Welcome Stranger to OCC!Login | Register

A Look at RX Vega 64 Efficiency

   -   
» Discuss this article (5)

Middle-earth: Shadow of War Results:

Pulling back the curtain a little bit, I did the testing in The New Colossus at the end of March, reviewed it to install Shadow of War in April, and then this specific testing in May. While it was from processing the data I had collected that revealed the issues and mistakes I made before, I did not really dive into the complete data (as in all OCAT and Adrenalin runs for a set) until I had already uninstalled The New Colossus. My point in saying this is that while it was too late to correct the issues with The New Colossus data, I was able to prevent these same mistakes with Shadow of War, so there are no missed runs or impacts from other programs, at least based on OCAT capturing them. I did however do these sets with my CPU at 3.7 GHz as I had a couple hard-locks while reviewing the game (which I did before these runs) at 4 GHz, and when other applications have had this issue, dropping the clock speed helped.

The run I did in Shadow of War was basically just to run along the top of the Nurnen region, with my plan being to cross the map around this perimeter, going from top-right to bottom-left. However, I was not going to just run this path, I was going to engage what enemies I found, just like when playing the game normally. The runs typically ended in the same location, on the North-West corner of this path, before turning South, so actually a bit before where the video ends. After making a run I would fast travel back to the start, change the FRTC setting via the Radeon Overlay, and start recording.

 

 

 

 

 

 

 

Something you will notice immediately when looking at the graphs for Shadow of War is that I did far fewer runs with it. This is because my RX Vega 64 was topping out at about 100 FPS, at the highest settings, for this game. In fact the averages of the Default runs do not even break 90 FPS, but I still decided to start at 100 FPS with FRTC so the impact would not be significant at first.

 

 

Now we are seeing undervolting having quite an impact on the power used by the GPU. While the Stock Default run averaged 218 W, the UV Default run is 190 W and UV +50 is 189 W. That is a drop of around 30 W, and these are not the only runs with such a drop. The 70 FPS and 80 FPS runs also show a 30 W drop between the Stock and either UV run, while 90 FPS had a drop of more around 25 W and at 100 FPS the difference is 20 W. At 60 FPS, these runs have around the same average power with Stock pulling 142 W, UV pulling 144, and UV +50 pulling 142 W.

Looking at the box plots for the frame rates, we see the 60 FPS, 70 FPS, and 80 FPS targets are all pretty tight, which indicates my RX Vega 64 was able to consistently achieve this level of performance in the game at all three Wattman configurations. Considering the Default runs of both the UV and UV +50 sets have approximately the same average power usage as the Stock 70 run, the value of the undervolt is very evident here.

 

 

While the power might be noticeably lessened with the undervolt, the fan speed does not look to have dropped as much with many of the runs across all three sets hitting against the 2400 RPM limit. That being said, the UV and UV +50 runs do show lower average fan speed than the stock runs. Of course all of the runs are still above 2000 RPM, which would speak to this being a challenging game to render, but still seeing a lower fan speed does indicate less heat was being generated.

 

 

Looking at these graphs of the average clock speed we can see again the undervolt having a significant impact on performance. While the Stock averages never broke 1500 MHz, the UV and UV +50 runs for 80 FPS and above all did. Looking back to the average fan speed graph and we can safely conclude there was thermal throttling in play for the Stock runs, forcing the clock speed down. Even at the 60 FPS and 70 FPS runs, Stock had a lower clock speed, though I think the throttling was less for these runs, based on the lower average fan speed seen in the previous graph. It is still clear that undervolting is worthwhile in Shadow of War.

 

 

Comparing the average clock speed and power use we can clearly see how the undervolt reduced power and allowed the clock speed to increase significantly compared to the Stock set. I am not sure what more can be said about these, so moving on.

