time spike in our results, it’s likely a delay that would affect when a frame reaches the display.
Aside from the Radeon RX 480, our test card stable is made up of non-reference designs with boosted clock speeds and beefy coolers. Many readers have called us out on this practice in the past, so we want to be upfront about it here. We bench non-reference cards because we feel they provide the best real-world representation of performance for the graphics card in question. They’re the type of cards we recommend in our System Guides, so we think they provide the most relatable performance numbers for our reader base. When you see “GTX 960” or “GTX 980” in our results, for example, be sure to remember that we’re talking about custom cards, not reference designs.
Each title we benched was run using DirectX 11. We understand that DirectX 12 performance is a major point of interest for many gamers right now, but the number of titles out there with stable DirectX 12 implementations is quite small. We’ve had trouble getting Rise of the Tomb Raider to even launch in its DX12 mode, and other titles like Gears of War: Ultimate Edition still seem to suffer from audio and engine timing issues on the PC. DX12 also poses challenges for data collection that we’re still working on. For a good gaming experience today, our money is still on DX11. That said, we probably need to revisit DX12 performance in the near future and see how the tables turn.
Sizing ’em up
Do a bit of quick math, and you end up with the theoretical peak performance numbers for the following graphics cards:
|Radeon RX 480||41||182/91||5.1||5.8||256|
|Sapphire Radeon R9 380X||33||133/67||4.2||4.3||256|
|Radeon R9 290||61||160/80||3.8||4.8||320|
|Radeon R9 Fury X||67||269/134||4.2||8.6||512|
|Gigabyte GeForce GTX 960||32||82/82||2.6||2.6||112|
|Gigabyte GeForce GTX 970||63||126/126||4.9||4.0||224|
|Gigabyte GTX 980||85||170/170||5.3||5.4||224|
We won’t be testing every card in the table above, but our theoretical numbers offer some interesting insight about how the RX 480 stacks up against its AMD stablemates and the Nvidia competition. Thanks in part to its ROP count, the RX 480 handily outpaces the Tonga-powered R9 380X in raw fill rate, but it falls a bit short of the enormous shader arrays on the R9 290 and the Fury X. The RX 480 also achieves higher theoretical int8 texturing rates than everything in the table save for the R9 Fury X, even if its performance on more complex textures is still limited by GCN’s half-rate throughput with fp16 data types.
Now that we’ve taken stock of the RX 480’s theoretical performance, let’s take a look at some actual numbers generated by the Beyond3D test suite to see how these cards behave in practice.
The RX 480 is clocked higher than the Tonga-powered R9 380X, and its slightly larger shader array gives it slightly more pixel-pushing power than that card. Nvidia’s cards maintain their long-running advantage here.
Hand the RX 480 an incompressible texture, and its memory bandwidth numbers unsurprisingly outpace everything else on the board. When the GeForces can employ their delta-color-compression, however, the competition heats up. Still, whatever new DCC mojo AMD has added to the RX 480 appears to be one of several factors contributing to the RX 480’s very solid performance increase over the R9 380X.
Hm. Despite having more texturing units onboard than Tonga does, the RX 480 seems to run into a wall at about the same peak rate as its predecessor. Perhaps there’s another bottleneck at work somewhere for this test.
Now here’s something interesting. In our past graphics card reviews, AMD’s cards have fallen behind in this polygon throughput test when we’ve presented them with work in a strip format. Here, the RX 480 sets itself apart by delivering similar rates for both formats, and at faster rates than even the R9 Fury X before it. Perhaps we’re seeing that fancy new Primitive Discard Accelerator at work.
Polaris 10’s theoretical peak shader performance is pretty potent, and our ALU tests confirm it. The RX 480 edges past even the GTX 980 in these tests.
Now that we’ve examined these cards’ theoretical performance, let’s put them to the test in some real-world gaming scenarios.
Grand Theft Auto V
As we learned in our recent review of Nvidia’s GeForce GTX 1080, Grand Theft Auto V runs pretty well on a wide range of hardware. This time around, we dialed back the resolution to 2560×1440 and left all of our other graphics settings the same. Pardon our wall of screenshots:
Click the buttons above to cycle through our frame time plots. As you can see, all of the cards we tested run GTA V quite smoothly at these settings—there are no large spikes at regular intervals that would suggest hitching, or wide swings up or down in the overall trend of the data. These graphs are the only place you want to see something flatlining, and all of the cards we tested turn in nearly-ideal results.
