Just a few months ago, the world was shown AMD’s new baby, the Radeon R9 Fury X. With a new architecture on board (codenamed Fiji), a new memory technology (High Bandwidth Memory, or HBM) and a fully enclosed water cooler attached, it was easily the most interesting product launch in recent years. Not only is its diminutive size a bit weird to consider especially given how well it performs, the form factor allowed it to fit into many more chassis and setups than a traditional, long flagship graphics card would have. Today, AMD has a variation of Fury X to share with you and it is yet another interesting product to consider, especially given its size, power setup and cost.
One of the issues that AMD faces today is that of perception. The perception is that the company is spinning the wheels furiously trying to gain traction in the market once more, fighting against NVIDIA’s aggressive rollout of Maxwell hardware and trying to woo the notebook designers back in at the same time. Dwindling OEM desktop sales don’t help the picture at all, but the company can take solace in the fact that the enthusiast market is always present and does, in fact, purchase more and more hardware year-on-year. Targeting enthusiasts first for the HBM and Fiji rollout was a smart move. These early adopters are the people who keep on talking about what AMD’s doing and what they could be doing a year from now, and Radeon R9 Fury and Fury X are both enthusiast-class products that are truly unique.
Trailing NVIDIA with 18% market share in Q2 2015 (Fury, Fury X and NANO shipments have yet to affect this graph), AMD is not in a position to gamble with their remaining market share. They’re in a place that restricts movement quite heavily because they can’t shift as many units as they’d like compared to NVIDIA, which means that their strategy now is really limited to disruptive technologies and picking out wins from niche, but lucrative or growing markets.
HBM is one of those disruptive technologies, but it has better applications outside of a 3D gaming-orientated product, which is why the gaming performance results tally up so closely to the Geforce GTX 980 Ti in most reviews of the Fury X. The potential for crushing performance is there, but AMD needs to extract it with the help of several other components, most notably Microsoft’s DirectX 12 API and the upcoming multi-platform Vulkan API.
Thus, NANO is a disruptive product for a specific niche, fitting into AMD’s current strategy of not trying to one-up the competition with products that are more or less similar as they’ve done in the past. It has to be. There’s simply no other way that it could exist without AMD being in the place they find themselves in the market today.
Radeon R9 NANO is a Fury X placed on to a smaller circuit board and mated to an air cooler. Its a fairly simple idea for sure, and one wonders why AMD didn’t launch with this alongside Fury X in the first place (my money is on prior agreements with Cooler Master on the bundled water cooler sales). With the circuit board only extending a small bit further over the PCI Express slot, R9 NANO will find itself in lots of different chassis and systems, both where space is limited and in places where it isn’t.
AMD told me that they engineered R9 NANO to be about 30% faster than the Radeon R9 290X, which is an odd comparison point, but one that makes sense in later slides. With a full compliment of 4096 stream processors on-board, this is a full Fiji implementation. No hardware is disabled to meet performance or thermal design power (TDP) restraints, although clock speeds are lowered to meet temperature targets. R9 NANO also appears to be cheaper to make for AMD, with changes in the board design possibly limiting how much performance is left for overclockers to eke out.
The comparisons to R9 290X began to make sense when AMD began to show off pictures of the actual card – they are basing their performance, power, and efficiency improvements on the stock or reference R9 290X with the reference cooler design. This muddles things up a bit for any comparisons to current generation products like the R9 390X, because there are several variants on the market today that are quieter and somewhat faster than the reference R9 290X cards launched back in October 2013. However, there are no variants of Hawaii cards that do fit into most ITX chassis, so there’s that.
Cooler and quieter than the reference R9 290X cooler is the gameplan here. AMD wants to rid itself of the stigma of those reference coolers as much as possible and they’re doing it with a slight leniency in terms of operating temperatures. The R9 NANO is designed for a 75º Celsius operating temperature and the cooler design exhausts hot air both into and out of the chassis. Like the middle-mounted blower design on the Geforce GTX Titan Z and GTX 690, this presents an interesting challenge for some builds because heat could potentially get trapped inside the case.
Lets get into the price very quickly – R9 NANO costs the same $649 (approx. R8500 today) as its Fury X counterpart. That sort of price point attracts high-end enthusiast buyers who prefer quality over a lower price point, which is why AMD makes use of premium materials for the cooler design. The price draws it right up with the Geforce GTX 980 Ti, but it matches the prices for the reference cooler version and not any of the custom offerings from NVIDIA’s board partners. Because the GTX 980 Ti reference cooler is plastic and no longer sporting a magnesium finish, this results in an interesting quality comparison between the two cooler designs. Preference for one over the other will come down to personal tastes. R9 NANO’s is certainly cheaper overall compared to Fury X which is saddled with a heavy, thick radiator.
