NVIDIA RTX Voice Benchmark on GTX vs. RTX GPUs, Gaming Performance, & Voice Quality

by birtanpublished on September 9, 2020

RTX voice has gotten a lot of coverage in a very short period of time largely thanks to our friend Barney Cleese who did a great job of showing how it's supposed to work at the basics of it and inspired us to dive deeper RTX voice is supposed to help with

Background noise it was trained on voice data with background noise and is eventually intended to target reduction of background noise for in-game voice chat or for streaming given the current situation and Vidya really accelerated

The launch of this product it's in beta and they wanted it to help out with things like conference calls so today's feature will test the performance impact of r-tx voice in games with r-tx devices and with gtx

Devices because by the way you can hack it to work on a GTX device before that this video is brought to you by the new be quiet straight power 11 platinum power supply series available from 550 watts to 1200 watts the new straight

Power 11 is Platinum certified operating more efficiently than before reducing heat and lowering noise levels as a result the be quiet Street power 11 platinum is up to 94.1 percent efficient and meets the new low power standards of

Just 0.16 watts when in standby learn more about this high end noise focused power supply at the link in the description below quick breakdown of today's content then first we're going to start with setting expectations of

What this thing is and how critically should we really look at it right now given its current status as early beta we spoke with Nvidia about their r-tx voice solution as well and have some information from them we're going to be

Next looking at benchmarks on our TX and on GTX video cards in game while also using the RT XP voice feature versus not using it so a be testing pretty simple stuff and then further we'll be looking at the voice quality impact if there's a

Change from using an RT X device to a GTX device and finally we'll give you some information on how to do the hacks to get it working on GTX setting expectations then is our first step so we spoke with Nvidia about this and and

Video was very open with us and emphasized very strongly that this is a super early beta the team that's behind it has been working on this for a while now but they weren't really ready for launch

And this was really intended to go out now but because of the current situation and unfortunately I'm going to have to you can figure it out what I'm talking out there thanks to YouTube but the current

Situation has inspired an outburst of conference calls and working from home and so our text voice was pulled forward ahead of all the other projects in the pipeline to try and meet the current high demand for better quality audio

Especially in the mainstream audience people who have maybe never played or streamed games before really on a PC but are doing some form of conference calls where they could use this GTX so before the outrage machine really gets going

The GTX future of our TX voice compatibility is currently uncertain we spoke with Nvidia about the performance results that we saw and NVIDIA hasn't really even had much time to do internal performance testing this is a relatively

Small team and the scope of things and it also has some logistical challenges with working from home right now the GTX does work the performance impacts spoiler alert is bigger and we'll talk about why we think that is later and the

Team isn't certain if they're going to enable it on GTX later but they also don't want to Nvidia doesn't want to worse and perform significantly in gaming with a feature that is initially targeted for gaming use so if it's not

Going to perform well then they probably won't turn it on we'll see if the hack remains in there though eventually this will become a very streamer or content creator targeted piece of software and the end goal is to optimize it more

Heavily and get it running more on the tensor Hardware tensor cores on our T cores in the device this runs on qu DNN which is an Nvidia solution and I think probably before giving any more background information on this let's get

Into the benchmarks and then we'll talk more about what it all means and setting the stage of expectations for RTX voice in the future well start with a benchmark of our TX voice running on our TX cards then we'll show how it does on

The GTX 1080 FTW this testing uses our standard test methodology for GPUs except for our TX voice when our TX voice is installed and enabled we're using a math drop PC 37 X headset with a 3

Thousand rpm industrial knock to a server fan pointed towards the wearer from about half a meter away the intent of this is to create a less reasonable representation of background noise to try and stress the software more if it

Even works that way we don't really know yet we then count throughout the benchmark with occasional pauses so one two three four and so forth to keep some regular noise that is

Repeatable and testing we attempt to give our TX voice something to do with this but do break at every 10 so we pause at 10 for a couple seconds we pause at 20 for a couple seconds and it's all to just try and create some

Stress workload that is repeatable this is with Windows voice recorder running which we found to have the lowest interference with benchmark results and which works with r-tx voice we had trouble getting the speaker setting and

R-tx voice to work so these benchmarks are only for RTX voice running with the microphone the impact should be bigger with speaker processing for a discord call which we did try by the way but we were unable to get it working in time

