Today, we consider NVIDIA Titan RTX for overclocking, gaming, thermal and acoustic performance, look at the first of two cards in the lab. We have a third card that is going to trade for a defective device, which works around the 1350MHz clock lock we detected, but it doesn't come until this review goes live. Titan RTX costs $ 2500, replaces RTX 2080 Ti by about 2x, but only allows for an additional 4 streaming multiprocessors. With 4 more SMs and 256 more tracks, there is not much to gain from game scenarios. The big win is in memory-based applications, as the Titan RTX has 24 GB GDDR6, a marked increase from 11 GB on an RTX 2080 Ti.
An example of a use case may be machine learning or deep learning or more Traditional, 3D graphics rendering. Some of our internal Blender project files use so much VRAM that we have to render instead of the slower CPU (instead of CUDA acceleration), as we run out of the 1
Finally, note that we were stuck at 1350MHz with one of our two samples, something we've been working with NVIDIA to investigate. The company now has our defective card and has dealt with a working card. We bought the defective Titan RTX, so it was a "real" retailer. We just wanted to help NVIDIA troubleshoot the problem and the company is now working on it.
The test methodology has completely changed from our latest GPU assessments, which were probably for the GTX 1070 Ti series cards. Mostly, we have reviewed the host exam and been updated with new games. Our game selection is cautious: Time is limited, and after analyzing our previous testing methods, we identified deficiencies where we eventually waste time testing too many games that did not meaningfully different data from our other tested titles. To better optimize our time and try "smarter" (instead of "more", which was one of our previous goals), we have selected games based on the following criteria:
- Game Engine: Most games run on the same group popular engines. By choosing one game from each major engine (eg Unreal Engine), we can ensure that we represent a wide sweep of games that only use the built-in engine level optimizations
- API: We selected a selected group DirectX 11 and DirectX 12 API integrations, as these are the most common at this time. We want to include more Vulkan API testing, as several games are sent with Volcano
- Popularity: Is There Something People Actually Play?
- Long Life: Whatever popularity, how long can we reasonably expect a game to go without updates? Game update can damage comparative data from previous tests, affecting our ability to cross comparison of new data and old, as old data may no longer be comparable to patch
Game graphics settings are defined in their respective charts.
We also test most games in all three popular resolutions – at least we are too high-end. This includes 4K, 1440p and 1080p, which allows us to determine GPU scalability across multiple screen types. More importantly, this allows us to begin identifying the cause of revocation, rather than simply saying there is performance abolition. If we know that performance increases harder at 4K than 1080p, we might call this the indication of a ROPs benefit, for example. Understanding that performance behaves as it does is crucial to the future expansion of our own knowledge, and therefore prepares our content for smarter analysis in the future.
For the test bench correctly, we now use the following components:
GPU Test Bench (sponsored by Corsair)
NVIDIA Titan RTX Overclocking
|NV Titan RTX Overclock Stepping | TimeSpy Extreme Stress Test | GamersNexus.net|
|Peak Frequency||AVG Frequency||Core Offset||Core Voltage||Memory Frequency||Memory Offset||Effect||Watt||Fan Speed ||Active Temp||Fan Speed / not||1905||1830||1905||1905||1905||1905||0.975 [1750.2] 114%|
|1860||1860||1860||1960||1975||1750 , 2%  [%]|
|2025  075] 2010||2040||2025|
|2025||3700||] 1,018||1750,2||114%||330||3700  66||N / A||N / A||200||N / A||200||N / A||N / A||N / A||N / A||N / A||N / A||N / A||F – Crash|
|Error – Application Crash|
|2040||2030||175||1,018||2030.2||1080||10 4%||330  3700||66||P|
Overclock knit epping comes up first. You need to know this information to join our gaming benchmarks. The Titan RTX overclocking process follows the same steps as 2080 Ti, and is instantly lubricated by the cooler under the exclusive fan curve built into the VBIOS. Under complete storage settings, we saw an average frequency of about 1800 MHz in TimeSpy Extreme, with a 1920MHz peak frequency. Power measured about 280W through GPUZ, although we have separate measurements later. The fan seems to stick to around 1515RPM to maintain a thermal measure of 75 degrees Celsius.
