AMD Radeon RX Vega 64
Can AMD win two wars at once
PRICE $ 899 ($ 699 SOON APPARENTLY) www.amd.com
If there’s any doubt that AMD has guts, 2017 has certainly put that out to pasture. Striking a major win against Intel with its Ryzen and now Threadripper CPUs, that’s not enough for the team in red. Now they’re taking on the might of Nvidia, but rather than stick to the budget and mid-range sector, they’re aiming high, taking on the best that Nvidia has got. Considering AMD is a fraction of the size of both Intel and Nvidia, does it have what it takes to win on two fronts at once?
THE BIGGEST UPGRADE IN 6 YEARS
Ever since 2011, AMD has been using its Graphics Core Next architecture for its GPUs. In that time it’s made five revisions to the technology, leaving us with the current version of GCN 1.5, aka Vega. According to AMD, this is the first major new GPU design from AMD since the release of GCN. The company has been talking about the new RX Vega design since last year, with initial promises of GeForce GTX 1080 Ti-beating performance. It’s no wonder gamers have been waiting eagerly for the arrival of the final product, and now it’s here. Before we look at the performance AMD has managed to deliver, let’s take a deep dive on what makes the RX Vega design tick.
FLIPPING THE LID ON VEGA RX
AMD is actually launching three versions of the RX Vega. At the top of the range is the Radeon RX Vega 64 Liquid edition, which is currently retailing in Australia for $1049, putting it on par with the GeForce GTX 1080 Ti. The card we received is the middle of the stack, the Radeon RX Vega 64 air-cooled edition, which is hitting shelves for $899 at the moment. AMD promises this price will drop to $699 in the very near future; it’s just that the first shipment has sold like hotcakes. At this price, it’s facing off against the GeForce GTX 1080. Finally we have the slightly cut-back Radeon RX Vega 56, which isn’t currently on sale in Australia. Expect it to face off against Nvidia’s GeForce GTX 1060 though.
So, how does each of the three differ? Well, in terms of design, both the liquid-cooled and air-cooled RX are identical. They each ship with 4096 Stream Processors, organised into 64 Compute Units. They also share 256 Texture Units, while the number of ROPs is identical at 64. The RX Vega 56 gets a few cut-backs to lower its price. It drops the number of Stream processors to 3584, arranged into 56 Compute Units, while the Texture Units drop slightly to
224. Yet it still has the full 64 ROPs of the other two cards.
All three cards use 8GB of HBM 2.0 memory, but the RX Vega 56 lowers the speed to 1.6Gbps, down from 1.89Gbps on the other cards. One of the massive benefits of HBM 2.0 is the incredibly wide bus width that it offers, and all three cards use a huge 2048bit bus. Each card is made from 12.5 Billion 14nm FinFET transistors.
The major difference between both RX Vega 64 cards is the frequency and TDP. The Liquid Cooled version runs at a base clock of 1406MHz, which Boosts to 1677MHz, while the Air-cooled version lowers the base to 1247MHz and the Boost to 1546MHz. Meanwhile the RX Vega 56 is slower still, with a base of 1156MHz and a Boost of 1471MHz. Now, all those transistors need some serious energy, and in this regard the RX Vega is a disappointment. The liquid-cooled RX Vega 64 requires a whopping 345W to power the board, while the air-cooled version is only slightly less at 295W. Even the RX Vega 56 requires 210W. These are extremely high compared to the 180W or so that Nvidia requires for its high-end products, and will likely mean you’ll need to upgrade that 750W PSU of yours, ensuring you’re got plenty of headspace for your other goodies. Obviously twin 8-pin plugs are found on both of the RX Vega 64 cards, and above these are eight small LEDs that light up to show the load the card is under. We haven’t seen the 56 yet, so can’t be sure of how many power plugs it will require. AMD isn’t really talking up CrossFire, though the cards are CrossFire compatible. Unlike Nvidia, the Radeon RX Vega series doesn’t require a hardware bridge, instead communicating over the PCIe 3.0 lanes.