For the benchmarks the first test is with shadow of the Tomb Raider at 4k with all our TX features disabled this is strictly rasterized game content so the RT cores and tensor cores are not taxed by the game the r-tx 20/80 super

Ran at 62 FPS average base line with our settings 54 FPS 1% lows and 53 FPS 0.1% lows frame time consistency is extremely good in this title for the 2080 super and that shows here enabling our TX voice dropped us 256 FPS average which

Is a reduction of 9.3 percent on the RT x video card this would imply to us anyway that there's overhead somewhere outside of the RT course since they are unutilized by the game the one percent and point one percent low reduction is

About 10.5% seeing performance hit in a non RT benchmark again indicates to us that something else on the GPU pipeline is loaded it seems heavy on copies right now looking at the resource monitoring and Nvidia told us that it's not yet at

The optimization stage but the company will be working on it as the application matures it's very clear that our TX voice isn't only running on our TX Hardware right now otherwise we should see nearly zero performance hit two

Games that leave the RT ax Hardware idle switching to the RT X 26 TKO at 1440p in the same game we measured a 72 fps at baseline with frame times again highly consistent on the nvidia card the drop was about eleven point seven

Percent from baseline to RTX voice enabled with a 14% hit to low frame rate performance we were next testing the GTX 1080 to reiterate an earlier point this isn't supposed to work on GTX and it is a hack so you get what you get there's

No Nvidia hasn't validated this internally the results are interesting the GTX 1080 falls from 71 point 4 FPS average at 1440p to 59 point 6 FPS average which is a sixteen point six percent performance hit that's more than

What we saw in either of the other two RT x 20 series benchmarks the one percent on point one percent lows were additionally hit and dropped 17 to 18 percent total off of their baseline our initial suspicion was that our TX voice

It might be actually leveraging our T or tensor hardware to some extent which would explain the difference in the impact despite the somewhat obviously non RT loading in the pipeline when we spoke with Nvidia to try and better

Understand why the performance Delta was higher on the GTX card the company noted that it hasn't really gotten into the performance testing stages yet and the product manager with whom we spoke didn't have any personal experience

Running performance tests against the GTX cards we have more benchmarks in a second but a quick note first and video really is in an early beta stage right now and the company is trying to get it working first so just to kind of preempt

The usual and video conspiracy theories will quote Gordon of PC world alright internet put the can of Budweiser down relax this is not Nvidia intentionally hindering GTX performance that's not what's happening here just to

Really make that clear it's not like Nvidia is trying to torpedo GTX performance on its software which isn't even ready for performance testing yet they don't even want the hit to be ten percent with our TX cards because that's

Obviously a massive hit for gaming especially if you're already in dealing with resource contention with streaming software with envy Inc or something else or whatever else is going on in the background so Nvidia doesn't even want

To be at the ten percent number that they're at now with our TX hardware and the GTX hit it's not like this is an intentional self-sabotage now and video might disable it for GTX in the future even though it may in the

Future improve and performance we have no idea if the performance improvement is somewhat linear on both devices let's say they bring the RDX 20 series cards down to 2 to 3 percent performance hits something like that and they can cram

All that processing in there a couple millisecond window while the games dealing with frames maybe the GTX card comes down to something like 5 to 9 percent somewhere in there if that's the case we'd like to

See Nvidia offer GTX for RTX voice but we'll see we'll see what they do either way anybody who tells you in the comments that this is an intentional hindrance has no idea what they're talking about

Because if you look at the software it's incredibly simple right in-house so it's not advanced enough to really do that yet our t-rex GPUs had architectural improvements outside of our tea cores too by the way so that's probably at

Play here and it may have to do with the bandwidth on the bandwidth difference between the two series of cards we tested it could also have to do with the way that our TX devices handled copies specifically this is a very copy heavy

Workload and if they're advantaged there then that would explain some of it the next test is minecraft RT X we recently published a benchmark video on this game and it's especially useful here like we said in the benchmark minecraft slow

Native graphics quality meant that a lot of the work could be pushed to our T cores without as much burden on things like the geometry pipeline for this one at 4k and with our TX force off the 20/80 super did about 37 FPS average at

