Nvidia recently announced DLSS 2.0 with their Geforce 445.75 game drivers, which is the latest update to their RTX DLSS released in 2018.
DLSS (Deep Learning Super Sampling) works by rendering the 3D scene at a lower resolution than what your graphics settings are set to, then upscaling it with AI deep-learning algorithms to enhance the details for a superior performance increase.
DLSS is only available with Geforce RTX graphics cards, due to their dedicated Tensor cores, which are specialized AI cores, that make use of new temporal feedback techniques to reconstruct details in the image, giving crisp details with massive performance gains over not using it.
If you do own an RTX card, you will be aware that it can support Raytracing (both shadows & reflections), but do note that, just because a game has DLSS, doesn’t mean that it has Raytracing, as they are exclusively different, while at the same time, fall under the RTX umbrella.
Check out my list of the Best RTX Raytacing games here.
What’s new is DLSS 2.0
The biggest changes to DLSS 2.0 over the previous version, is that it doesn’t need to be trained on a per game basis, but instead using one large, non-game specific network, meaning that it promises easier implementation and better results, allowing more devs to take advantage of the technology, which is great.
Nvidia also states that they have achieved higher Tensor core usage optimization, allowing it to run faster (apparently), while reconstructing details in the image.
With DLSS 1.0, you could only turn it on or off, like in Shadow of the Tomb Raider. However, now with DLSS 2.0, it can be set by choosing one of three quality settings. Each setting provides a different degree of upscaling to 4K (3840×2160) from the following resolutions:
- DLSS Quality (44% of 4K at 2560×1440)
- DLSS Balanced (33% of 4K at 2227×1252)
- DLSS Performance (25% of 4K at 1920×1080)
In other words, your game is actually rendered natively in the lower resolution of your choosing, from which point DLSS uses its Deep Learning AI to upscale it to 4K, gaining you significant boost in performance, as your PC only has to work on rendering the lower resolution.
Note, that in order to take advantage of DLSS 2.0, you will need to install the latest Geforce 445.75 game drivers or above, as the previous version, 442.59 will yield you DLSS 1.0 results only.
PC Benchmark Setup
With all that said, let’s see how much gain in frames per second, we can achieve with DLSS by running some benchmarks. We will then compare those benchmarks to the new DLSS 2.0 benchmarks, to see if Nvidia has delivered on their promise for better performance, compared to DLSS 1.0.
These DLSS Mechwarrior V Benchmarks were done with all graphics settings set to Ultra at 4K (3840 x 2160) resolution, using the following PC system specs:
- Graphics Card : Nvidia RTX 2070 Super ($525.44)
- CPU : AMD Ryzen 3700x ($205.00)
- RAM: Corsair CMW32GX4M2C3200C16 Vengeance RGB PRO 32GB ($114.00)
- Motherboard : Gigabyte X570 Aorus Pro Wifi ($399.99)
- Storage Drive : Western Digital 8TB Ultrastar DC HC320 ($150.00)
- Recording Drive : Samsung 970 EVO SSD 1TB M.2 NVMe ($157.99)
- System Drive : Samsung SSD 860 EVO 1TB SATA ($139.00)
- Power Supply : EVGA SuperNOVA 750 G3 ($285.00)
- Monitor :
Samsung 49″ RU7300 – 4K Smart TV (DEPRECIATED)
PC Setup Extras
- Keyboard : Corsair K70 RGB MK.2 Mechanical Gaming Keyboard (Price not available)
- Keyboard Rubber O-Rings : Rosewill Mechanical Keyboard Rubber O-Rings (Price not available)
- Mouse : Corsair M65 Elite – FPS Gaming Mouse ($42.85)
- Gaming Controller : Microsoft Xbox Elite Wireless Controller Series 2 ($150.99)
- PC Case : Phanteks Eclipse P600S (Price not available)
- PC Case Fans : Corsair LL120 RGB, 120mm RGB LED Fan – Triple Pack ($69.99)
- PC Case RGB LED Strips : Speclux Computer Magnetic Addressable RGB LED Strip Kit ($16.99)
- PC Case RGB Controller : Corsair iCUE Commander PRO ($82.90)
- CPU Cooler : Cooler Master Hyper 212 RGB Black Edition CPU Air Cooler ($99.99)
Microphone : Rode NT1-A Anniversary Condenser Microphone (DEPRECIATED)- Microphone Stand : RODE PSA 1 Swivel Mount Studio Microphone Boom Arm ($98.50)
- Audio Interface Device : Steinberg UR22C 2×2 USB 3.0 Audio Interface ($194.99)
- Headphones : Beyerdynamic DT 770 Pro 80 Limited Edition Black ($200.00)
- NVMe Expansion : ASUS Hyper M.2 X16 PCIe 3.0 X4 Expansion Card V2 – 4x NVMe M.2 ($47.99)
- HDMI Switch : KVM 4K60p HDMI USB 4 Way Switch (Price not available)
- Stream Deck : Elgato 15 Key Stream Deck (No products found.)
