Today, AMD is prepping a new GPU lineup based on a refresh of its Polaris (aka GCN 1.4, aka 4th-generation GCN) architecture. We’ll cover the other announcements, like the RX 550, RX 560, and RX 570 in a separate post. In this review, we’re diving into the RX 580 and how it performs relative to 2016’s RX 480 and the GTX 1060.
The RX 580 will be familiar to anyone who followed AMD’s GCN 1.4 architecture, aka Polaris. It’s built on the same core as the RX 480, with exactly the same features: 2304 GPU cores, 144 texture mapping units (TMUs), and 32 render outputs (ROPs). It’s a well-fed GPU, with 256GBps of memory bandwidth, and a full 8GB of RAM (a 4GB option will also be available).
AMD says the only adjustment they made to the base design were some silicon changes to enable higher clock speeds. On paper, the RX 580 is a modest improvement over the RX 480, with a base clock of 1257MHz and a boost clock of 1340MHz, compared with 1120MHz and 1266MHz for the old RX 480. That’s an increase of 1.12x on base clock and 1.08x on boost clock, for those of you playing along at home. With the RX 580, however, AMD has introduced GPU core clocks that float above their official specifications. This is nothing new — Nvidia does something similar with its Pascal GPUs — but it’s the first time we’ve seen Team Red take this tack. Our Gigabyte RX 580 runs at a rock-solid 1425MHz, which gives it a 1.13x advantage over the RX 480. That’s enough to conceivably matter, so let’s take a look at what we’ve got.
Gigabyte RX 580 Aorus OC 8GB
Gigabyte’s RX 580 Aorus OC is best described with terms like “beefy.”
If you recall the original reference designs for RX 480, AMD ran into some trouble with regard to power delivery over the 12V rail. There will be no such problems here — AMD’s original RX 480 shipped with a single 6-pin PCIe power connector, while the RX 580 Aorus OC has an eight-pin and a 6-pin connector. No, the GPU doesn’t consume anything like that much power; that’s just how overprovisioned this GPU is. The twin coolers keep the GPU quiet, even under heavy load and the Aorus holds a steady 1425MHz without a single fluctuation in any title we tested.
We’ve reviewed two Gigabyte Nvidia GPUs in the past year, but this is the first time in several years I’ve reviewed one of Gigabyte’s AMD cards, and I’ve got to say, I’m impressed. The coolers are quiet, clock speeds hold steady, and we had no issues whatsoever with the GPU.
All our GPU benchmarks were run using a Gigabyte X99-Ultra Gaming motherboard, 16GB of Corsair DDR4-2400 RAM, a 500GB Samsung 850 EVO SSD, and Nvidia’s 381.65 Game Ready driver. All AMD GPUs were benchmarked on the release driver for the RX 580. While we’ve included a GTX 1070 for reference purposes, we won’t be saying much about its overall performance. At $ 379, it’s too expensive to be a direct competitor against GPUs meant to compete in the $ 200 to $ 270 price bracket. Our benchmark results are shown below; each graph can be clicked to open it in a new window.
Here’s where we come to the proverbial fly in the ointment: The RX 580’s power draw. While the GPU is measurably faster than anything AMD has previously fielded, its power consumption is significantly higher. We retested all of the GPUs below for this review, since we’ve shifted from the Core i7-5960X to the Core i7-6900K for our benchmarking. All measurements were taken at the wall during the third benchmark run of Metro Last Light Redux (Very High Detail, SSAA enabled).
Oof. AMD’s RX 580 may be roughly 10% faster than the old RX 480, but it’s paying for those clock increases, big time. Bringing the GPU clock up by 1.13x increases power consumption by 1.3x compared with the RX 480. What about power efficiency, as measured by how many watts of power are required per each frame of animation?