4k and that was with our tested settings of 8 chunks and dl SS on the drop was 234 0.5 FPS with our TX voice on that's not a lot in an absolute sense since no one's really going to notice that but it is a percentage difference about 6.7%

This will scale up to 1080p or 1440p mind you so it should be fairly similar across the board you might still have a 7% hit at 1440 the performance impact is lower than in non r-tx games so we assume that our hypothesis is

Potentially being somewhat proven with low disproportionately shifted to our T Hardware on the GPU we're now choking the pipeline outside of usual areas in choke in a game like geometry our TX utilization

The game should have resulted in more resource contention if r-tx voice were heavily leveraging r-tx hardware and so the FPS impact we think should have been higher if that were the case the results we got tells us that it may be relying

More on the traditional cuda cores rather than the RT course or a tensor course the RT x 2080 supras voice quality was about the same as in other games so we didn't lose any performance in the application on the voice quality

Either we next tested at minecraft with the RT x 20 60 k out benchmarked at 1080p instead this card already struggled so the results aren't very surprising the performance hit is about 5% here again it's all about relative

Numbers not just raw fps 2 FPS doesn't mean a world of difference and is frankly sort of irrelevant but we're more using this as a tool to try and figure out where the performance loss is happening in the pipeline the RT x 2060

Kayo is much more stressed on its RT hardware than the 2080 super wise where it's dealing with a lot of pixels as well as this is at 4k so we again think that this is indicating towards maybe more CUDA utilization than not finally

Our last test is time spot for GT 1 & 2 which is a synthetic workload that doesn't use our T Hardware our score 4 times pi extreme was 53 52 points for RTX voice off with the 2082 / or 48 22 points with our TX boys on the

Difference is about nine point nine percent not shown here GT one is more of a tessellation workload while GT 2 is more of a memory workload and in both instances we're at about nine point nine percent performance reduction times by a

Non extreme with the 2060 kayo applauded about an eleven point five percent performance hit 74 fifteen points down to 65 64 GT one fell eleven point four percent from about forty eight to forty two point five and gt2 fell about 11.5%

From forty two point eight to thirty seven point nine in time spy that's a big FPS hit the benchmark cares about every fraction of a frame the last test is for the GT x 1080 FTW won and times by where we saw a 17 point seven percent

Reduction in score or 17.1% in GT won the tessellation load and 18.2% in gt2 the more memory intensive load this follows the gaming results pretty well and is about what we saw in the previous tests

Another reason that Nvidia might not want the 1080 supported or the GTX 3 supported for RTX voice would be if the sound quality is worse where it becomes actually just a hindrance to usage rather than a benefit so we compared the

Twenty eighty super to a 1080 in the same test configuration we had an ox with 3000 rpm fan about half a meter away so obnoxiously close at this point this is no longer realistic this is not representative of how any one sets the

Computer up it was strictly a torture test for RTX voice so Knox about 3000 rpm server fans about here and blowing more or less straight into the headset microphone and it's if it survives that then it'll survive a lot of things so

That was the test set up the 1080 had severe issues with dropping out and sort of robotic sounding audio where you can hear the artifacting in it it's the same thing anyone who's worked with any kind of noise suppression and Adobe Premiere

You're very likely familiar with this as well and we'll play clip in a second the twenty eighty super seems to do better but it was still flawed it certainly wasn't perfect and also we should note that we're not really positive how

Accurate run to run RTX voice is we did do a couple of different tests and it seems consistent but we can't with certainty state if a gtx card is always going to be worse quality than an RT x card however

Observational e with fairly controlled as far as the world of audio goes testing goes it was definitely worse on the gtx card so this was certainly not as amazing as initial coverage had us believing but it's still impressive

Overall just the gtx card was really bad so anyway here's a clip of the gtx 1080 with our TX voice disabled this will give you an understanding of the room audio and then we'll play it unable then point out the mic one two three four

Five six seven eight nine ten and here's a clip of the GT X 1080 with r-tx voice-enabled while running times by in the background so it's effectively gaming while also running RT x voice it's one of the most

Intensive possible loads that we could generate for it GT X 1080 testing times for extreme with the ax voice on 1 2 3 4 5 6 here's the RIT X 20 80 super end time spy with r-tx voice enabled Archie X 20

