Inside the AMD R9 290

AMD and Nvidia have played leapfrog with product performance for a long time now, if you include the ATI era. With an established six-month window between their respective releases, domination is often fleetingly short for either side. Over the past decade, it's often been Nvidia out in front, with AMD/ATI chasing. With the arrival of the new R9 series, AMD squarely aims to take back the initiative.

Inside the AMD R9 290

Top Dog
Technically, for those following GPU development, the AMD R9 290 isn't the flagship card of this new series. That accolade goes to the R9 290X. That said, in all but a few very minor clock values and shader numbers, this is best AMD has on offer, and a card that many gamers will be salivating over in the coming months.




What's also important about this and its R9 290X brother is that while AMD launched a complete range of new hardware, from the R5 210 upwards, these are the only ones with entirely new GPU designs in them. All the other cards, including the 280X and below, reworked cores from the 8000 series and before.

The new Hawaii Pro and XT cores are a major leap in complexity for AMD, containing 6.2 billion transistors, a 512-bit memory pathway and more than 2560 shaders.

Given all this power, when I was first presented with this card in early October it was positioned against the Nvidia Geforce GTX 770 - an impressive card but hardly the best the opposition offers.

Since that time, new drivers and increasing confidence from the AMD camp have now shifted the R9 290 to a more direct comparison with the GTX 780 series that includes the Titan class cards. By definition, that moves the R9 290X into an as yet unmatched performance level above the Nvidia 700 series range.

AMD has also released pricing for the R9 290, which at £319 seems oddly cheap for such a high end device.

So what's the R9 290 all about, and does it live up to any of AMD's hype?

The AMD R9 290
With the level of complexity that a modern GPU involves, all new designs are by definition derivative. The Hawaii Pro therefore is a direct offspring of the Tahiti XT that preceded it. except it's enhanced in almost every aspect.

The key features of the new R9 290 GPU include:

  • 2,560 stream processing units
  • Up to 947MHz engine clock
  • 6.2 billion transistors
  • 4GB GDDR5 memory
  • Up to 5Gbps memory clock speed
  • 320GB/S memory bandwidth (maximum)
  • 4.85 TFLOPS Single Precision compute power
  • API support for DirectX 11.2, OpenGL 4.3 and Mantle


While more shaders is always helpful, probably the most critical enhancement that this design sees over the HD 7970 is the amount of bandwidth available: a staggering 320GB/S. This is achieved with a relatively low GDRR5 clock speed by having a monster 512-bit wide memory pathway - something that could easily see AMD offer a +400GB/S in a later model. If you're intending to support 4K Eyefinity gaming where you're aiming to render a virtual display of 12888 x 2048 at more than 60fps, then you'll need that much.

Having this amount of power in a card isn't without the odd catch. The reference board is 27.5cm long, which is wider than a full ATX motherboard. It also requires both an eight- and six-pin PCIe power lines, delivering the 230 watts that the card needs as a minimum. At idle the power management features of this design reduced the power footprint to just over 62W, which is almost identical to the prior HD 7970.

When stressed, the wattage that the card consumes is converted mostly into heat, and the massive heatsink/shroud combination is designed to vent that out through the double end plate.

As AMD cooling designs go. this isn't radically different from the HD 6970 layout and relies on a single fan to drive air down over the GPU and GDDR5 to stop it getting too hot. A typical idle temperature is 35C. and when the pixels are being crunched, it can easily go up to 95°C. Temperatures like that cause the fan RPM to climb and stay high, so I'd expect the retail cards to focus on cooling the card so it doesn't get that warm.

If AMD can solve that problem, then the sensor-driven throttling the card exhibits could allow a thermally well managed design to inherently outperform a less effectively cooled one.

Much of what I've talked about so far isn't a major departure from previous AMD video cards, so what has AMD put in here that's actually new?

New Thinking
Along with a good dash of the old and familiar, these cards do also have some radical new things that we've previously not seen on AMD products. Rather than concentrating on one specific area, AMD has spread these enhancements around to encompass a wider scope than just those who play games. Here are some of the highlights:

Enhanced Eyefinity
I've been a big fan of this virtual display technology from the outset, but I've been disappointed by a number of significant catches that AMD seems to slide underneath what's generally a great feature.

The R9 290 supports six displays out of the box, with four connections provided on the card. The standard configuration is dual DVI and single HDMI and DisplayPort outputs. That allows you to get triple displays without using DisplayPort, and if you own a monitor that accepts that input, you can chain it on to a maximum of six.

Enhanced CrossFire
The back-channel method of connecting the frame buffers using ribbon cables is now dead, because with the advent of PCIe 3.0 the bus has more than enough bandwidth to accomplish the task. In this new mode slave cards will render their frames directly into the memory of the master card. AMD claims this is much more efficient and allows dual cards to approach 2x speed scaling.

True Audio Technology
If you've not noticed, the sound card as a concept has died. That was because Microsoft kept messing with the sound API, and Realtek put a good enough audio chip on almost every motherboard from the outset.

AMD sees game audio as an opportunity, and on the R9 290 it's added True Audio Technology, a dedicated programmable audio DSP. The purpose here is to improve the audio experience by processing the available audio streams independently, providing strong spatial effects even with stereo headphone output.

Normally the calculations for offering 3D audio effects or real-time reverb over headphones and 2.1 speakers are too great for typical game use. but with the TrueAudio DSP, it can be achieved with little or no significant overhead. GenAudio, which developed this technology, intends to provide plug-ins for the common PC audio middleware so that developers can easily incorporate the option in their titles.

Mantle API
Another new aspect to these Volcanic Islands designs is the introduction of the Mantle API, a software interface layer designed to get more out of the available hardware.

For those who don't live and breathe game coding, this is the means by which most games are connected to their game engines. By having the API layer sitting between the game code and the hardware, the generation of 3D imagery can be managed in a way that allows developers to worry less about the vast range of video card capabilities.

The caveat to doing this is that software rarely fully exploits the hardware, because the API is designed for all possibilities, not the specific one in your PC.

In a presentation that AMD gave to journalists on its new hardware, it was explained in the context of the AMD video hardware that it provided for the Xbox 360 and upcoming Xbox One and PS4. People have long questioned how the Xbox 360 can deliver such high-quality graphics and consistent frame rate given its modest GPU and CPU facilities.

The answer, according to AMD, is that when you've got complete control over the hardware and are not working through an API like DirectX, the GPU hardware can be run much harder than it is currently on a PC.

AMD's answer to that is the Mantle API, where the software developer can get direct hardware access and therefore make a PC seem much more like a games console. The trouble, as is often the issue with APIs, is getting developers to use methods that they're not familiar, having invested massively in either DirectX or OpenGL previously.

EA has been convinced, via a sum reportedly in the $5m to $8m range, to bind Mantle into its recent flagship release. Battlefield 4. The official Mantle patch for Battlefield 4 won't be out for a month or so. and when it arrives it will be massively scrutinised. Should the Mantle version of Battlefield 4 deliver a substantial boost over its traditional API version, then a very large number of developers are likely to take a significantly greater interest.

Just before I finished writing this article, it was announced that joining the two original converts to the Mantle cause, EA and Activision, are Cloud Imperium Games, Eidos-Montreal, a part of the Square Enix Group.and Oxide Games. Perhaps that hints that the Battlefield 4 Mantle release will yield dramatic improvements, even if we've yet to see these numbers officially.

The release of Mantle enhanced titles from these and others could open the door to PC games playing much more like console titles and side line the DirectX API.