Going by the measure of potential performance that average FPS provides, most of our cards start off with a strong showing in GTA V. The Radeon RX 480 and GeForce GTX 970 are within a hair’s breadth of one another, while the R9 380X is further back and the GTX 980 pulls far ahead. Average frame rates don’t tell the whole story, though, so let’s have a look at the 99th-percentile frame time each card delivered during our tests. This measure shows how much time the card took to produce each frame for 99% of our benching run.
Not a bad result here, either. The GTX 970 delivers an ever-so-slightly better 99th-percentile frame time than the RX 480, while the R9 380X and GTX 960 bring up the rear. The GTX 980 is clearly in a class of its own with GTA V at this resolution, maintaining well over 60 FPS for most of our test.
These “time spent beyond X” graphs are measures of “badness”—the amount of time that a card was displaying animation that may have been less than fluid, or at least less than perfect, during our test.
If a frame takes longer than 50 ms to produce, for example, the time the card has to spend churning away on that difficult scene represents a drop below 20 FPS—a slowdown that most users will definitely notice. If a card consistently produces frames without taking more than 33 ms on each one, we know it’s running at 30 FPS. If frames take longer than that to get through the graphics pipeline, that slowdown can produce judder and other ugliness with vsync enabled on a 60-Hz monitor. Finally, if a card is completing frames in 16.7 ms or better, the corresponding animation on screen is moving along at 60 FPS or more. 60 FPS is the golden mark we’d like to see a card achieve (or surpass) for each and every frame.
Happily, none of our cards spend any time past the 50-ms or 33-ms thresholds. The GTX 980 also gets a gold star here for not spending any time past the 16.7-ms mark. The GeForce GTX 970 spends just a fraction of a second under 60 FPS, while the RX 480 spends, well, a larger fraction of a second there. Both the Radeon R9 380X and the GeForce GTX 960 have a far harder time of it—they spend considerable amounts of time working on frames that caused the fluidity of animation to drop.
Despite its 2013 release date, Crysis 3 is still plenty capable of putting the hurt on modern graphics cards. In fact, our GTX 960 was so overwhelmed by the game that it only sustained an average of 22 FPS at our chosen 2560×1440 resolution, for some reason. We’re omitting it from these numbers as an outlier.
Once again, the RX 480 is neck-and-neck with the GTX 970, both in average frame rates and in its 99th-percentile frame time. Moving on.
In our measures of “badness,” the R9 380X spends a fair amount of time on frames beyond the 50-ms and 33-ms marks, and it’s the only card that has any kind of problems with keeping frame times below 16.7 ms for an extended period. Our data reveals that the GTX 970 and RX 480 both had some troublesome frames to churn through, but nothing to the degree of the R9 380X. The GTX 980 turns in a great performance.
Rise of the Tomb Raider
Lara Croft’s latest quest is one of the more demanding games we’ve come across recently. It posed a considerable challenge for even the GeForce GTX 1080 in our recent review, so we’re dialing it back to 1920×1080 for this batch of cards.
Sensing a trend yet? The GTX 970 and the RX 480 deliver pratically identical performance in Rise of the Tomb Raider. The GTX 960’s frame time plot gets a little furry, but the R9 380X’s is well and truly all over the place. This is one situation where the value of frame-time benchmarking really shows itself: despite having identical average frame rates, the R9 380X and the GTX 960 have vastly different 99th-percentile frame times. As you might expect, playing RoTR on the R9 380X isn’t nearly as smooth an experience as it is on even the GTX 960.
The R9 380X spends a noticeable amount of time past the 33.3-ms mark. Moving to the 16.7-ms threshold, the GTX 980 is the champ, while the RX 480 and GTX 970 spend similar-but-significant amounts of time working on difficult frames that would result in a drop below 60 FPS. The R9 380X and GTX 960 bring up the rear.
Like GTA V, Fallout 4 is another game that’s not terribly hard for most graphics cards to run well. We tested this game at 2560×1440 while leaving all of the visual quality settings the same as in our GTX 1080 review.
None of the cards in our tests have any trouble running Fallout 4 smoothly. The RX 480 overtakes the GTX 970 in our average frame rate measure, and it also matches the beefier GTX 980 in our 99th-percentile frame time metric. The GTX 970 takes ever-so-slightly longer to deliver most of its frames, while the GTX 960 and R9 380X bring up the rear.
In our “badness” measures, the action really starts to happen at the 16.7-ms threshold. The RX 480 and GTX 980 are neck-and-neck, while the GTX 970 is just behind. To be clear, none of these three cards are having issues that would severely impact animation smoothness in-game. The R9 380X and GTX 960 continue to skew our graphs, though.