The watts-per-frame metrics aren’t any better. Nvidia’s GTX 1060 uses 64% of the electricity per frame of animation than the RX 580 does. No AMD GPU currently matches Nvidia’s cards for power efficiency, but the RX 580 uses 1.18x more watts per frame of animation rendered than the RX 480. AMD may have hit higher clocks with their re-tuned version of the RX 480, but they had to give back most of its power consumption improvements to do it. The RX 580 is still more power efficient than the R9 390, but the gap between the two GPUs has slipped considerably and not in a positive way.
What the RX 580 does prove, beyond any shadow of a doubt, is that the GCN architecture in its various incarnations from 1.0 – 1.4 is tapped out. Polaris got a kick from a new process node and better clock scaling, but AMD is slamming into the same problem they had with GCN at 28nm: This GPU is not designed to hit high clock speeds, and it can’t do so without blowing out the power consumption improvements that the RX 480 delivered almost a year ago.
What about price?
Theoretically, the Gigabyte RX 580 Aorus OC we reviewed today will be priced at $ 259. A standard RX 580 Aorus (no OC) is $ 229, the RX 580 Gaming 8GB is $ 219, and the RX 580 Aorus 4GB is $ 199. For $ 20, the difference between 4GB and 8GB of RAM is worth paying for, but we aren’t making any statements on price today and price isn’t weighed in our conclusions.
Our reason is simple: Last year, Nvidia and AMD collectively launched a farcical product refresh cycle, where hardware went out to reviewers literally months before it could be purchased for anything close to the suggested retail price, if you could buy it at all. We didn’t hit the GTX 1080 Ti from this angle, since AMD doesn’t have anything remotely competitive and it’s a halo product with a limited customer base. But in areas where the GPU companies are competitive, we’re suspending all launch-day price comparisons until GPUs are confirmed in-market at something close to the prices both AMD and Nvidia state they should sell for. If AMD and Nvidia find that disagreeable, they’re cordially invited to have products on store shelves on launch day and consistently thereafter.
When AMD launched Polaris last year, it was a significant improvement over the company’s previous midrange hardware, with much higher frame rates and better performance-per-watt. While it didn’t always beat its big brother, the R9 390, it delivered a much better experience compared with AMD’s R9 380. It’s not unusual for these mid-cycle updates to deliver relatively meager performance improvements; the gap between the GTX 680 and the GTX 770 was even smaller than the RX-480-versus-RX-580 gap. If you’ve been holding on to a midrange AMD card from the GCN 1.0 or 1.1 days, you can upgrade to a chip like the RX 580 without looking back, provided your power supply can handle it.
Data from Steam’s Hardware Survey suggests that AMD is targeting its own lower-end users, since the HD 7700 series is listed as AMD’s most popular DX12 GPU, followed by the HD 7900 family. If you have a GPU from either of these product lines, the RX 580 is an enormous upgrade for you.
Unfortunately, higher-end AMD customers aren’t going to find a lot to like here. If you didn’t consider the RX 480 sufficient reason to upgrade from an R9 290/290X/390/390X, you probably won’t be sold on the RX 580, either. And the comparison against the GTX 1060 just isn’t great: The RX 580 is 96 percent as fast as the GTX 1060 at 1080p and 98 percent as fast at 1440p across both DX11, DX12, and Vulkan. In DX12/Vulkan, the situation improves somewhat. Here, the RX 580 is 1.03x faster at 1080p and just 1.01x faster at 1440p. It’s hard to get too excited about AMD matching the GTX 1060 tit-for-tat when the latter is nine months old and the former only manages it at drastically increased power consumption.
If you’re a dedicated AMD gamer with a midrange GPU and no particular concern about power consumption, the RX 580 is a great card. If you’ve been waiting for either company to field a midrange card that specifically goes head-to-head with the GTX 1060, but offers an 8GB frame buffer (and don’t care much about power consumption), the RX 580 is a great card. If you’re evaluating both raw performance and power consumption, the RX 580’s power curve makes it difficult to recommend against what Team Green has available. 14nm GCN variants can hold down lower price brackets for awhile longer, but AMD needs Vega in-market badly.