AMD Radeon Fury X Review – We’ve known about Fiji, AMD’s latest and greatest GPU, for months. There’s its High Bandwidth Memory (HBM), capable of an impressive 512GB/s of memory bandwidth. And there’s its 4,096 shader units, compared to the previous 2,816 in Hawaii’s 290X/390X. So we expected impressive results. Considering the GPU clockspeeds are similar, the Fury X should be around 35–45 per cent faster than the R9 390X. Indeed, there’s a chance the Fury X could reign as the highest-performance desktop GPU around. AMD even released some preliminary results, showing it besting the 980 Ti and Titan X in a variety of games. It looked like AMD had the Next Big Thing™.
But there were concerns – the biggest being memory configuration. HBM uses a silicon interposer, like a simplistic microchip, to route traffic from the HBM modules to the GPU. This is required because each HBM module has a 1,024-bit interface, which would be very difficult to route using traditional methods. But the interposer must be large – large enough for the GPU core along with the HBM modules. And Fiji is a big chip to begin with.
The net result is that space limited AMD to four HBM modules. That in turn limited the Fury X to 4GB graphics memory – less than the R9 390/390X and the 980 Ti. If you run games at settings that exceed 4GB VRAM use, performance will suffer.
Despite AMD tests showing the Fury X leading across a collection of 12 games, we couldn’t corroborate those results. Out of our eight games, AMD only wins in Hitman: Absolution, and then only at 1440p and 4K, and in Batman: Arkham Origins, at 4K. One-and-a-half out of eight ain’t bad, right? It’s not what we were hoping for. The 980 Ti has a 3 per cent lead at 4K, increasing to 7 per cent at 1440p and 11 per cent at 1080p, plus the Fury’s performance on newer titles (GTA V and Witcher 3) is further off the pace than on older titles.
And there are other factors to consider. We can’t shake that 4GB VRAM limitation. When the R9 290/290X launched as AMD’s top GPUs 18 months ago, 4GB VRAM was considered a good choice. But we’re starting to see games utilise more than 4GB RAM. GTA V, for example, can use more than 4GB even at 1080p.
For better or worse, it also appears Nvidia is doing well at getting developers to use GameWorks libraries, some of which are Nvidia-specific features that won’t work without an Nvidia GPU. If one of your ‘must-have’ games works better on AMD hardware, that could easily sway your buying decision, but that works both ways, and recent major releases suggest Nvidia has had more ‘wins’ than AMD.
GTA, Arkham Knight and Witcher are perhaps the worst-case scenario for AMD. In short, they show what happens when the wheels come off the Fury X. GTA V uses too much memory at higher-quality settings; Arkham Knight apparently needs some patching and additional drivers; and The Witcher 3 with HairWorks taxes the tessellation hardware and can experience a big hit to performance.
Nvidia is winning the drivers wars of late. By our count, Nvidia had 10 WHQL driver releases in 2014, and has already added eight in 2015. By contrast, AMD had four in 2014, and one so far in 2015.
But we feel there’s room to improve Fury X performance with driver updates. It has 33 per cent more memory bandwidth than the already well-fed 390X, and shader computational performance should be up to 45 per cent faster than the 390X. So why then does the Fury X only average 18.5 per cent faster than 390X? Drivers.
The Fiji architecture is the first new high performance architecture for AMD since late 2013. (We don’t count Tonga, as it was effectively a lateral move from Tahiti.) It changes the playing field. Not only does it sport a third more shaders, but it also has a different memory subsystem. Yes, 512GB/s of bandwidth is great, but if latencies and other elements have changed – and they almost certainly have – then previous ‘best practice’ driver code may no longer be properly tuned.
It wouldn’t be the first time this has happened. AMD discovered it missed lots of potential Tahiti performance more than a year after its launch. Hopefully, AMD can improve Fiji performance much more quickly this time around. If so, it may still come out on top. Provided you don’t need more than 4GB VRAM.
There are also other factors. Nvidia’s Maxwell 2.0 cards (Titan X, 980 Ti, 980 and 970) have proven capable overclockers. The Fury X so far proved disappointing, netting only a 7 per cent core clock increase in our testing. This resulted in performance improvements of just 4 per cent on average. All of the Fury X cards will be running the reference configuration with a liquid cooler, with reference clocks as well, so unless something changes, we don’t see overclocking as a serious draw.
There is good news. Besides being competitive, if not chart-topping, the Fury X uses similar power to the 980 Ti. Where the 390X and GTX 980 offer similar performance, in our tests, the 390X ended up using 125W more power during gaming sessions than the 980 – that’s almost enough to power a second GTX 980! The Fury X, on the other hand, used 45W less power under load than the 390X, with power use falling just slightly above the 980 Ti (by 5W). Considering the 275W TDP, the similar real-world power use is good news.
Other elements could be good or bad. The built-in cooler, for example. It’s awesome to see a GPU this fast packed into a 7.5-inch card, but the closed-loop cooler ups the space requirements. It’s a potential problem for anyone wanting to create a 2-way or 3-way CrossFireX setup. That said, AMD does have a cheaper, air-cooled Fury on the way.
All up, the Fury X lands a few solid punches, but the 980 Ti simply has too much. The Fury X is a product with plenty of guts, but also bad habits. With additional drivers, it could become a force to be reckoned with. The question is whether that will be in a few weeks, months, or after so long that the 980 Ti and Titan X will have been replaced by even more formidable hardware.
- High bandwidth memory
- lots of shaders
- 4GB VRam limitation
- Poor overclocking chops
- Driver updates needed
AMD Radeon Fury X Specifications
- GPU Fiji
- SMs 64
- Lithography 28nm
- Shaders 4,096
- Texture units 256
- ROPs 64
- Core clock 1,050MHz
- Memory capacity 4GB
- Memory clock 1GHz
- Memory bandwidth 512GB/s
- TDP 275W