 

 

Now that we are not looking at broad averages we can see the Stock Default run was reaching up to 225 W of power, a limit I have observed for my GPU throughout these tests and other times I have recorded power use through Adrenalin. Looking it up, this corresponds with the 220 W limit of the primary BIOS for the RX Vega 64 using the Balanced profile. (The RX Vega 64 and Vega 56 both come with two BIOSes with the second configured to use less power.) While The New Colossus was not pushing the card to that level for any of the runs, this game does for Stock. Remembering the fan hitting its limit for some Stock runs, it looks like the GPU was both thermal and power limited for those runs. We can see the regression line turns down at the end of the Stock set, which is just a curious artifact of the data and not reality. The power usage will not start decreasing as clock speed increases without changing other variables. The 225 W limit is no doubt influencing this as well. Also I suspect the lack of additional data points at the lower end of clock speed and power I think might be skewing the data some.

Looking to the UV set and we see some interesting behavior with the regression line. Again we see it start to go down again at the end, even though we are not at the 225 W limit, but the reason is certainly the same as with stock; there are more points coming in below the line than above, making it turn down. Realistically, it should be continuing to trend up.

What is more interesting in this set, and it is also present in the UV +50 set, in an undulation in the regression line. Realistically, a line showing the relationship between clock speed and power would not have such a profound curve to it, and here I think it is just a result of how little data there is at those clock speeds. You can clearly see it is the 60 FPS run mostly influencing the line where the undulation is and I suspect if I even just repeated the 60 FPS run a couple times or did 50 FPS and 40 FPS runs, it would flatten out here. I wanted to stop at 60 FPS as that is the refresh rate of most monitors.

While the undulation does make spotting the inflection points easier here, we also know the regression is not as informative as we may wish it was. Still, we can see the inflection point for the Stock set is around 1200 MHz, which is pretty low, in my opinion. For the UV and UV +50 sets the inflection point looks to be around 1325 MHz, which is much better. Looking back a bit, this is also around where it was for The New Colossus undervolted sets. With more runs to flatten out the regression line, the inflection point might shift, but my guess is it will stay around there, partly because this is what we see with The New Colossus.

 

 

 

 

To show how well the GPU managed to hit the FRTC targets, here are the graphs of the frame rates and display rates, based on the OCAT data. Looking at the frame rate graphs we see the 60 FPS and 70 FPS targets were being hit quite well, and a nice spike for the 80 FPS run as well, though not as nice. For Stock the 90 FPS run barely has a spike, but it is clearly defined, if not very tall for the UV and UV +50 sets.

Turning to the display rate graph we can get a look at the experience I had while playing, with Stock apparently never managing to constantly send out 90 frames in a second to my display. Both the UV and UV +50 sets have a spike, though it is quite short, telling us the GPU was struggling to maintain that target.

Looking at the 80 FPS runs, we see the Stock run was having a harder time sending frames to the display at that rate. If each of the 80 frames a second were being sent to the display at that rate, the spike would be taller than the 70 FPS spike purely because there are more frames to count, but for Stock we see the spike is lower. This tells us it was struggling to hit that target, but the UV and UV +50 sets have their 80 FPS spikes stretching above the 70 FPS spike.

 

 

 

 

Looking at the course graphs for the frame time and display time, we again see Stock was struggling with 80 FPS more than the other two sets. We can also see in the Display Time graphs that Stock never managed to maintain the 90 FPS target either, while the other sets do show flat sections.

Time to move on to the final game I did testing in and, well, things are about to get weird.




  1. RX Vega 64 Efficiency - Introduction
  2. RX Vega 64 Efficiency - Procedure
  3. RX Vega 64 Efficiency - Wolfenstein 2: The New Colossus Results
  4. RX Vega 64 Efficiency - Middle-erath: Shadow of War Results
  5. RX Vega 64 Efficiency - Killing Floor 2 Results
  6. RX Vega 64 Efficiency - All Graphs and P-State Tables Part 1
  7. RX Vega 64 Efficiency - Conclusion
Related Products
Random Pic
© 2001-2018 Overclockers Club ® Privacy Policy
Elapsed: 0.4477829933   (xlweb1)