Conclusion; The Gauntlet Drops
Conclusion; The Gauntlet Drops
The Fury X has a lot riding on its shoulders. This is AMD’s shot to beat or at the very least catch up to NVIDIA after two years of trailing behind their Kepler and now Maxwell architectures. So did they accomplish these loft goals or did the Fiji architecture require more time on the drawing board? Once the dust settled the Fury X ended up being an accomplishment worthy of excitement, one that puts AMD back on the map among the hard-to-please enthusiast crowd.
To bring the Fury X and its associated Fiji architecture to market, AMD had to get creative since they needed a drastic increase in performance without exceeding the nuclear-style thermal figures of their previous Hawaii cores. To accomplish this while still retaining the 28nm manufacturing structure, AMD turned to High Bandwidth Memory and a process node that’s now extremely mature. The end result is a card that features substantially more processing power than its predecessor while consuming less power. If anything Fiji is the first step towards a new generation of integrated GPU / memory designs that facilitate board production while also minimizing bandwidth constraints.
I know you are all here for the raw numbers so here they are….and some are pretty unexpected. Let’s start with the 1440P results first since they exhibit the best and worst of what AMD has to offer here. On one hand the Fury X can offer absolutely very competitive framerates in titles like Far Cry 4, Assassin’s Creed: Unity, Hitman, and Thief. However, it struggles to keep its head above water in Battlefield 4, Shadow of Mordor, Dying Light and GTA V. In some cases it loses quite spectacularly.
Normally a few hiccups could be overlooked but we’re talking about sub-optimal performance within one of AMD’s own “sponsored” engines in Frostbite and two of this year’s most popular games. To me, that points to two potential issues: one being immature drivers and the other being a lack of Render Back End resources which may be bottlenecking the architecture in some situations. However, both are purely speculation on my part but nonetheless there’s obviously something cutting off framerates at the knees in some titles.
At 1440P at least this leads to a tit for tat battle against the GTX 980 Ti with the Fury X remaining close or tied in many games but ultimately falling short due to unexpectedly low performance. However, against the Hawaii-based R9 290X and Grenada-based R9 390X, Fiji offers tangible benefits. But would I recommend this card for 1440P gaming? While making that call will largely depend on the games you play, the GTX 980Ti and even the R9 390X offer much more consistent performance at their relative price points.
While 1440P gaming performance had me wondering if AMD had lost their marbles with the Fury X’s $649 price, the 4K numbers are nothing but vindication for what’s been created here. In UHD scenarios this card is easily able to trade blows with the GTX 980 Ti (if it wasn’t for a lack of optimizations in Dying Light, it would actually be ahead by 2%) and gives the uber expensive TITAN X a swift kick in the pants. Remember, this feat was accomplished at 4K, putting to bed any thoughts about how 4GB of HBM would behave when the flag dropped at ultra high resolutions. Meanwhile, the R9 390X, R9 290X and GTX 980 rapidly vanish in the rear view mirror and aren’t even a factor.
That's not to say the Fury X exhibits dominant 4K gaming behavior in every single game. GTA V, Shadow of Mordor and BF4 still provide a challenge.
I could hypothesize about what’s happening here for another two thousand words but you’ll be spared that. In short it looks like despite the inclusion of a relatively “small” memory footprint of 4GB, HBM’s benefits are beginning to shine through. At the very least it looks like AMD’s driver team is starting to take the first steps towards efficiently allocating memory resources based upon available bandwidth rather than capacity. This process will likely continue to morph into enhanced performance in other applications as the technology matures. I wouldn’t be surprised to see it start positively affecting 1440P results at some point as well.
AMD has certainly met and exceeded some of their goals but there are still some pressing concerns about the Fury X. The performance on tap is undeniable but the fact that it needs a water cooling unit to achieve relative parity with a GTX 980 Ti at 4K while falling behind at 1440P says a lot. It speaks volumes about the challenges AMD faced when trying to integrate HBM modules into an interposer shared with a massive 28nm core but it doesn’t necessarily preclude the possibility of an air-cooled Fury X in the future (hint, hint!).
Despite extolling the compact nature of the Fury X, that all-important All in One water cooling unit will force some users into a tough decision, particularly anyone who uses a smaller form factor chassis. This issue is compounded by the fact that the additional reservoir on AMD’s radiator is quite a bit taller than traditional designs and can effectively interfere with other elements of a build.
That clunky water cooling setup not only adds to installation difficulties but also exhibits a noticeable amount of pump whine. These setups are meant to lower temperatures and quiet down otherwise loud heatsink fans but this one only achieves the former. AMD has stated the whine is a pre-production issue that should be ironed out in retail cards but until we see that first hand, NVIDIA’s solutions will still be quieter in our charts, particularly in idle scenarios.
There are some other limitations as well. An SFF PC alongside a good 4K UHD TV are staples of a next generation gamer’s living room environment but the R9 Fury X’s lack of an HDMI 2.0 port feels decidedly last-gen. Granted, there are some DisplayPort to HDMI 2.0 adapters in the pipeline but that won’t cut it for a $649 card these days. This is a relic left over from previous architectures that should have been addressed.
Another pressing question is whether or not the Fury X and its $649 price tag will put any downwards pressure on NVIDIA’s GTX 980Ti. Given the current situation, I really don’t think that is going to happen since we’re being presented with a statistical tie at 4K and a clear loss at 1440P. Fiji’s availability has yet to be determined as well and if anything, that will ultimately dictate whether or not NVIDIA feels the need to adjust their pricing.
Overclocking headroom wasn’t anything to write home about either as there seems to be just 8-10% of additional core speed headroom and we’ve heard of some samples hitting a wall at a paltry 5%. There’s also a complete lack of available memory increases. Now to their credit AMD claims the massive bandwidth of HBM doesn’t need any overclocking to keep up with the Fiji core and I’ve seen no reason to dispute that.
The R9 Fury X is exactly what AMD needed: a card that directly competes with the best NVIDIA has to offer after nearly two years of playing catch-up. It is fast, surprisingly power efficient, well designed and infinitely easier to live with than the R9 290X. Granted, there were some minor hiccups along the way but every one of them can be easily overlooked in most circumstances. For example, the pump whine is drastically lower when the card is installed into a case, educated case shopping will eliminate any installation problems and overclocking may be improved if AMD allows board partners to feed a bit more voltage into their cores.
In many ways the Fury X and its associated Fiji architecture are well ahead of their time. They point towards what the graphics market will look like in the future rather than what we’re used to right now. Hopefully the newness of the technology behind the R9 Fury X doesn’t hinder its availability because this card has every right to sell like wildfire from day one.