UL Benchmarks Updates 3DMark with Ray-Tracing Feature Test

[btarunr]

The launch of AMD Radeon RX 6000 Series graphics cards on November 18 will end NVIDIA's monopoly on real-time raytracing. For the first time, gamers will have a choice of GPU vendors when buying a raytracing-capable graphics card. Today, we're releasing a new 3DMark feature test that measures pure raytracing performance. You can use the 3DMark DirectX Raytracing feature test to compare the performance of the dedicated raytracing hardware in the latest graphics cards from AMD and NVIDIA.

Real-time raytracing is incredibly demanding. The latest graphics cards have dedicated hardware that's optimized for raytracing operations. Despite the advances in GPU performance, the demands are still too high for a game to rely on raytracing alone. That's why games use raytracing to complement traditional rendering techniques. The 3DMark DirectX Raytracing feature test is designed to make raytracing performance the limiting factor. Instead of relying on traditional rendering, the whole scene is ray-traced and drawn in one pass.



DOWNLOAD: 3DMark v2.15.7078

The result of the test depends entirely on raytracing performance, which means you can measure and compare the performance of dedicated raytracing hardware in the latest graphics cards.

Compared with traditional rendering, raytracing can more accurately model how light interacts with the aperture of a camera.

In this feature test, camera rays are traced across the field of view with small random offsets to simulate a depth of field effect. The frame rate is determined by the time taken to trace and shade a set number of samples for each pixel, combine the results with previous samples and present the output on the screen. You can change the sample count to see how it affects performance and visual quality. The rendering resolution is 2560 × 1440.

The result of the test is the average frame rate in frames per second. You can read more about the test in the 3DMark technical guide.
Interactive mode

The 3DMark DirectX Raytracing feature test includes an interactive mode that lets you move freely around the scene and take screenshots. You can control the focus point and aperture of the camera to explore different depth of field effects using raytracing.

Test your graphics card with 3DMark
3DMark includes several tests to help you measure and compare the performance of the latest graphics cards.

• Use the 3DMark DirectX Raytracing feature test to measure the performance of dedicated raytracing hardware
• Run the 3DMark Port Royal benchmark to test graphics performance with a game-like mix of traditional DirectX 12 rendering and real-time raytracing
• Run 3DMark Time Spy Extreme to test and compare DirectX 12 performance
• Use 3DMark Stress Tests to check GPU stability and cooling performance over longer periods of heavy load

3DMark Advanced Edition
The DirectX Raytracing feature test is available now as a free update for 3DMark Advanced Edition.

3DMark Advanced Edition owners who purchased 3DMark before January 8, 2019, will need to upgrade to unlock the latest raytracing tests. The 3DMark Port Royal upgrade DLC adds Port Royal, the DirectX Raytracing feature test, and the NVIDIA DLSS feature test. Find out more about 3DMark updates and upgrades.

3DMark Professional Edition
The DirectX Raytracing feature test is available as a free update for 3DMark Professional Edition customers with a valid annual license. Customers with an older, perpetual license will need to purchase an annual license to unlock the test.

View at TechPowerUp Main Site

 

[ratirt]

Wonder how will that work with NV and AMD and if this one turns out to be a good benchmark for the Ray Tracing performance.
@btarunr Is TPU going to use it as a RayTracing performance indicator in the graphics cards testing suite?

 

[Sandbo]

Wonder how will that work with NV and AMD and if this one turns out to be a good benchmark for the Ray Tracing performance.
@btarunr Is TPU going to use it as a RayTracing performance indicator in the graphics cards testing suite?

I guess benchmarking cards for ray tracing will be a norm, and TPU has already started doing that for the newer NV GPUs.
While most games might not look different, some gams like ray-traced minecraft is completely gaming changing for me.

 

[ratirt]

I guess benchmarking cards for ray tracing will be a norm, and TPU has already started doing that for the newer NV GPUs.
While most games might not look different, some gams like ray-traced minecraft is completely gaming changing for me.

We all know that. My question was if TPU is going to use this 3Dmark as a base for measuring the RT performance. Not individual games but this benchmark in particular.
I've seen the RT games and played a bit. You are right. The RT brings something new to the game but still the support and DXR is something of a premium. Also, we need to look at the implementation in games to be slightly easier or should I say, more common than it is now.

 

[W1zzard]

@btarunr Is TPU going to use it as a RayTracing performance indicator in the graphics cards testing suite?

