Table of Contents
Navigation
Page 2:
Page 3:
Page 4:
Introduction
AMD's new Radeon R9 290X is a new video card that is being heavy discussed in the enthusiast PC community as it gives performance that is close to or better than the NVIDIA GeForce GTX Titan, but at roughly half the cost. One of the features of the R9 290X is the inclusion of a hardware switch that changes the card between two fan profiles called "Quiet" and "Uber". The differences between these profiles is only evident when the GPU is under load and works by changing how the GPU keeps itself cool when it starts running hot.
In our testing, we found that these two profiles change the fan speed and core clock speed by the following rules:
| Quiet Profile | Uber Profile |
|
|
Since the Quiet profile only ramps the fan up 2200 RPM (which on our cards is about 44%) before reducing the core clock, there is obviously going to be a performance hit in many situations if you use the Quiet profile. This has been shown in almost every R9 290X review currently published, but the most of the benchmarks show very little performance difference between the Quiet and Uber profiles. However, when we started our own testing, we saw a much larger difference. To understand what might be causing our results to vary so much from other published reviews, we decided to expand beyond our normal testing.
Test Setup
To examine the performance of the AMD Radeon R9 290X, we performed successive benchmark runs with the following hardware:
|
|
||||||||||||||||||||
We suspect that the chassis cooling is impacting our results, so in addition to comparing the Quiet and Uber fan profiles we will also be switching the chassis fans between 5v and 12v to simulate both a quieter, low airflow system and a louder, high airflow system. To see how these cards perform when cooling is not an issue, we will be also be liquid cooling the cards in order to test the R9 290X in a best-case cooling scenario. Although it shouldn't matter with the cards being liquid cooled, we will be leaving the cards on Uber mode to make sure they do not throttle the clock speed.
Since the R9 290X is becoming increasingly popular and debated online, we wanted to get initial results out as quickly as possible. In the future, we may benchmark more titles but for right now we will only be using two benchmarks. In addition, we will only be using a single R9 290X while benchmarking at 1080p, and a pair of R9 290X cards in Crossfire for testing at 4k resolutions.
This does mean that our results won't be 100% accurate for every game currently available, but it will still be a very good indication of how the performance of the Radeon R9 290X is affected by the overall cooling of the system in both Uber and Quiet mode. Since we are running multiple, successive benchmark runs, we will also be able to see the performance difference from when a benchmark first runs and when the system has reached full load temperatures.
The two benchmarks we will be using are Unigine Heaven Pro 4.0 and Metro: Last Light. Both are very modern benchmarks that are easy to setup for successive benchmark runs with little to no time in between each run.
Performance – R9 290X @ 1080p
| Max GPU Temperature | Max GPU Fan Speed | Avg. GPU Frequency | Max GPU Frequency |
|
|
|
|
|
Our very first benchmark – Unigine Heaven 4.0 running at 1080p – already shows that in some instances the performance of the Radeon R9 290X decreases after the initial benchmark run. The liquid cooled setup and Uber profiles did not show any variation, but both of the Quiet profile setups show a very noticeable performance drop after the very first benchmark run.
Although the Quiet profile does perform slower than the Uber profile, if you look at the fan speed logs you can see the benefit of using the Quiet profile. With the chassis fans at 5v, the Quiet profile keeps the fan running about 20% slower than the Uber profile, which is a very large and audibly noticeable difference. One other thing to point out is that with the exception of the liquid cooled setup, each configuration runs pretty much constantly at 94 °C. So if you are planning on purchasing a R9 290X, don't have a heart attack when you find that you can almost boil water on your video card while gaming.
| Max GPU Temperature | Max GPU Fan Speed | Avg. GPU Frequency | Max GPU Frequency |
|
|
|
|
|
Metro: Last Light gives us very similar results to Unigine Heaven, although we actually see a very slight performance drop after the first run even with the Uber profile and Liquid cooled setups. It's not much (between .43FPS and .69FPS), but the fact that it happens on every single configuration indicates that there is a very small performance drop even in the best cooling scenarios.
The Quiet profile again shows a performance loss over time, but the difference is much larger than what we saw with Unigine Heaven.
Analysis – R9 290X @ 1080p
As expected from the graphs in the previous section, the Uber profile and liquid cooled configurations don't show much of a performance drop in either Unigine Heaven or Metro: Last Light at 1080p. This means that for a single R9 290X using the Uber profile, online benchmarks should pretty much always be accurate – at least at 1080p.
The Quiet profiles, on the other hand, showed a performance drop of about 5.5% with the chassis fans at 5v. In other words, when using the Quiet profile there is the possibility that online benchmarks may be off by as much as 5.5% at 1080p if the reviewer did not perform multiple benchmark runs, discarding the first four or so benchmark runs.
