The AMD Radeon RX Vega Review: Vega 64, Vega 64 Liquid, Vega 56 Tested
Clocks, Power Consumption, Overclocking
Before we dive head-first into the world of benchmarks, it is important to look at how the technology and cooling capability that AMD built around the RX Vega graphics cards turned out. Let’s start with a look at how the measured clock speed of the three new cards, the Vega 64, Vega 64 Liquid, and Vega 56 compare.
The Vega 64 Liquid has the highest sustained clock speeds and averages 1663 MHz over the course of our Unigine Heaven loop. The standard Vega 64 averages about 150 MHz lower at 1513 MHz and the Vega 56 brings up the back with an average clock speed of 1433 MHz. If we compare that to the rated clocks of the card from AMD, which we now know are supposed to represent the “typical” clocks users will see while gaming, all three are lower than expected. The Vega 64 Liquid is rated at 1677 MHz, with our results just behind that by 14 MHz. That difference is within a reasonable margin.
The Vega 64 standard card was rated at 1546 MHz, and our 1513 MHz result is 33 MHz lower. The Vega 56 is rated at 1471 MHz but in our testing, we saw it 38 MHz lower. AMD is closer than they have been and I still think the move to reporting an “typical” clock speed is a welcome change. Hopefully over time they can tweak this process to rate more accurately.
A delta of 150 MHz between the RX Vega 64 cards in air- and liquid-cooled scenarios is a substantial difference and one that will without a doubt result in performance increases. Though the price is $200 higher, the liquid cooler can keep the power hungry and hot Vega 64 GPU under control and maintain clock speeds that are nearly 10% faster on average.
Detailed Power Consumption Testing
One of our architectural concerns when testing the Radeon Vega Frontier Edition was power consumption. While we don’t expect things to change much for the RX Vega 64, I was very curious to see how the new Vega 56 would stack up.
If you don’t know how we measure power, check in on one of our previous reviews. Our methods measure power directly to the graphics cards through the PCIe bus and the auxiliary ATX power connections. This is not a basic “at the wall” measurement.
In both The Witcher 3 and Rise of the Tomb Raider, there is a distinct difference between the RX Vega GPU and the GeForce Pascal-based products. The RX Vega 64 Liquid uses 350 watts in its default configuration while the standard RX Vega 64 consumes around 290 watts. The Vega 56 more or less hits its expected 210-watt TDP.
On the NVIDIA side of things, the primary competition for the RX Vega 64 is the GTX 1080 that uses 170-180 watts under a full gaming load. The GRX 1070, which is the target of the new RX Vega 56, consumes around 140 watts in both games here. The GTX 1080 Ti sits right at its 250-watt power draw rating, but will outperform the RX Vega 64 and GTX 1080 by a wide margin.
Power consumption differences show the efficiency of the performance of each of these GPUs. Assuming the standard RX Vega 64 matches performance with the GTX 1080, the Pascal GPU can offer similar performance but at 37% less power. While enthusiasts and gamers are well known for their ability to overlook power draw, running at lower power means you can run at lower temperatures, lower noise levels, and possibly fit the product in smaller spaces. The RX Vega 56 uses 33% more power than the GTX 1070, so the same analogy would apply if performance is equal. We’ll dive more into that on our performance analysis pages.
Overclocking RX Vega
Overclocking these three cards produced very different experiences. Below is the graph of my peak overclock and resulting clock speeds for each.
Starting at the flagship level with the RX Vega 64 Liquid, I was only able to squeeze another 15 MHz out of the card. In fact, even increasing the clock speed by 1% in the Wattman slider would result in a crash or a black screen, even with the temperature maxed out at 70C and the power target slider moved to +50%. Therefore, the differences you see for the RX Vega 64 Liquid in the graphs is the result of adjusting the power slider only – pretty disappointing. Clearly AMD has pushed the Vega 10 GPU to its limits already to get the Vega 64 Liquid to its current performance levels.
The standard RX Vega 64 card was able to run at 3% higher clock speeds in Wattman, resulting in an average clock speed that is 86 MHz higher than stock settings. This includes a +50% power target adjustment.
Overclocking on RX Vega 56 - Note that GPU-Z still reports the clock speed as the MAXIMUM, not the "typical." This will be updated in the future I'm told.
The RX Vega 56 saw the biggest jump, accepting a 7.5% clock speed increase with the same +50% power target shift, resulting in a 94 MHz increase in average clock speed. Though the card remained stable even moving the clock slider to +10%, clock speeds never seemed to go above what we show here.
Another interesting trait that I found, that may be a result of the current GPU-Z implementation for Vega, is that the clock speed steppings when overclocked were as small as 2-3 MHz. In stock settings though we found those steps to be much larger, on the order of 100 MHz in some case! I don’t know what in the design would be causing this kind of change, but it would be interesting if overclocking the card slightly allowed for less variance in frame times by lowering the clock speed steppings between states.
The power results of this overclocking deserve discussion as well. The RX Vega 64 Liquid card was pushing 440 watts with the power slider adjusted over, even though the performance increases we saw were near zero. The RX Vega 56 (grey line) moves the 210-watt card up to 310 watts, near the 50% increase we would expect with the power slider change on a product with headroom.
The air-cooled RX Vega 64 had similar problems as the Vega Frontier Edition in regard to temperature throttling. Leaving the fan settings untouched, we saw frequent and regular throttles from 340 watts of power draw down to 200 watts, which caused clock speed dips and noticeable stutter. By manually moving the fan speed up to 3400 RPM, which is very loud, the card could run unthrottled at those settings, providing more consistent performance and lower power consumption to boot (~10 watts).
Though a single sample of each does not provide enough information for me to make any strong statements on the overclockability of these RX Vega family, my assertion is that the RX Vega 56 will have a high amount of headroom due to its out-of-box power set to 210 watts, even though it is essentially the same GPU and thermal solution as the RX Vega 64. Vega 64 on the air-cooled variants has headroom to reach higher clocks nearing the liquid-cooled option if you are willing to put up with the noise levels of a very loud blower fan. (Partner cards with better coolers might work around this.) As for the RX Vega 64 Liquid, it seems that AMD is already getting the Vega 10 GPU to its near peak capabilities, so I don’t expect that much headroom for it even with the fancy cooler design. That 150 MHz delta between the air and liquid versions seems to be near the pinnacle of Vega 10 in its current state.