80 super test 2 was on 1 2 3 4 5 6 7 8 9 10 and finally a reminder of what our TX voice sounds like when it's off with just the room noise and this time with 20 80 super although it's irrelevant at this point because it's not doing

Anything on a software level it's just how the noise sounds to the mic let's play that one 1 2 3 4 5 6 here's a quick clip of the r-tx 20 60 KO running with our TX voice enabled while playing a game 1 2 3 4 5 6 7 and here's

A final recording of our TX voice enabled while just at desktop with a bit of a cleaner audio track that we recorded 20 super at the desktop 1 2 3 so as a reminder this is all very new software and video isn't able to really

Help us remotely with any kind of insight to testing yet because they haven't done much of that themselves yet this was pulled way ahead in the pipeline in a more normal use case so with fairly annoying radiator and GPU

Fans blowing but not into the microphone so testbench positioned about 1 meter away or let a little bit less and with a neck 62 with fans at full speed and with an RT X 2080 super with fans intentionally set to full speed on the

Founders edition card you end up with probably at that distance something around 45 DBA or so and and in that instance without the wind deploying straight into the headset microphone like the knock who a fan was doing it

Works pretty well and overall the sound quality is good ish now you still get that robotic the aberrations and weird noises that happen in the background but it is certainly better than some of the tests

We just showed with the nakta fan blowing into the mic so this technology it's it's good but right now this is not something we would use for streaming its its beta it should be probably called alpha and there's a performance hit that

Is significant in gaming applications right now that should change we're told and there's also kind of questionable gains and audio quality sometimes where you almost rather have the more natural sounding voice rather than the weird

Clipping that you get otherwise all that said this is still overall shaping up to be impressive technology it has a little bit to go but it's going the right direction if you want to set this up on GTX here's what you do it's pretty

Simple so we found these steps in an online form then we modified them for more clarity you download and run the r-tx installer first step to navigate to C Drive temp and VRTX voice we'll put this on the

Screen env a FX is the final folder and then you copy RTX of voice dot n VI as a backup keep that and open up our TX voice nvi with notepad plus plus or something delete attacks that will show on the

Screen but it is in the constraints section and that text says constraints property name feature our TX voice and so forth so you delete that whole block make sure the rest of the formatting and space and is maintained the same on the

Sections above and below it and after that go back up a level to n VRTX voice and the temp folder that's in and run septa Exe as admin if that doesn't work navigate up a level and grant folder permissions to your user account to that

Folder or if you want to be lazy about it you can grant all permissions to it and an attempt again and maybe reset your permissions after that that will allow it to run and should work well I mean should should work as well as it'll

Be allowed to work the GTX entity doesn't always sound garbage but it sounds garbage when you point a fan at it more so than the twenty eighty super does and we don't have an explanation for why we don't know if that's because

The RT and the tensor cores or if it's weaker hardware or just the software wasn't doesn't have any parameters in it for Pascal no idea but it didn't sound the same and it was under roughly the same test conditions so someone's going

On and it might be because it's not supposed to work a couple notes we found a few bugs the r-tx voice outside of the application itself one of them was repeat initialization errors that were really frustrating where it would not

Detect our audio devices we finally figured it out and it's simple when you go to install r-tx voice make sure that your audio device is mic and headphone are both plugged in and on or active otherwise at least in our case it bugged

Out and once you connect to them later it didn't understand what they were so they have to be in there when its installed we also notice that on boot it indicates that it's failing to initialize again even if we don't launch

It so it starts automatically which you go kill and that was config if you want but starts automatically and then fails to initialize and also if you want to exit the application you have to kill it in task manager killing it in the system

Trade doesn't actually kill it double final notes here we've seen some confusion about what how this works so there's a lot of talk about AI and deep learning and all this all the buzzwords they're actually used but the thing that

People commonly misunderstand is this is not live while you're using it sending your data back to the server to be analyzed and then sending it back to you to be played back so you you will not be sending your information to Nvidia

Unless you opt into it on their feedback form on the website they're not listening to your calls as far as we understand obviously r-tx voice and vidya told us does not store or transmit any of the data that it encounters back

To a server it doesn't do any any remote data processing it's all done locally and done on the card that you have installed and all of the deep learning testing that this marketing talks about in the pages they quickly threw together

