Overclocking the Radeon 5870
Last week we used ATIís Overdrive utility to take both of our Radeon 5870 boards to speeds of 900MHz core/1300MHz memory
: these are the maximum
speeds provided by Overdrive in ATIís Catalyst 9.10 beta driver.
Seeing both 5870 cards scale so well was an encouraging sign Ė this was a feat RV770 wasnít able to pull off on its launch day Ė but we were also disappointed in the sense that we couldnít see how far these cards could ultimately be pushed in terms of clock speeds. We got a brief snapshot into how far RV870 scaled, but we really wanted the complete picture.
ATI told us they were toying with the idea of perhaps offering unlimited OCíing in Overdrive for the 5870 at some point in the future, but that a decision hadnít been made and that we may not get such a feature.
Fortunately AMDís Overdrive team does provide such a utility; their GPU Clock Tool has been providing this capability for some time, although it doesnít offer the officially sanctioned supported of Overdrive in CCC.
This is a use at your own risk, donít blame us if something doesnít work type of program thatís intended for enthusiasts and experienced overclockers.
The GPU Clock Tool also comes with multiple downsides: it doesnít save your OCíed speeds, so youíll have to manually dial in your clock speeds every time you boot up your PC, support for voltage adjustment is listed but it doesnít work with the 5870, and the same goes for fan speeds. Also, employing GPU Clock Tool makes Overdrive built-in to CCC go haywire; you wonít be able to monitor temps or adjust fan speeds in Overdrive. In fact Overdrive doesnít work at all with GPU Clock Tool.
In GPU Clock Toolís defense though, it is the only way to get no-limits OCíing with the 5870 at this time. You can also watch RV870 temps and voltages with GPU Tool.
GPU Clock Tool for Radeon 5870 was just released last Thursday and can be found at techPowerUp
So how far were we able to push our cards with GPU Clock Tool? The first 5870 board hit 930MHz core/1350MHz memory. We could actually run many apps with the graphics core running at 950MHz, but needed to drop down to 930MHz to achieve complete stability. We tried everything we could to get 950MHz stable, including pulling off the stock cooler and swapping out the thermal paste for Arctic Silver 5, but it didnít improve stability. We did manage to get some nice shots of the 5870 card sans cooling though:
Still an OC of 80MHz (9%) on the graphics core is pretty good with such an early engineering sample Ė our board was manufactured at the beginning of this month. With voltage and/or fan speed adjustment, we wouldnít be surprised if 950MHz ran stable with this particular board.
The second card didnít scale nearly as far. We hit speeds of just 912MHz core/1005MHz memory before stability became compromised. This board didnít run any hotter than our first card, so temps definitely werenít the issue. It just didnít want to go any further.
To track the impact GPU and memory OCíing has on performance, we decided to OC each of these components by a fixed amount Ė 9%. That equates to a GPU clock speed of 930MHz, and memory speed of 1318MHz. We were curious to see which OCíed component delivered the most tangible performance benefit. As youíll see in the benchmarks, this will also help put to rest some of the thinking out there than RV870 is constrained by its memory bandwidth. Letís get to the results!