Maximum power measure is 114%, giving us a limited number of rooms to increase performance. We ended up with an 1830MHz average, up 30MHz, and it's without core frequency shifts. We were also instantly throttling at 88 degrees, causing the clock to jump around between 1780 MHz and 1850 MHz. A few steps later, we blown the fan speed to 3700RPM, or maximum speed, and so performance abated significantly. This is an illustration of how cooling coolers are limited. Now someone has traditionally complained to us that we were too hard on NVIDIA blower coolers for this type of card, because their idea was that you put a bunch of blower cards in a tight box for reproduction, in which case it is more efficient and removes heat from system than axial coolers. This time, when NVIDIA has switched coolers, this argument is no longer standing. It's just a pretty worse cooler than any partner would give, yet doesn't even give the only benefit a blower would offer, giving better performance in multiple card back scenarios. We finally crashed with a 200MHz offset, then found the core clock stability with a 175MHz offset. The maximum frequency was 2085MHz when it was around 62 degrees Celsius, and found a resting point at the 2040MHz peak below 66 degrees Core. Average frequency landed at 2030MHz, with the maximum stable memory frequency offset at 1080MHz, which we think is one of the memory loops.
We could push this card harder with greater power interference. With 100% fan speeds, noise levels are unacceptable for most users, but thermal performance is acceptable for overclocking. The problem is that, just like the previous RTX cards, we are bound by an over-protective force target and voltage limitations. We can try and against VBIOS to see if we can surpass this, or maybe just tear one of the shunt resistors off the board and make Buildzoid's courage.
Either way, it's our overclocking setting for the tests. Overclocking is limited to this and it is entirely due to the power limit. The GPUZ was reading the power train at 330W when it was overclocked, but the VRM can handle far more than this, as our 2080 Ti reference PCB analysis shows, and this is the same PCB as that.
Titan RTX vs 2080 Ti, SLI, & 1080 Ten Benchmarks – Sniper Elite 4
Our gaming standards begin with the Sniper Elite 4, which is one of the best-built games with a modern API. Using DirectX 12, Async Compute and 4K / High settings, NVIDIA Titan RTX ends on 112FPS AVG, with low 90FPS and 87FPS 1% and 0.1% low. By comparison, the 1080 Tis in the SLI makes about 170FPS AVG, or the SLI 2080 Tis makes the 210FPS AVG. We also observed 2080 Ti at approx. 108FPS AVG, establishment of a difference of approx. 4FPS, or 8.96ms average future for Titan versus 9.2ms future for 2080 Ti. We cannot think of many people, if any, who can identify a 240- microsecond difference in frame-to-frame intervals.
Overclocking of Titan RTX to approx. 2040MHz core, it gets to 126FPS AVG, which surpasses the overclocked 2080 Ti by about 3%. The next thing we need to test is NVLink Titan RTX, which we will be working on immediately after this.
Here's a look at the future between the shares Titan RTX and stock 2080 Ti cards. As a reminder, the futures are the most accurate representation of frame-to-frame pacing, or the time interval from one frame to the next, and are the best way to objectively illustrate the fast-paced experience without any average. Both cards perform about the same. Lower is better and more consistent is best, but these cards are both low in the future and consistent in the future. NVIDIA has done well with the future consistency of this generation, for the most part, and none of these cards experience a remarkable future spike or hang. It would be difficult to tell the two apart. For reference, 16.667ms is 60FPS with 8 at 120FPS. What we care about here is that there is never more than an 8ms deviation from the average, and it is therefore unlikely that the user detects any hitching or interruption of fluidity. Some call this "smoothness."
Overclocking Titan RTX introduces more variance in the future, which you will see in its line plot, but nothing serious. We regularly hit 3ms future over baseline, from 7m to 10m in some cases, but this is still relatively fluid and generally useless to the user.