This brings up the question of how much of the maximum performance possible you should you actually expect when using the Quiet or Uber profiles in various cooling setups. The good news is that with the Uber profile, you are likely to see pretty much full performance regardless of your cooling setup (within reason of course).
However, if you use the Quiet profile even in a very high airflow chassis with the fans running at 12v – which is actually louder than the Uber profile with the chassis fans at 5v – you should only expect about 95.3-97.6% of the maximum possible performance of the R9 290X. If you want to use the Quiet profile in a quieter chassis with the fans at a much more reasonable 5v, you should expect to only see about 93.2-93.7% of the maximum performance. This is a significant performance loss, but is actually not bad considering that the GPU fan runs about 20% slower when in Quiet mode.
Performance – Crossfire R9 290X @ 4k
| Max GPU Temperature | Max GPU Fan Speed | Avg. GPU Frequency | Max GPU Frequency |
|
|
|
|
|
Running two R9 290X cards in Crossfire at 4k we see some really dramatic results. The added heat of a second video card means that even in Uber mode with a side fan blowing directly onto the cards, there is a performance drop if the chassis fans are only running at 5v. It is only about a 1.75% drop in performance, but this shows that if you want to get the full performance of two R9 290X cards in Crossfire and not require headphones while gaming, liquid cooling is the way to go. Keep in mind that this is on a motherboard with two empty slots between the video cards, so the performance drop will be even larger on motherboards that stack the cards right on top of each other.
The results for the Quiet profile with the chassis fans at 5v look strange at first glance since the performance actually increases after the second run. The reason for this performance increase is actually because the cards are overheating. Yes, for the R9 290X in Quiet mode, overheating actually gives you better performance. The cause of this becomes apparent when you look at the GPU fan speed and average GPU frequency graphs. Since the GPU fan is limited to 2200 RPM when using the Quiet profile, the average frequency drops to about 750 MHz. At that average, the card is actually dropping as low as 730 MHz which "unlocks" the Quiet profile, letting the fan ramp above 2200 RPM. The weird thing is that once the fan is unlocked, the average frequency actually jumps up a bit to about 780 MHz. In other words, once the card overheats and forces an override of the fan speed limitation, you actually end up with better performance (at the cost of the louder fan speed).
| Max GPU Temperature | Max GPU Fan Speed | Avg. GPU Frequency | Max GPU Frequency |
|
|
|
|
|
For Metro: Last Light, we actually don't see much of a performance drop using the Uber profile with the chassis fans at 5v. However, the fan speed log makes it look like the card didn't actually hit it's max thermal load in five benchmark runs, so it is very possible that with a few more runs the performance would drop a little more in that cooling configuration.
Using the Quiet profile with the chassis fans at 5v, we see very similar results to Unigine Heaven, only slightly delayed. Instead of the cards overheating after the second run, they make it all the way to the fourth run before the overheat protection kicks on and overrides the Quiet profile fan limitation.
Analysis – Crossfire R9 290X @ 4k
Just like when analyzing our single card results, there are two different ways we can use to examine our results in more detail depending on what you are interested in. The first is to compare the result from the fifth benchmark run to the result from the first benchmark run. This will show how large the potential performance difference is for each profile and cooling setup and should indicate how much of a difference there is between a review that only used a single benchmark run to a review that used multiple, sequential benchmark runs, discarding the first four or so runs.
The second way to interpret the results is to compare the fifth benchmark result (or the result once the system was fully warmed up) to the highest result we saw for that benchmark/setting regardless of the fan profile and chassis cooling. This will show what kind of performance you should expect in each cooling configuration relative to the maximum performance possible with the Radeon R9 290X.
Starting with the comparison of the fifth benchmark run to the first benchmark run, let's take a look at the performance of two R9 290X cards in Crossfire for each cooling configuration:
*Results taken from the last benchmark run before the GPU temperature forced the card to override the Quiet profile. Due to this, these results may be slightly higher than the actual minimum performance
As expected from the graphs in the previous section, the liquid cooled setup and the Uber profile with the chassis fans at 12v doesn't show much of a performance drop in either Unigine Heaven or Metro: Last Light when running two R9 290X cards in Crossfire at 4k. Unlike our single GPU 1080p results, we actually did see a performance drop using the Uber profile with the chassis fans at 5v. This drop in performance indicates that there is the possibility of online benchmarks being off by as much as 1.75-2.75% when using the Uber profile compared to what you would see in your own system if you do not have a high airflow chassis.