The Witcher 3
Aside from running this title at 1920×1080, we left our graphics settings the same as they were in our GeForce GTX 1080 review. Ignore the resolution in our first setings screenshot.
GeForce GTX 960 aside, none of the cards we tested The Witcher 3 on have any problems running Geralt of Rivia’s adventures smoothly. The RX 480 turns in a higher average frame rate than the GeForce GTX 970, but its 99th-percentile frame time is a tad higher. Still, the GTX 970, RX 480, and R9 380X are all pretty closely matched here. The GTX 960 is off in a corner somewhere, while the GTX 980 delivers somewhat better average frame rates and a slightly lower 99th-percentile frame time than the rest of the peloton. No surprises there.
The RX 480 and the GTX 970 both spend about the same amount of time past 16.7 ms working on challenging frames in this title. Once again, we can say they’re pretty closely matched. On to the next.
If our GTX 1080 review is anything to go by, Agent 47’s most recent missions are quite taxing for most graphics cards to render smoothly. We tested this demanding title at 1920×1080.
Here, the RX 480 really stretches its legs, even in DX11 mode. Its average FPS figures are just slightly behind the GeForce GTX 980, and its 99th-percentile frame time is even a smidge better than that card’s. The GTX 970 delivers a solid 99th-percentile result, but its average FPS numbers suggest that its performance potential is considerably lower than the RX 480’s in this title. Meanwhile, the Radeon R9 380X nearly matches the GTX 970, while the GTX 960 falls behind.
At the critical 16.7-ms threshold, the GTX 980 leads the pack, but the RX 480 doesn’t spend much more time working on frames that take longer than that to render. The GTX 970 and the R9 380X have a somewhat harder time of it, while the GTX 960 just isn’t up to the job. We’re not sure what’s up with our particular sample of GTX 960, but it doesn’t seem all there for some reason. Might have to take it out behind the shed after this review is over.
The power consumption of Polaris chips is another major point of interest for us. We’ve seen AMD demo systems running Star Wars Battlefront before with tantalizing numbers on the power meters beside them, so we were excited to replicate that experience in our own testing environment. To test power draw, we fire up a real game, Crysis 3, to show how much juice each graphics card needs under a typical workload.
Well, that wasn’t quite what we expected. The RX 480 doesn’t draw much more power at idle than the competition, but it’s worth noting that Polaris aside, all of the cards we’re testing are using GPUs fabricated on a 28-nm process. Polaris 10 doesn’t seem to benefit much from the move to 14-nm fabrication as far as idle power draw is concerned.
Fire up Crysis 3, and the RX 480 draws as much power to do its thing as the GTX 970. If we consider the Radeon R9 290 (or the R9 390) the GTX 970’s most natural competitor on the 28-nm process node, AMD has shaved anywhere from 100W to 140W off that card’s power draw while delivering the same performance, if our past reviews are any indication.
An improvement that large is impressive until one considers that the GTX 1080s in the TR labs need 265W-300W of total system power to do their thing. To be fair, our power numbers are one measurement taken under one particular load and in one particular testing environment, and modern power management is a complicated thing with many input variables. All that said, our gut impression upon seeing these numbers is that Pascal is frighteningly efficient, more so than the GTX 1080 taken in isolation might have suggested. Our completely wild hunch is that Nvidia has tons of headroom to play with in designing a Pascal GPU to target this price class, if it wants to.
Noise levels and GPU temperatures
Before we take a look at the RX 480’s cooling performance, we should make a quick note about our test environment. The closed-loop liquid cooler on our test system’s CPU produces about 40 dBA on its own, so our noise floor is artificially high. Noise numbers near or below 41 dBA indicate that a given graphics card isn’t exceeding that rather high value to begin with, at least. The ambient temperature in our testing environment was about 73° F during our tests.
We’re not fans of blower-style coolers for the most part, and the RX 480’s isn’t doing anything to change our minds. While the card runs at or below our 40-dBA-ish noise floor at idle, its load noise level climbs to 51 dBA—just short of the triple-fan cooler on our Gigabyte Windforce GTX 980. Despite all the sound and fury, the RX 480’s load temperatures reached 83° C under load, too.
One of the characteristics we’ve come to associate with efficient graphics cards is polite manners in the noise department, but the RX 480’s reference design doesn’t deliver. The blower fan isn’t pleasant-sounding, either—it’s grindy and obtrusive. We hope AMD’s board partners have custom coolers in the works that deliver a better experience in the noise, vibration, and harshness department.
