Welcome Stranger to OCC!Login | Register

A Look at RX Vega 64 Efficiency

   -   
Category: Video Cards
» Discuss this article (6)

Introduction:

Back when I was working on the Wolfenstein 2: The New Colossus Performance Analysis I decided to include some tests using the Radeon Chill feature that is part of AMD's drivers. This feature allows the user to set an upper and lower frame rate limit the drivers will try to keep a game running between. The upper limit acts as a normal frame rate limit while the lower limit is a floor the frame rate can be reduced to when there is less action in the scene. Not every moment in a game needs to be rendered at a high frame rate for a good and smooth experience, so with Chill dropping the frame rate at times, the power usage and temperature of the GPU can also be reduced, potentially by a significant amount. In theory the limiting feature of Chill may improve responsiveness, based on the delay or latency between a frame being rendered with the latest information and when it is shown.

When I did those tests inThe New Colossus and looked at the data, it showed my RX Vega 64 was able to render the game, using its highest settings, at between 60 FPS and 90 FPS while only using about 75 W. I found this very interesting as 75 W is the amount of power the PCIe slot is specified to provide. In theory, this modern title with the full glory of its graphics could be enjoyed by the RX Vega 64 with only a single 6-pin power connector supporting the slot, instead of the two 8-pin connectors my reference model has. That is if more power is not just pulled through the slot because while the specification is only for 75 W, many motherboards tend to be overbuilt, making it possible to safely supply more than the specification. (This is something I learned when following the news about the RX 480 graphics cards and how much power they pulled.)

I thought this was an interesting discovery, so I wanted to investigate further by measuring the performance and power consumption of my RX Vega 64 at different performance targets. So often I see the Vega 10 GPU described as 'power hungry' or 'inefficient,' but I would not consider a GPU able to provide 60-90 FPS in a modern game at max settings for around 75 W inefficient.

The Radeon Adrenalin drivers have the ability to record various performance measurements for AMD GPUs, including ASIC power use, fan speed, core clock speed, HBM clock speed, and more, and it is this feature I used to record that information, though I used OCAT to record frame time data in parallel. Obviously this same feature cannot be used on the GTX 1080 I have, which would mean a third-party utility would be necessary, and while they do exist, I have an aversion to using them. These additional utilities will add more hooks into the games, which risks stability and may interfere with performance. I have seen games crash from trying to use OBS Studio and/or OCAT. Additionally, unless the power usage of the GTX 1080's GDDR5X is also recorded, the power usage numbers will be incomparable as the ASIC power reported by Adrenalin includes both the GPU and HBM2 power usage. I like the idea of limiting this examination to what is possible out-of-box with the RX Vega 64, as the features I will be using to influence the power usage of the graphics card are part of the drivers, so all users can do the same, if they wish.

Naturally for this project I wanted to continue working with The New Colossus but other games should also be investigated. I decided to use Middle-earth: Shadow of War and Killing Floor 2. Due to the large install size of The New Colossus and Shadow of War I was only able to have one installed at a time, which was especially unfortunate as I failed to notice I had missed a recording until after Shadow of War was already installing.

Why Shadow of War and Killing Floor 2? I wanted to do a total of three games and I wanted one of the games to be modern, which Shadow of War satisfies, having released in October 2017, and to be free of any potential bias for the RX Vega 64, which both should satisfy. Initially I wanted to use Serious Sam Fusion 2017 for the third game, as I have worked with it in the past for other performance statistics focused articles (Serious Statistics and Serious Statistics 2: The Sync Encounter), but it was not behaving as I needed it to. (I suspect this might partially relate to its support for multiple graphics APIs confusing things.) Thankfully Killing Floor 2 behaved as I wanted, and its age also allows it to act as a best-case-scenario for being a light load, at least compared to the two modern titles.

The next page of this article is going to go through my process for making the recordings I collected. After that I will get into how my RX Vega 64 performed at stock, when undervolted, and when undervolted but with the Power Limit increased by 50%.




Related Products
Random Pic
© 2001-2018 Overclockers Club ® Privacy Policy
Elapsed: 0.0364968777   (xlweb1)