- Monitor RGB LED Light Strips : GIDERWEL RGBW LED Strip Light,16.4ft SMD5050 ($15.99)
- RGB LED Light Strip Controller : GIDERWEL Home Smart Zigbee RGBCCT (Philips Hue Compatible) ($21.99)
- SATA Cables : 10 Pack 16 Inch SATA III 6.0 Gbps Cable (Price not available)
- Figurine (Storm Trooper) : Revoltech Star Wars Stormtrooper 6.7″ Action Figure ($181.64)
- Figurine (Darth Vader) : Revoltech Star Wars Darth Vader 6.7″ Action Figure ($64.57)
DLSS 1.0 Benchmarks On vs OFF
First I ran DLSS 1.0 benchmarks using the Geforce 442.59 game drivers, comparing 4 variables, being DLSS OFF, DLSS Performance, DLSS Balanced and DLSS Quality to see how the FPS changes in MechWarrior 5: Mercenaries.
As MechWarrior 5 doesn’t have a built-in Benchmarks tool, I decided to simply run 3 walking tests in the hangar, which actually has a really detailed location, as well as being quite large, allowing for some distance rendering to also play a role.
DLSS 1.0 Benchmarks – View 01 (Overclocked)
In the first view, I walked from the hangar door in-between the mechs legs. I actually overclocked this view, just to show the difference in overclock between other two views, so that you guys can also see how much extra frame performance you can get from an overclock.
Using MSI Afterburner, I set my Core Clock Speed to 105mhz and Memory Clock to 900mhz, as well as Power and Temp to max.
- DLSS 1.0 Performance benchmarked at 82fps, which is a whopping 110% increase over DLSS OFF at 39fps.
- DLSS 1.0 Balanced benchmarked at 70fps, which is a 79% increase over DLSS OFF at 39fps.
- DLSS 1.0 Quality benchmarked 64fps, resulting a 64% increase over DLSS OFF at 39fps
Thus, there is a 28% increase in fps between DLSS Performance and DLSS Quality.
But more importantly, there was more than double the performance (110%) using DLSS compared to not using it. This alone is groundbreaking for anyone like me who plays 4K games and wants to hit that 60fps mark using all ultra settings, which this easily does.
DLSS 1.0 Benchmarks – View 02
I changed the angle up with the second view, walking across the hangar instead, while benchmarking at stock speeds this time.
- DLSS 1.0 Performance benchmarked at 67fps, which is a 97% increase over DLSS OFF at 34fps.
- DLSS 1.0 Balanced benchmarked at 59fps, which is a 73% increase over DLSS OFF at 34fps.
- DLSS 1.0 Quality benchmarked 53fps, resulting a 56% increase over DLSS OFF at 34fps
Thus, there is a 26% increase in fps between DLSS Performance and DLSS Quality.
One thing to immediately note here, is that it seems that the Overclock resulted in slightly bigger margins between DLSS OFF vs On, which is rather interesting. By this logic, the more power your graphics card has, the larger the percentage gains you will see between DLSS being On and OFF.
As my 2070 Super card obviously had the same Tensor Core amount during all benchmarks and only the clock speeds changed, it seems that brute force has a lot to do with this result.
DLSS 1.0 Benchmarks – View 03
The last view was the opposite of the second view, walking from the other side, but this time walking all the way from one end to the other. The results are almost identical to the previous view.
- DLSS 1.0 Performance benchmarked at 65fps, which is a 91% increase over DLSS OFF at 34fps.
- DLSS 1.0 Balanced benchmarked at 58fps, which is a 70% increase over DLSS OFF at 34fps.
- DLSS 1.0 Quality benchmarked 51fps, resulting a 50% increase over DLSS OFF at 34fps
Once again, there is a 27% increase in fps between DLSS Performance and DLSS Quality.
As we can clearly see, DLSS is a superb technology that has really come out of its teething period and blossomed into a beautiful, non-pixelated swan, showing over DOUBLE the performance when not using it.
Now let’s see how the new DLSS 2.0 compares.
DLSS 1.0 MechWarrior 5 Benchmarks Summary
Putting the above DLSS 1.0 benchmarks together into one graphs we get the following:
And for those interested, here are the performance gains we see when comparing the Overclock results to the stock, non-overclock results
- DLSS OFF (OC) benchmarked at 39fps, which is a 14% increase over DLSS OFF (Stock) at 34fps.