Titan RTX Gaming Benchmark vs 2080 Ti – F1 2018
F1 2018 gives us a look at the EGO engine with DirectX 11, goes back to more common APIs and away from the gleaming example of Dx12 which is Sniper Elite 4.
For F1 2018 at 4K and ultra high, we are clear GPU bound with high ceilings set by the CPU. In our CPU reviews, you see FPS as high as 300 with the right CPU, so this is a good way to really test the limits of the GPU and avoid the limiting influence of the CPU. This is further illustrated by the SLI 2080 Tis by 169FPS AVG, which clearly pulls away from the package. For the Titan RTX, which ends up on the 110FPS AVG stock, place it just before the 20 FE Tiets 99FPS AVG, or 2080 Ti XC Ultras 105FPS AVG. The maximum gap is no greater than 10%, and it is reduced to approx. 5% with the AIB partner model 2080 Ti. Overclocking the Titan gets it to 118FPS AVG, just before the overclocked 2080 Ti FE at 114FPS AVG.
At 1440p, the overclocked Titan RTX runs at 185FPS AVG and low at 87FPS 1%, and adds just over 2080 Ti FE Hybrid overclocked cards by about 3%. Titan RTX stock cards hit the 175FPS AVG, melt it between the overclock 2080 Ti and make 2080 Ti XC Ultra at 170FPS AVG, or make 2080 Ti FE at 160FPS AVG. Differences here are minimal, which is likely to be expected when Titan RTX only adds 4x SM to the GPU, moving to 72 SM from 68 SM, so the lack of performance makes sense.
At 1080p we clearly enter a bottle neck at approx. 213FPS AVG. This becomes a CPU-bound scenario, and the results here are therefore relatively meaningless. All we learn is that this CPU can't keep up with the 1080p cards.
The shadow of Tomb Raider – Titan RTX Benchmark
The shadow of Tomb Raider is a DirectX 12 title that is still relatively recent. For this, note that the SLI tests were performed without anti-aliasing as the TAA causes alternative frame mapping problems due to the lack of previous frame data on each card.
Titan RTX ends up at 72FPS AVG for stock performance, which is not much better than 2080 Ti FEs 67FPS AVG. 7.6% performance cancellation is undesirable in view of a 100% price increase, but note again that the bigger reason for buying this card is for its VRAM increase, not for any other reason. It comes mostly to the benefit of non-gaming applications. Overclocking the Titan RTX puts it on the 78FPS AVG, which is about the same as overclocking a 2080 Ti card, even though our 2080 Ti was able to push higher in the core clock. This is a lot of silicon lottery and probably part thermal density.
At 1440p, the Titan RTX runs on the 119FPS bearing, which is functionally tied to the 2080 Ti FE when there is no thermal limit on Hybrid mode. 2080 Ti XC Ultra sits at 116FPS AVG, with FE at 114FPS AVG. Overclocking puts 2080 Ti in the lead, again due to a higher clock, with the Titan RTX just behind at around 128-130FPS AVG.
At 1080p, the cards still show scaling and have not yet entered a hard limit, but we see tha Titan RTX and 2080 Ti XC Ultra – and 2080 Ti FE, for that matter – are about the same performance level. There is no difference between these devices.
Titan RTX Benchmark – GTA V
GTA V gives us an interesting spin on the results. For this, first tested at 4K, the Titan RTX runs on 93FPS AVG, which surpasses both stock and overclocked 2080 Ti FE cards. 2080 Ti XC Ultra ends up on 88FPS AVG and will surpass Titan RTX with an overclocking. Overclocking the Titan RTX gets it to 98FPS AVG, which is still better than the two GPU classes above it. Although the Titan card looks better here than in some games, it is still clearly not worth using for gaming.
At 1440p, Titan RTX ends near the top of the 156FPS AVG, where we begin to be tied by the game engine. This 187.5FPS engine frame makes it impossible to see how much more the top space it is for the SLI cards, for example. However, the Titan RTX stock card even has an RTX 2080 Ti FE overclock, which makes no meaningful difference.
