This RX Vega 56 versus GTX 1070 overclocking showdown is part of our RX Vega “Unleashed” series. Today, we have optimized our overclocks with all performance options set to their highest limits to get the most performance from each card.
Previously, we found that the stock-clocked $399 reference version of the RX Vega 56 is about 8-9% faster than the $399 stock-clocked reference GTX 1070 FE using 28 PC games. This time, we will test 20 PC games, and overclock the GTX 1070 FE and the RX Vega 56 each as far as they will go to see where they stand in relation to each other fully overclocked.
Undervolting & Overclocking the RX Vega 56
Overclocking the sample of the reference RX Vega 56 that we borrowed from Tech of Tomorrow benefited from undervolting unlike our RX Vega 64 Liquid Cooled Edition which did not. The goal with undervolting is to maximize Vega’s overall overclock by concentrating on overclocking the HBM 2 as far as possible using the maximum 1200mV that WattMan allows.
Since it is not possible to supply enough voltage to the core and to the memory simultaneously to get the highest possible overclock without putting either the power usage through the roof or accepting uneven clocks, we have to drop voltage to the core while still trying to keep the highest overclock we can manage. And of course, we supplied the maximum +50% Power Limit, turned the temperatures to their maximum limits, and set the fan to 100% to try and approximate what aftermarket RX Vega 56 cards may achieve with better cooling and power delivery.
First we decided on the maximum core overclock on stock voltage that made the most sense with regard to energy draw and performance bang-for-buck which for this RX Vega 56 sample is 2.5%. Then we found the maximum memory overclock that was attainable on this GPU (985MHz) with the maximum sustained memory voltage allowed partly by underclocking the core voltage. Here is the final stable overclock that we were able to maintain with the reference version of Vega 56 for 20 games.
Issues with the RX Vega 56 Overclock
First of all, we had to revert to AMD Crimson ReLive drivers 17.8.2 from 17.9.1 to get any overclock at all. Evidently 17.9.1 is a bug fix driver, but it does not allow overclocks to be set.
The easiest way to tell if an overclock is really successful or not is to reset WattMan’s reporting tool under Avg before running a benchmark, and afterward check to make sure the memory clock is a straight (purple) line. After running 20 games we began to notice that the memory clocks no longer held steady at 985MHz as previously. We continued to decrease the core overclock and the core voltage but these temporary measures to keep the voltage up to the memory eventually failed and the HBM2 clocks finally became erratic as below.
Of course with the memory clocks dropping and rising, the performance tanked. We tried reinstalling the drivers, and even moved the card to a different drive, but our reference RX Vega 56 would no longer hold any overclock on its HBM 2 and we called an end to our testing at 20 games instead of our usual 28. We did not flash to a Vega 64 Liquid BIOS to attempt to gain an additional 50mV but we used WattMan’s tool to supply the maximum 1200mV allowed by it. Fortunately, the card still runs OK at stock settings but it will not hold any memory overclock now.
For our 20 overclocked games, we settled on a 2.5% overclock to the core and the memory clocks were overclocked from the reference 800MHz to 985MHz with the fan allowed to ramp to 100%. Temperatures never reached even close to 80 C, but at 100%, the fan is quite annoying. We found that higher Vega 56 memory clocks gained significant performance in games, much more so than from overclocking the core. However, it appears that the reference RX Vega 56 edition is already running very close to its maximum core clocks.
Overclocking the Founders Edition of the GTX 1070
We devoted a separate evaluation to overclocking the GTX 1080 Founders Edition which you can read here, and the methodology is very similar to overclocking the GTX 1070 FE. We settled on a final stable overclock of adding 140 MHz to the core which settled in around 2012MHz with GPU Boost, and we added 520MHz to achieve a 4519MHz final stable memory clock.
We did not need to adjust the fan profile, but left it on automatic. The GTX 1070 memory overclock greatly contributed to the increased performance. The fan never became obtrusive as we were able to leave it at stock and the GPU remained relatively cool in the mid-70s C.
Our testing platform is Windows 10 Home 64-bit, using an Intel Core i7-6700K at 4.00GHz which turbos to 4.6GHz for all cores as set in the ASRock Z170 motherboard’s BIOS, and 16GB of HyperX DDR4 at 3333MHz. The settings and hardware are identical except for the two cards being tested.
We normally feature 28 games but were only able to test 20 games before the RX Vega 56 memory overclock became unstable. We also test VR with Futuremark’s Orange and Blue Room benchmarks and will compare the performance at 1920×1080, 2560×1440, and at 3840×2160 resolutions with maximum settings.
Before we run benchmarks, let’s check out the test configuration.