That was done on site before the software shipped so it's not being done live with your data going back to the server just wanted to clear that up if you wanted to contribute to Nvidia's development of r-tx voice they do have a

Feedback form open it is not only opt-in but you have to go find it on the internet so as far as we're concerned that's fine because it's it's all voluntary by the user but if you want to feed data into it go for it they have a

Process where they have both a machine oversight and an engineering oversight because to quote the rap garbage data it produces garbage results so they need to make sure that the data that comes in is actually useful to them and not just you

Holding a leaf blower against the microphone and then playing never gonna give you up in the background I'm sure they'll get a lot of those submissions now though and that more or less recaps it so it's overall okay if you have a

More reasonable amount of background noise like your computer and it's not a blower fan pointed at the mic it seems like a fine thing to use you will have a performance hit and games we probably wouldn't recommend this for streams

Where it's maybe you're a source of income for you because it's beta software and it's buggy but for conference calls if you have people complaining about audio quality well then it seems like it's an okay thing to

Try out in the least the concern we would always have with this type of software is that any kind of noise suppression you deal with potentially having parts of your speech cut out without you knowing it the recipient

Will know because they don't understand what you're saying but that's always the concern we encountered a little bit of that but it was under those extreme test conditions and not under the reasonable ones with

Just a bunch of computers in the room making noise indirectly hear the mic so that's it for this thanks for watching fun stuff to test we'll look at this some more as it develops it's certainly interesting and it seems like there are

Real applications for this but we're gonna give it some time to develop some more subscribe for more go to store documents next it's not net or patreon.com slash gamers Nexus tops out directly and we'll see you all next time

Related Videos

because the Atari VCS is definitely something that's coming out we need to do a teardown on one to get some perspective and Atari if you don't know it&...
Hey how's it going guys Jack I'm out here with the toasty rose and today we're going to be doing a four hundred and fifty dollar PC that any of you ...
The biggest rule in testing coolers is to never trust anything don't trust the numbers don't trust the software don't trust the firmware and definit...
Hey what's up guys Jack and Matt here with the toaster brothers and today we're gonna be finding out if this 350 dollar laptop can game this is gonna be...
Everyone welcome back to another episode of ask GN sorry I'm not doing the intro with snowflake for this one we're flying to Taiwan and we need to get t...
Everyone welcome back to another hardware news recap for the week we are in Taipei right now working on a factory tour series and we actually have a point for y...
Hey how's it going guys Jack and Matt here with the toaster bros and today we're gonna be doing our first 20 20 thousand dollar gaming PC but guess what...
Coursers a 500 has been an embarrassing show of performance for the company's first revisit to air cooler isn't a long time but the upside is that it ma...
When I received the thread Ripper 3990 X on loan from a yet unnamed youtuber we quickly threw together an extreme overclocking stream and got the CPU to about f...
Every one so we're in Taiwan right now we're doing our factory tour series and while we were out here we stopped by Lian Lee's offices and they want...
Hey what is up guys Jack and Matt here with the wait what do you do it on sex you back your desk back to your desk sorry about that okay hi Jack and Matt here w...
We thought we'd found the limits of viewer interest when he posted our tour of a screw factory last year as poor Freight it was more of a screw shop than a ...
Everyone welcome back to another hardware news recap for the week we're in Taiwan where we venture outside to film things that's always a nice change of...
Hey how's it going guys Jack and Matt here with the toasty rose and today we're gonna be doing a 350 dollar gaming PC build it's gonna be really awe...
out of all the factory tours we've done until today we'd never gotten an opportunity to see how water blocks and open-loop cooling components like fitti...
Paints is one of the chief processes to dial in for product manufacturing and today's tour will bring us to a taiwanese factory replete with a mix of automa...
hey what's up guys Jack I'm Matt here with the tasty bros and today we're gonna be showing you our 2020 CES recap live from our CES Airbnb but befor...
except you can't wipe your ass with it well I mean we're going to take a gamble on the YouTube D monetization system and talk about the impact of corona...
everyone we're closing out our trip in Taipei for our factory tourists we have a lot of factory footage live already and we have more coming to the the chan...
hey what's up guys Jack and Matt here with the Thai stereos and today we're gonna be finding out is a six hundred dollar laptop worth it in 2020 it'...