We have 1080p results, but it is CPU bound to a point that everything is equal. There is nothing to learn here, so let's move on.
Titan RTX Benchmark & Overclocking – Far Cry 5
Some recent game updates made it clear that it was worth testing in Far Cry 5, so we overturn 2080 Ti FE numbers for this. Titan RTX ends up on 70FPS AVG, with the overclocked variant getting 5% performance for 73FPS AVG. 2080 Ti tested lower than a few months ago, instead placed on 66FPS AVG, with XC Ultra on 67FPS AVG. Titan RTX ends with 6-7% when everyone is tested on these drivers and with this game update.
When both are in stock, 1440p scale places Titan RTX as 4.9% before 2080 Ti FE. Scaling is reduced as the resolution decreases. Our original 2080 Ti result was around 126FPS AVG, although the new drivers, game update, and small Windows update affected scoring.
1080p offsets the results. We become CPU-bound, here, so all the best results are within reasonable margin of error and are covered by the CPU.
NVIDIA Titan RTX Power Consumption
For power consumption testing, we measure between the wall and the system for total system consumption. The test platform is 100% controlled, including control of all minor railway system voltage, fan quantity, speed and type, keyboard and mouse, and every other part of the system. Lack of control even though some of the motherboard tensions would shed these readings and leave them inaccurate. We also log over time so you can see a real look that includes tops and downs, rather than an average sum of the test.
This chart is with Ashes of Singularity: escalation under a 4K / crazy workload, pushing the GPU to its limits. Running the Titan RTX card, our total system consumption is consumed at 480W, and pulls out the 2080 Ti Founders Edition system with 24W. The only thing that pulled out the Titan RTX card was our vigorous modded Vega 56 contraption, which used a 250% power measure to burst the current past the inventory and allow the overall system to retract around the 640W peak. Vega 56 is typically closer to the 350W series for total system power.
NVIDIA Titan RTX Thermals
Thermal testing puts Titan RTX under load with either FurMark or 3DMark, depending on the test. Thermocouples are attached to a hotspot GDDR6 module and the hotspot MOSFET, with GPU-Z being used to log the rest of the performance counters. For our power virus workload, we monitored a maximum GPU temperature of about 80 degrees Celsius, with the hottest GDDR6 module at around 76 degrees. The mosfet ran around 80 degrees. The GPU temperature is warm, definitely, and will affect the frequency. We already know that the frequency steps are increasing every couple of degrees, so being in the 80's for the stock cooler is unimpressive. This becomes more of a problem for anyone who stacks several of these cards in a matrix. As for the VRAM temperature, it's okay. 90-95 degrees is the specification, and the MOSFET can take 125-150 degrees, so both are within the operating specification.
With Firestrike endurance testing we saw the frequency beginning at approx. 1935MHz when it is 40 degrees, then drop to approx. 1830 -1845MHz and fluctuates based on core temperature. The temperature seems to hit around 84 degrees before it seals a little around the clock to beat 80 degrees.
Noise levels for Titan RTX
This is the final chart for the review. Noise for Titan RTX runs a 33.9dBA level when set to 41% speed, which is the slowest possible. It is much higher than most other partner models during idle situations. At an average fan speed of 60%, we are at 42.9 dBA. Coming to 100% push 58dBA. Expect to be around 43 dBA in most cases. This card is acoustically better than the table partners, but no card sponsor Titan RTX card will exist.
It should not surprise anyone that this is not a valuable purchase for gaming or enthusiastic users. We can test Titan RTX in Blender or other applications later – first and foremost, after the professional applications provide the ability to utilize RT and Tensor cores – but it's beyond the radar for today. If you really have to spend $ 2500 on a GPU, you get two 2080 Tis instead, though we don't necessarily recommend it either.
Editorial, Testing: Steve Burke
Video: Andrew Coleman, Keegan Gallick