One disappointment I had with this reveal is that the face plate is no longer easily replaceable. You can take off the front part of the metal shroud, but this also drags up the fan with is. AMD told me that custom designs are possible and that the 3D schematics are available to anyone who asks for them. You’ll just have to mount on the fan yourself if you decide to switch it out. I suppose another is that the Radeon logo doesn’t have a red LED to light it up when the card is running, nor are there GPU Tach LED status lights near the power connector. Very odd choices considering that this is a premium product.
In a nutshell, here’s what this card is all about. AMD is targeting a 175 watt TDP, putting it just over the typical power requirement for many mini ITX-class graphics cards in the same segment that it sits in (size-wise, not price-wise). Utilising a 8-pin PEG power connector, the NANO is just a little over the 150 watt power draw recommendation for the ATX specification, which it makes up for by drawing power from the PCI Express slot.
With another 50W potentially available for overclocking, we end up with what is probably a board design peaking at 225 watts. R9 Fury X tops out at a typical power draw of 275W, with 375W being possible with some modifications and overclocking applied. That distinction becomes important later, because AMD told me something else that was interesting with regards to how this card overclocks.
Performance-wise, AMD expects R9 NANO to be on average 30% faster than a typical Geforce GTX 970 designed for mini ITX chassis. With Gigabyte’s GTX 970 “mini” retailing for $300 in the US, the performance comparisons are interesting, especially when you notice that AMD is comparing UltraHD 4K results here. A 30% performance increase at 4K is fine, but that’s not the resolution that most people will use this card at. I’m expecting that reviews of NANO will show that the gap closes considerably at 1440p and 1080p, where the GTX 970 performs better, and might be faster overall.
As you can see, the specifications for R9 NANO largely tally up with the R9 Fury X. However, there is one change that tells us a little about the expected performance of this card – the clock speed goes up to 1000MHz, whereas Fury X had an upper limit of 1050MHz. When I was listening in on AMD’s conference call about Fury X before its launch, they were confident that Fury X would hit that limit just about all of the time thanks to the temperatures being kept in check. With NANO, that’s not the case, as I was told that the expected average clock is “a little higher than 900MHz in a typical scenario”.
The bigger picture
There’s a catch with NANO’s size and power efficiencies – AMD told me that R9 NANO will more or less match R9 Fury at stock clocks for performance comparisons. If you ever encounter a R9 Fury with a custom board design that lets it fit into most ITX chassis, there’s no real reason to spend extra for a R9 NANO, especially if you’re not going to be aiming for playing games at 4K. Without overclocking involved, there is performance left on the table that you’re not getting from the Fiji GPU. The thermal and power constraints on the GPU leave it a bit underpowered compared to the R9 Fury X, which costs the same amount of money.
There are some concerns about overclocking headroom that potential buyers might have as well. When I was told that increasing the power limit of NANO should allow for performance to improve to almost match match the Fury X, I began wondering about that 8-pin connector. Is it possible that AMD is allowing R9 NANO owners to overdraw on the ATX power specification if they want to match the Fury X’s stock performance? I certainly think so. Will that result in identical performance compared to Fury X on this card? Perhaps, perhaps not. R9 NANO only has four power phases compared to Fury X which has eight by default, so I have no idea how much headroom there actually is.
UPDATE: When asking AMD in a series of follow-up e-mails about the power delivery of R9 NANO, this is the response I received:
“Most modern high end PSUs with merged rails can supply 300W through an 8-pin power connector. While we have chosen to respect the 150W ATX guideline for the default product configuration it should not be difficult to find a power supply that supports a 300W 8-pin connector. This is MORE than enough headroom to play with the overdrive settings.”
So that answers that. If you want to overclock R9 NANO to Fury X clock speeds and performance, you’re going to need a power supply that can safely overdraw from the 8-pin PEG connector. Most high-end name brand 500W PSUs rated for at least 80 Plus Gold efficiency with 8-pin PEG connectors and decent quality cables should be enough for this card.
Overall, it seems like AMD is on the right track with this release and NANO is certainly in a league of its own when it comes to performance in a small package. The price compared to Fury X is slightly disappointing because performance isn’t on par between the two products, but I can understand the drawbacks of sticking a GPU capable of a 375W power draw into a chassis that is roughly 15 litres in size.
The final line, I guess, is that if you have the space for the attached radiator, always buy the Fury X. It costs nothing extra, and you’re not limited by power draw constraints. If you don’t have the space for the radiator, buy the R9 NANO, accepting the power tradeoffs that come with it unless you also have a high quality PSU to go along with it. If you don’t have space for a radiator, but play at 4K with some lowered settings and want to save $100-ish on your purchase, buy the R9 Fury if it fits.
Towards the end of the media briefing with AMD, we were all told that initially the R9 NANO rollout will be based only on reference designs, but that partners would later be able to put out their own versions of this SKU. You know what that means, right? There’s clearly a Fury X variant on the horizon that isn’t eight inches long and weighted down with a radiator. That product might be interesting.