- DLSS 1.0 Performance (OC) benchmarked at 82fps, which is a 26% increase over DLSS 1.0 Performance (stock) at 65fps.
- DLSS 1.0 Balanced benchmarked at 70fps, which is a 20% increase over DLSS Balanced (Stock) at 58fps.
- DLSS 1.0 Quality benchmarked 64fps, resulting a 25% increase over DLSS QUality (Stock) at 51fps
DLSS 1.0 Image Quality
Besides the speed gains when using DLSS, image quality is also something Nvidia are very proud of, so let’s see if DLSS improves, not only the speed of the images, but their details as well.
For this comparison I used the yellow mech from the first view, cropping 100% into his upper body.
DLSS OFF vs DLSS Performance Image Quality
We can clearly see a significant increase in quality of the image over using even the lowest quality DLSS Performance setting, compared to not using DLSS, especially on the mech’s front armor and the background details
DLSS Performance vs DLSS Balanced Image Quality
The difference between DLSS Performance and DLSS Balanced is less noticeable, but if you look closely at the lines below the yellow trianles, you will see that DLSS Balanced offers better definition and more details here.
The dents on the mech’s front armor are also more detailed.
DLSS Balanced vs DLSS Quality Image Quality
The difference between DLSS Balanced and DLSS Quality is also very difficult to spot, but if you look at the grey metal bars (that look like a ladder) on the left side, you will see that they are slightly more defined with DLSS Quality. The
Yellow stripes to the left of this ladder are also clearer, as wells the yellow pillar in the background.
With these observations, DLSS seems to do very well at sharpening straight edges the higher you go up the quality levels.
The Mech’s right arm (our left), at the part where it bends “inner elbow” is also much better with DLSS Quality.
DLSS OFF vs DLSS Quality Image Quality
Then just for fun, let’s look at the two extremes and compare DLSS OFF vs DLSS Quality.
The results really speak for themselves here, as everything form the mech to the background, is far more detailed with DLSS Quality turned on, than having DLSS off.
DLSS 2.0 Benchmarks On vs OFF
For the DLSS 2.0 Benchmarks, I installed the Geforce 445.75 game drivers, restarted my PC and ran the same benchmark tests in MechWarrior 5: Mercenaries.
DLSS 2.0 Benchmarks – View 01 (Overclocked)
The first view is again the under leg view, using the same Overclock settings.
- DLSS 2.0 Performance benchmarked at 72fps, which is a 105% increase over DLSS OFF at 35fps.
- DLSS 2.0 Balanced benchmarked at 64fps, which is an 83% increase over DLSS OFF at 35fps.
- DLSS 2.0 Quality benchmarked 56fps, resulting a 60% increase over DLSS OFF at 35fps
Thus, there is a 28% increase in fps between DLSS Performance and DLSS Quality, which corresponds exactly to the 28% difference in DLSS 1.0.
DLSS 2.0 Benchmarks – View 02
The second view benchmark is again halfway down the hangar passage, running the RTX 2070 Super at stock, non-overclock settings.
- DLSS 2.0 Performance benchmarked at 64fps, which is an 88% increase over DLSS OFF at 34fps.
- DLSS 2.0 Balanced benchmarked at 57fps, which is an 67% increase over DLSS OFF at 34fps.
- DLSS 2.0 Quality benchmarked 52fps, resulting a 53% increase over DLSS OFF at 34fps
Thus, there is a 23% increase in fps between DLSS Performance and DLSS Quality, which is 4% less than DLSS 1.0.
DLSS 2.0 Benchmarks – View 03
The third and last view benchmark is again from the opposite side of the hangar passage, walking all the way from one side of the other, using stock non-overclock settings.
- DLSS 2.0 Performance benchmarked at 61fps, which is a 90% increase over DLSS OFF at 32fps.
- DLSS 2.0 Balanced benchmarked at 55fps, which is an 72% increase over DLSS OFF at 32fps.
- DLSS 2.0 Quality benchmarked 49fps, resulting a 53% increase over DLSS OFF at 32fps
Thus, there is a 24% increase in fps between DLSS Performance and DLSS Quality, which is 2% less than DLSS 1.0.
Not exactly the 2x enhanced performance Nvidia promised, but let’s put the two results together into another graph, to better compare the results.
DLSS 2.0 MechWarrior 5 Benchmarks Summary
Putting the above DLSS 1.0 benchmarks together into one graphs we get the following:
And for those interested, here is the performance gains we see when comparing the Overclock results to the stock, non-overclock results
- DLSS OFF (OC) benchmarked at 35fps, which is a 9% increase over DLSS OFF (Stock) at 32fps.
