We recently published step by step instructions on how to enable RTX Voice for all your work from home and gaming needs and some users asked us just how big of a performance impact this tool would have. Considering most games don't really use the Tensor cores in an RTX GPU, our knee jerk answer to that would have been virtually none but fortunately, we decided to run our own tests. As it turns out, there is a performance impact in games while enabling RTX Voice. We will look at some benchmarks and then discuss why this is happening at the end.
NVIDIA RTX Voice Performance Impact: Typically Between 6% to 11.6% Depending On The Graphics Card And Application, Under 1% in Some Games
Before we begin, an advisory: while I spun up some titles that I had lying around for some quick and dirty benchmarks, I did not include these in the results here. The reason being that I saw extremely large deviations in performance hits and including just a few gaming titles would have been tantamount to cherry-picking. Ideally, I would run this test over a sample of at least 25-30 modern AAA titles but my macros aren't properly set up right now, it is the middle of the night, and I am choosing to take the easy way out: synthetics.
By using the standard synthetics, we can get the figurative ball rolling on the conversation and let our readers and other third party testers delve deeper from here. A small sample of synthetics is almost always better for generalizations than a small sample of gaming titles. The test bench we were running consists of an Intel Core i9 9980XE (stock), 32 GB RAM DDR4 3000, RTX 2080 Ti and RTX 2060. For testing purposes, both the outgoing mic and incoming audio stream are being filtered.
Let us start with the lower end of RTX first, the RTX 2060 (an EVGA RTX 2060 XC in this case):
|RTX Off||RTX On|
|Unigine Heaven 1440p||1449||1279|
|Superposition 4K Optimized||6190||5456|
The RTX 2060 takes a moderate hit in performance when RTX Voice is turned on. The worst-case scenario we saw was a performance hit of around 11.6% while as the best-case scenario in synthetics was around 7.4%. Interestingly, however, some gaming titles showed very slight performance hits of under 3%. It seems that the amount of performance hit will vary depending greatly from application to application. The average hit for our sample of the RTX 2060 while using RTX Voice was in the 9% range.
We also tested the high end of RTX with an RTX 2080 Ti (an ASUS RTX 2080 Ti):
|RTX Off||RTX On|
|Unigine Heaven 1440p||2613||2422|
|Superposition 4K Optimized||11870||10902|
The flagship of NVIDIA's RTX lineup with the largest Tensor core cluster, the RTX 2080 Ti, takes a low to moderate hit with RTX Voice enabled. We saw 9% as the absolute worst-case scenario (just 1 data point) and 6.7% as the best-case scenario in synthetics. In gaming, there was actually a title that saw under 1% hit in performance (which is negligible and not significant as it could easily be the result of standard deviation between runs). This is remarkable because it shows RTX Voice running as intended under 'ideal' parameters. For anyone interested to run their own tests, the title in question was GTA V. On average, the RTX 2080 Ti saw a performance hit of around 5-6% with RTX Voice On.
Interpreting these results
So what exactly is going on here? To begin with, there is a very large deviation in results, to the point where I think this application is not completely optimized as of yet. Considering COVID caught everyone by surprise, NVIDIA's development team might have seized an opportunity and fast-tracked development of RTX Voice. In other words, performance will almost certainly improve as the application undergoes polish. This also might be why the app is still in its beta phase *cough* obviously * cough*.
I would also hazard a guess and say the vast majority of the performance hit right now is stemming from this lack of optimization. The reason being that certain games saw very little performance impact while others saw a large impact and this is something that can only be explained by the above (if the performance hit was a function of GPU power the app absolutely had to use then there would be no title with low-performance hits).
That said, RTX Voice's use case will always exist in the less-than-perfect-real-world where running any hardware that is on the same package as the shader cores will result in some non-zero performance deterioration. TDP and thermal envelopes are a sensitive system and it is quite possible that your performance hit will vary greatly depending on the amount of cooling capability and power the graphics card has access to.
All in all, using RTX Voice for work from home use-cases is still a no brainer, and if you are not using a 100% of your GPU right now then it is even feasible for gaming. I am also sure that most of this performance impact will be mitigated when RTX Voice exits beta. The best-case scenarios we are seeing right now should easily become the norm.