The Quiet profiles results are muddied by the fact that the primary card was overheating when the chassis fans were set to 5v. Given that the largest performance drop we saw in that configuration was about 13.5%, and the largest performance drop when using the Quiet profile with the chassis fans at 12v was also about 13.5%, we can reasonably assume that until the cards overheat and override the 2200 RPM limit, benchmarks published online may be off by as much as 13.5% when using the Quiet profile. This is somewhat of a worst-case scenario since our testing shows that you really shouldn't use the Quiet profile when running the R9 290X in Crossfire, but it certainly is useful to know when looking at benchmarks.
*Results taken from the last benchmark run before the GPU temperature forced the card to override the Quiet profile. Due to this, these results may be slightly higher than the actual minimum performance
As far as how much of the maximum performance you can expect from two R9 290X cards in Crossfire in either Quiet or Uber mode in a high or low airflow setup, the difference is pretty dramatic. Once the system has fully warmed up, you can expect to get full performance by using either liquid cooling or Uber mode in a high airflow system. If you do not have a high airflow system and use the Uber profile, you should expect to get about 96.5-98.4% of the maximum performance.
For the Quiet profile in a high airflow chassis, the drop in performance appears to vary greatly depending on the application. For Unigine Heaven we only saw 80.6% of the maximum performance possible, but for Metro: Last Light we saw a more reasonable result with the card running at 94.2% of the maximum possible. With the fans set to 5v, our results are muddied by the fact that the card overheated and overrode the 2200 RPM limit. Metro: Last Light ran at about 86.3%, which is better than Unigine Heaven, although we suspect that this is due to the thermal override occurring in the middle of a benchmark run rather than at the end of a run. In Unigine Heaven, however, we saw the biggest drop in performance from all our testing with the card only running at about 80.4% of what the R9 290X is capable of performing when being adequately cooled.
Conclusion
The purpose of this article was to find an explanation for why benchmark results available online for the Radeon R9 290X in Quiet mode vary so much from our own testing. By expanding our normal testing procedure, we found two potential reasons for the difference in performance results. Both are very possible, and it may very well be that it is actually a combination of both that is causing the performance discrepancies.
The first potential explanation we found was that the performance of a single R9 290X with the Quiet profile can drop by as much as 5.6% once the system has reached its full load temperature. Putting this in terms of comparing two fictional reviews, Review1 may show the performance of a R9 290X as 58.5 FPS in a specific benchmark if they only ran the benchmark once. Review2, on the other hand, may show the performance of same card in the same setup with the same benchmark settings as 55.8 FPS if they let the system get to full load temperatures before logging the results. Both are accurate, but one is an unrealistic number since you never are going to game for 5 minutes at a time, letting the system cool down between each session. Even if you do loop the benchmark five times like we did in our testing, if you simply take the average of those five runs you would get a result of 56.5 FPS, not the more accurate 55.8 FPS.
If you add a second video card and start testing in Crossfire, the difference is even larger. Even discounting the times when the cards overheated and overrode the Quiet profile's 2200 RPM fan limitation, we saw up to a 12.3% drop in performance once the system was at full load temperature. This time, Review1 would show 91.5 FPS, but Review2 would only show 80.8 FPS. This is a huge difference, and although both are still technically correct, one is clearly an inaccurate representation of what performance you should reasonably expect from a pair of Radeon R9 290X cards using the Quiet profile in Crossfire.
|
|
|
|
|
In addition to the question of whether reviewers are correctly letting the system get to full load temperatures before recording benchmark results, this brings up a second problem that is somewhat new to the art of benchmarking video cards. Considering how much of a difference the overall cooling performance of the system makes on a R9 290X in both Quiet and Uber mode, it calls into question what setup reviewers should use when testing video cards. Should they use a high airflow chassis to find the maximum performance, or a more moderate airflow setup that will give lower performance numbers, but be more accurate for the average user?
Add this to the fact that testing multiple airflow setups is not currently the norm, and it calls into question just how meaningful these benchmarks really are. If the results are obtained by using a high airflow chassis, they might be higher than you would see in your own more moderately cooled chassis. And since other video cards are not as susceptible to heat as the R9 290X, you may end up purchasing a card that ends up performing much lower than you expected. Even to the point that a different video card may have been a better choice. Again, this is not the case for all video cards, but the fact that we saw as much as a 7% difference between a pair of R9 290X cards with the Quiet profile in a high airflow setup versus a low airflow setup is absolutely huge. Even in Uber mode, we saw almost a 3% variance depending on whether we had the chassis fan at 5v or 12v.
Any benchmark you find online is always going to just be an approximation of the performance you should expect from the given component, but with how much cooling affects the performance of the Radeon R9 290X, it might be time to make reviewers jobs just that much harder and introduce testing with variable cooling setups.