- DLSS 1.0 Performance (OC) benchmarked at 72fps, which is a 18% increase over DLSS 1.0 Performance (stock) at 61fps.
- DLSS 1.0 Balanced benchmarked at 64fps, which is a 16% increase over DLSS Balanced (Stock) at 55fps.
- DLSS 1.0 Quality benchmarked 59fps, resulting a 20% increase over DLSS Quality (Stock) at 49fps
Consistent benchmark results across the board.
DLSS 2.0 Image Quality
Let’s take a look at how DLSS 2.0 performs in terms of image quality.
This time I used the larger black mech, making two 100% crops – one of his right leg and some background and the other of his gun.
DLSS OFF vs DLSS Performance Image Quality
Looking at the DLSS 2.0 image detail comparisons, we can again see DLSS not only offering more fidelity, but also better highlights on the background metal.
DLSS Performance vs DLSS Balanced Image Quality
All three DLSS on setting pretty much look the same to me, with such minute differences that you will never see them in game, especially while moving.
DLSS Balanced vs DLSS Quality Image Quality
Taking in account the very little gain in quality you can see, I would recommend playing using performance for greater gains in fps,as the tiny amount of details gains for the loss in frame when using either balanced or quality don’t justify it.
DLSS OFF vs DLSS Quality Image Quality
Then again, just for fun, let’s look at the two extremes and compare DLSS OFF vs DLSS Quality.
DLSS 1.0 vs DLSS 2.0 Benchmarks Results
Ok so let’s put it all together and see if DLSS 2.0 delivers superior performance over DLSS 1.0, as Nvidia claims.
DLSS 1.0 vs DLSS 2.0 Benchmarks – View 01 (Overclocked)
Firstly, let’s test the drivers themselves, by comparing the two DLSS OFF Benchmarks. The older 442.59 drivers using DLSS 1.0 output 35fps, while the newer 445.75 only achieved 35fps, resulting in a 10% lower performance.
As DLSS isn’t a factor in this result, my conclusion can only be that the drivers are holding the results back, as I did a clean restart before each set of driver tests, with the same startup items in each test, so that there wouldn’t be any outside influence on the results.
Turning DLSS ON, in Performance, Balanced and Quality yielded similar results, whereby DLSS 2.0 actually performed significantly worse than DLSS 1.0.
The difference in loss using DLSS Performance was up to 10 frames, with a 6 frame loss using DLSS Balanced and 8 frames lower using DLSS Performance.
This is unfortunate, as Nvidia certainly aren’t delivered twice the performance here, but at the same time interesting when we look at the next view.
DLSS 1.0 vs DLSS 2.0 Benchmarks – View 02
The second view benchmarks, gave similar results, although there was no difference between the DLSS ON and OFF, as both delivered 34fps, instead of a difference of 4 frames, from the first view benchmarks.
However, looking at the three DLSS ON benchmarks, DLSS 2.0 followed the same pattern as the previous results, falling behind the DLSS 1.0.
Although, due to this not being and overclocked benchmark, the difference was far less, being only 3 frames for DLSS Performance, 2 frames for DLSS Balanced and 1 frame for DLSS Quality.
Very interesting how an overclocked effected the results between DLSS 1.0 and DLSS 2.0.
DLSS 1.0 vs DLSS 2.0 Benchmarks – View 03
In the last benchmark view, DLSS OFF results again showed DLSS 2.0 falling behind DLSS 1.0 by 2 frames (6% lower).
Much like the second view benchmark results, the difference in frames between DLSS 1.0 and DLSS 2.0 remains constant, with DLSS 1.0 taking the lead, with 4 frames lost in DLSS Performance, 3 frames lost in DLSS Balanced and 2 frames lost in DLSS Quality.
DLSS 1.0 vs DLSS 2.0 Benchmarks Conclusion
Unfortunately DLSS 2.0 isn’t any improvement over DLSS 1.0 in terms of performance, actually resulting in worse performance, although this could either be the result of one of four reasons:
- The new drivers need tweaking and optimization.
- MechWarrior 5 could be the culprit and need tweaking and optimization
- DLSS 2.0’s new unified AI network needs time to collect more data and will get better over time.
- All of the above
Either way, whether you are using DLSS 1.0 or 2.0, one thing is for sure, using DLSS offers a faster performance as well as superior imaging quality compared to not using it, making RTX graphics cards a must have for the future of gaming, whether you care about ray tracing or not.
Another interesting find was the fact that when using an overclock, the gains of DLSS were higher as well.
I hope you guys enjoyed this DLSS 2.0 vs 1.0 Benchmark test. Let me know your thoughts in the comments below.
Happy Gaming
Ozarc (¬‿¬)