AMD RADEON RX VEGA 56 & 64
It’s been a long time coming, so is AMD’s new Vega graphics tech ahead of its time, or just late to the party? Jarred Walton finds out.
Let’s be frank: the Vega 56 and 64 are what we wanted to see last year, right around the time AMD was busy launching Polaris for the mainstream market. But Vega wasn’t even close to being ready. So why did it take so long for AMD to get a high-end replacement for Fiji and the R9 Fury X out the door? AMD has talked about numerous architectural updates that have been made with Vega, and in the professional world, some of these could be massively useful, but as we’re APC, we’re at least equally interested in how these new cards stack up as a gaming device. So does it?
RX VEGA 64
Let’s quickly talk specs. Vega includes 8GB of HBM2 memory in two stacks, which deliver the same 512GB/s bandwidth as the four stacks of HBM1 in last-gen’s Fiji, but with just two stacks, the silicon interposer doesn’t need to be as large, and HBM2 densities have allowed AMD to double the amount of memory. But AMD isn’t just calling this HBM or VRAM; it’s now a High-Bandwidth Cache (HBC) and there’s also a new High-Bandwidth Cache Controller (HBCC). The difference is important, because the HBCC plays a much more prominent role in memory accesses. AMD calls this a “completely new memory hierarchy”. That’s probably a bit of hyperbole, but the idea is to better enable the GPU to work with large data sets, which is becoming an increasingly difficult problem.
Vega also has a new geometry pipeline, with over twice the throughput per clock as AMD’s previous architecture. The compute unit has also been improved, with native support for packed FP16 operations, which should prove very useful for machine learning applications. AMD has improved the pixel engine, too, with a new Draw Stream Binning Rasterizer that helps cull pixels that aren’t visible in the final scene.
We didn’t get the liquidcooled Vega 64 for testing, but it should be up to 8% faster than the air-cooled version, based on boost clocks. If you look at the raw numbers, Vega holds promise — with higher theoretical computational performance and more memory bandwidth than the GeForce GTX 1080 Ti — but AMD’s GCN architecture has traditionally trailed behind Nvidia GPUs with similar TFLOPS. DirectX 12 games, meanwhile, tend to track a bit closer to the theoretical performance, and certain tasks (like cryptocurrency mining) can do very well on AMD architectures. But the power requirements of AMD GPUs have traditionally been higher than the Nvidia equivalents, and that remains true with Vega.
TAKING ON NVIDIA’S BEST?
Many in the gaming world were hoping Vega would be able to take down the heavyweight champion, the GTX 1080 Ti. In testing, however, that didn’t happen, as the 1080 Ti remains out of reach, though to be fair, it also remains in a higher priced bracket. Based on cost alone, the direct competition for the Vega 64 is Nvidia’s GTX 1080, while the Vega 56 is looking to take on the GTX 1070.
Vega 64 runs games fine — more than fine, as it’s now the third-fastest consumer graphics card, and we’d
expect nothing less for $800. At best, Vega 64 is just a hair faster than the GTX 1080, but at worst, it can be about 30% slower — on average, the GTX 1080 leads by just under 10%.
The Vega 64 really feels like a one-generation upgrade, despite the fact that AMD skipped updates to the high-end market last year. We often recommend skipping a generation or two of hardware between upgrades, which means upgrading every 2–3 years on graphics cards. Each generation will typically improve performance by 30% or so, and that’s not usually enough to warrant an upgrade for most gamers. If you skip a generation, you potentially end up with a 60–70% performance jump. If you’re already running a higher-end AMD card from last generation (like an RX 570 or RX 580), then our recommendation to wait another round still stands.
What about power use? While it’s a wash at idle, while gaming, the Vega 64 continues AMD’s trend, using 478W at the outlet on the test system compared to 370W for the GTX 1080.
RX VEGA 56
Vega 64 was clearly designed to go after Nvidia’s top parts. It came up short of the 1080 Ti, but is relatively close to the 1080. The Vega 56 doesn’t aim quite as high or cost as much, so is it a better bet? While the power draw is still a bit high on this slightly lower-end card (as we measured 400W while playing games, compared to 314W for the GTX 1070), looking at just the Vega chips, the Vega 64 is only about 9% faster than the 56 overall, but it uses around 80W (20%) more
power. In terms of system performance per watt, the Vega 56 is, therefore, 10% better than the Vega 64, and if we look just at the graphics card, it’s around 25% better.
AMD has included a BIOS switch along with a powersaving profile — but that comes at the cost of performance, of course. And if you want to overclock it, you still can, though that will affect the overall efficiency in the reverse direction.
OVERCLOCKING PERFORMANCE
We performed some preliminary overclocking, and things are far more interesting on the Vega 56 card than the 64, as you’d expect. Simply raising the HBM2 clockspeed to a 930MHz base (close to the same as the Vega 64’s 945MHz) and cranking the power limit to 50% improved performance to where Vega 56 is effectively tied with the 64. Going further, we toyed with the percentage overclock slider and managed to push it up 20% higher. In both cases, system power draw while gaming basically matched the Vega 64 — 480W at the outlet. But in limited testing, all that added clockspeed actually did very little for performance.
The Vega 56 isn’t just more efficient than the Vega 64, though — when stock levels stabilise (at the time of writing, it was completely sold out worldwide), it should end up being a little cheaper than the GTX 1070, while delivering nearly the same performance overall.
The 56 ends up being a good alternative to the GTX 1070, and an even better one if it’s actually readily available at $550–$600, considering the 1070 currently goes for $600– $650 or more due to mining demand.
STOCK PROBLEMS
Given the shortages on AMD’s RX 570/580 cards, which are typically selling at prices 50% or more above the MSRP, many have feared — and miners have hoped — that the RX Vega would be another excellent mining option. In Ethereum mining, so far, it’s been pretty lacklustre, however, considering the price and power use. The Vega 56 manages around 31MH/s and the 64 does 33MH/s. Overclocking the VRAM helps boost both cards closer to 40MH/s, so at launch, that’s not super promising.
Part of the problem is that most of the mining software (whether for Ethereum, Zcash or some other coin) has been finely tuned to run on AMD’s existing Polaris architecture. Given time, we could see substantially higher hashrates out of Vega, and there are rumours that the right combination of VBIOS, drivers and mining software can hit mining speeds more than double what we measured. All we can do is hold our breath and hope that mining doesn’t drive prices of Vega into the stratosphere in the future.
Considering the significantly higher clockspeeds, we had hoped Vega would perform far better than what we’re seeing today. It’s not a bad GPU by any means, but it’s not going to dethrone Nvidia, at least not right now. What Vega does have going for it are reasonable prices, and AMD users will certainly appreciate having something that’s clearly faster than both the RX 580 and the R9 Fury X. The real target would be gamers who are still running R9 290/290X (or older) hardware.
“The Vega 56 ends up being a good alternative to the GTX 1070, and an even better one if it’s actually readily available at $550–$600.”