THE big CHangEs in gCn 1.5
So we know that AMD claims this GPU architecture has the most radical redesign in six years, but what are they? There’s a handful of key technologies that GCN 1.5 introduces, starting with the Next Generation Compute Units, or NCU for short. Without getting too tech heavy, this allows for a pair of FP16 instructions per FP32 ALU, doubling the number of instructions if they’re identical. It offers 512 Operations on 8-bits per clock, 256 for 16-bits and 128 for 32-bits. Each NCU is organised into a batch of 64 NCUs, which it calls a Compute Unit. They’ve also doubled the L2 cache available in total to the GPU,
Next up is the part designed to handle all that memory – the High Bandwidth Cache Controller, or HBCC. This allows the GPU to access external system memory at a much faster rate than before, with support for data sets of up to 512TB in size. Remember that the card only has 8GB of memory, so being able to access that system
will likely mean you’ll need to upgrade that 750W power supply unit of yours
memory faster is a good thing, especially as today’s games require so much more memory. For example, AMD showed that The Witcher 3 allocates three times the data that is being used on the GPU. The HBM 2.0 is also a major improvement, offering up to eight times the performance of HBM 1.0, and capacities of up to 32GB (HBM 1.0 was limited to just 4GB).
RX Vega also introduces a new programmable geometry pipeline. Recently, game devs have been relying on the shader pipeline to handle geometry shaders, but it’s not the most efficient way to do the job. Consider how rendering a scene works; the game engine has to render every single polygon in that scene. However, sometimes only 1% of those are actually viewable to the player, as the other polygons may be hidden behind other polygons. This is why RX Vega now has a programmable geometry pipeline, which AMD claims offers up to twice the throughput per clock of previous designs.
AMD is well known for its focus on DX12, and claims the RX Vega architecture is the most complete solution for DX12 on the market. Unfortunately we didn’t have time to test these claims, but look for a DX12 vs Vulkan vs DX11 article in the near future. As for outputs, AMD has gone with its usual design – triple DisplayPort 1.4 ports and a single HDMI 2.0b. There’s no sign of a DVI-I output this time around though, but we expect vendors will likely add this. FreeSynch 2.0 is also included, which focuses on delivering High Dynamic Range (HDR) colour at a lower latency. As for the cooler included on the AMD edition card we were sent, to be frank we were a little disappointed. Considering it’s got so much heat to move around, a single fan blower design is a bit underwhelming. No wonder it hit 56dB during our game tests, but again we can look forward to third-party manufacturers releasing much more exotic cooling solutions with multiple fans. It also supports AMD’s new Radeon Software Crimson ReLive software for overclocking, but early results have shown that despite an extra 100MHz or two can be gained, it has very little impact on games. We can’t wait to see if the RX Vega 56 changes that.
PERFORMANCE MATTERS
On to the bit you care most about – performance. We tested in our stock testbench, an Asus Maximus VIII Hero with Intel Core i7-7700K and 2 x 8GB DDR4-2666MHz of Ballistic memory. However, we used an AMD supplied PSU rated to handle 1250W, but stuck with our usual array of three SSDs for the OS and benchmarks. We ensured that all competing cards were downclocked to Nvidia’s default speeds.
As you can see from our new benchmark suite, unfortunately the RX Vega 64 lost out to the GeForce GTX 1080, but not by a huge margin. It did however wipe the floor with all cards in the Rise of the Tomb Raider test, which is known to like AMD products. When compared against the 1080 Ti there’s simply no competition, but they’re in different price ranges. As mentioned, the card is extremely loud under load, so you’ll definitely hear it when installed in your system.
NO VICTORY TO RX VEGA
It’s great to see AMD come so close to the GTX 1080, yet a little sad that it can’t beat a year-old product. There’s also the issue of power consumption and noise, which can’t be ignored in this day and age of power efficiency and silent systems. Finally, at the time of print, the pricing was simply insane. AMD has promised this will be resolved with the next shipment, dropping to the RRP of $699, so we’ll just have to take their word for it.
It’ll be interesting to see what the third parties can do with this card. Whack a better cooler on and it should allow for better overclocks, which could see it beat the GeForce GTX 1080. Until then, it’s a case of close, but no cigar.
the card is extremely loud under load, so you’ll definitely hear it when installed