Before we get too deep into our thoughts and feelings about Polaris and the Radeon RX 480, let’s break out a couple of our famous value scatter plots. We’ll first look at the performance-per-dollar these cards offer going by the potential measure of average FPS. To accurately reflect the changes in price many of the graphics cards in our tests have experienced since their release, we’ve averaged the prices of those cards on Newegg right now. We used the RX 480 8GB card’s $240 suggested price for these results.
The RX 480 8GB card we tested delivers a hair more performance potential than the GTX 970, but at a significantly lower price than the average going rate for that card on Newegg right now. We’d have to test the 4GB RX 480 to be truly sure of its value proposition, but just imagine a similar dot at $200, and AMD might have quite the hit on its hands.
Going by the measure of performance potential that average FPS provides, the RX 480 is sometimes slightly faster than the GTX 970, and it’s sometimes a little slower. Those familiar with the long-running battle between the GeForce GTX 970 and the Radeon R9 290 should be getting a sense of deja vu right now. What’s nice about the RX 480 is that AMD is extracting that kind of performance from a die that’s roughly half as large as Hawaii with what is, in many respects, a smaller engine inside.
Next, let’s take a look at our 99th-percentile-FPS-per-dollar graph. We take the geometric mean of the 99th-percentile frame time each card delivers across our test suite and convert it into FPS to make our higher-is-better logic work.
The RX 480 just barely squeezes past the GTX 970 in this measure, but its appealing price tag helps it plant a flag that’s the highest and leftmost on our 99th-percentile FPS chart. Once again, imagine a little dot in a similar place on the $200 line. That’s pretty incredible performance in our advanced metrics for a card at this price point.
Indeed, what’s most notable about the RX 480 compared to past Radeons of any price is its consistently smooth frame delivery. Where AMD’s older cards have trailed the GeForce competition in delivering smooth gameplay—often by wide margins—the RX 480 chalks up a huge improvement in both our advanced 99th-percentile frame time and “badness” measures compared to the Radeon R9 380X. We’re completely comfortable calling the RX 480 the equal of Nvidia’s GeForce GTX 970 in those regards. That’s excellent progress from the red team, and we hope that whatever mojo is responsible for this turn-around works its way into every future AMD graphics card.
On the other hand, the RX 480’s power consumption and noise figures aren’t as rosy as we might have expected them to be. Our power consumption tests today aren’t perfectly comparable to those in our older reviews, but one Radeon R9 290-powered system with a similar CPU and motherboard drew about 400W under load when we reviewed that graphics card as part of a larger test a while back. The Radeon RX 480 delivers similar performance to that card while shaving about 140W off the total power draw of our system. TR readers helpfully point out that using board power as a rough guide, the RX 480 is about 90% more efficient than the R9 290 before it, considering the performance we measured. Either way, that figure seems to fall short of the 2.8X performance-per-watt increase that AMD often touted with Polaris.
If you care about noise and heat, the performance of the reference cooler on the RX 480 will probably leave you wanting, too. It’s quite loud at full tilt, and it lets the GPU underneath get quite hot under load. Perhaps AMD needs to set the designer of the Wraith cooler loose on its graphics cards, as well. Like the Founders Edition GeForce GTX 1080 we just reviewed, we think most buyers will probably be best off waiting to see what sort of custom cooling AMD’s board partners have in store before dropping two Benjamins or more on a reference card.
Like we’ve noted, the 4GB version of the RX 480 delivers efficiency and performance figures that are both pretty similar to a GTX 970, and it pulls off this feat for a $200 suggested price. We think that’s a nice place to be. A performance jump like this hasn’t happened around this price point for a long, long time, and it’s quite welcome. The RX 480 8GB offers a bit of “future-proofing” in memory-hungry games like Rise of the Tomb Raider for $40 more. Either card should meet Oculus’ and HTC’s recommended specs for a Rift or Vive, so aspiring VR junkies can put the $65 to $150 extra that a GTX 970 would command right now toward a VR headset. Regular gamers can just enjoy fast, smooth gameplay in traditional titles and pocket the cash.
Right this second, the RX 480 sets a new bar for performance and smoothness at its price point, and it’s undoubtedly the midrange card we’d recommend to most—at least, once AMD’s board partners get their hands on it. It’ll be interesting to see what Nvidia’s answer to the RX 480 will be, but for now, we’re excited to see where AMD will go now that it has its eyes on the stars.