Unlikely, I have always preferred actual games instead of synthetic benchmarks

 

[ratirt]

Unlikely, I have always preferred actual games instead of synthetic benchmarks

Considering there's been disputes about, NV uses something else with the DXR than AMD. That games need special optimization for each system either NV or AMD. (the latest news say that some current RT games may not work on AMD) I really would like to see the performance for the RayTracing of each card (others probably as well). At least, are you going to give it a try and see how this 3Dmark RT app compares to your actually gaming suite?

 

[W1zzard]

Considering there's been disputes about, NV uses something else with the DXR than AMD. That games need special optimization for each system either NV or AMD. (the latest news say that some current RT games may not work on AMD) I really would like to see the performance for the RayTracing of each card (others probably as well). At least, are you going to give it a try and see how this 3Dmark RT app compares to your actually gaming suite?

Does it matter in the end? What matters is how well games work, and if they work better with AMD's RT implementation, for whatever reason, then isn't that why people should buy one or the other?

 

[Vya Domus]

Does it matter in the end? What matters is how well games work, and if they work better with AMD's RT implementation, for whatever reason, then isn't that why people should buy one or the other?

I guess what he wants to know is how fast each card could be when in order to detect when a game runs poorly because of the implementation and not because of the capability of the card.

 

[W1zzard]

Oh, theoretical performance. Yeah this could be interesting. For rasterization AMD has had a FLOPS advantage, and it didn't really help.

 

[ratirt]

Oh, theoretical performance. Yeah this could be interesting. For rasterization AMD has had a FLOPS advantage, and it didn't really help.

Yes, I know it is theoretical but still it does count. The games depend on the implementation and these can work better or worse. I'm not saying it is not important since games is what matters here but RayTracing is not just games you know. There is no harm in having theoretical RT performance for cards versus games individually.
Besides, if RT, sooner or later floods the market and we you will have 30 games supporting it or more, will you evaluate and test each one individually or just certain games? How many RT supporting games in a review you will be able to test maximum? Just a point here. If not all how will you pick these games?
Besides, it would be a great deal to evaluate how this RT 3Dmark stacks up with the performance against your own measurements by individual game suite of 10 for example. Will the measurement match or be totally off.

 

[Vya Domus]

Thing is RT is much more straight forward than rasterization, there isn't much that can be optimized to improve performance, it's mostly about sheer throughput. So it's going to stick out when a game performs much worse than expected.

 

[ratirt]

Thing is RT is much more straight forward than rasterization, there isn't much that can be optimized to improve performance, it's mostly about sheer throughput. So it's going to stick out when a game performs much worse than expected.

I wouldn't say that is 100% correct. You can state that rasterization is straight forward as well but just as rasterization and RayTracing is you can do these in slightly different ways which will give you same outcome. You may use different hardware (we can see that with Nvidia vs AMD both graphics but not 1 to 1 same) I think, RT is just launching and starting to matter. I'm sure there will be more apps, API, hardware solutions, plugins etc. speeding things up and improve things. God knows what's in store for us in that matter. All in all it is interesting and I'm really curious how things will develop.

 

[Jinxed]

@W1zzard Would you be also looking at quality of the RT output of games on AMD vs Nvidia graphics? There seem to be signs of lower raytracing quality in various AMD demoes, for example shadow shimmering in this one from the presentation of RX 6000 (you have to buffer it in 4K to see, look for the shadow cast by the plant on the desert ground)(go to 16:23):

Or the extremely low resolution reflections that do not seem to update in every frame for the AMD metallic/robot raytracing demo (look at the reflection of the robot in the mirrors next to the control console it's accessing)(go to: 0:48):

And also from the RX 6000 presentation, blocky raytraced shadows, that seem to be computed from a low-poly model instead of the high-poly model drawn by the rasterization pass (look at the shadow the character casts when it gets close to the light source with RT enabled)(go to 18:30):

I don't think it would be fair/valid to compare performance of two implementations, if one of them produces inferior quality. I think Steve from GamersNexus mentioned having to do something like that, but I don't remember whether that was in relation to raytracing quality or maybe the future DLSS alternative from AMD.

EDIT: I had to unlink the videos, as they were not starting at the specified timestamps.
EDIT 2: That did not help, so I just added the timestamps as text. Is there a way to turn off the youtube video widget somehow and just share a plain link?

 

[Sandbo]

@W1zzard Would you be also looking at quality of the RT output of games on AMD vs Nvidia graphics? There seem to be signs of lower raytracing quality in various AMD demoes, for example shadow shimmering in this one from the presentation of RX 6000 (you have to buffer it in 4K to see, look for the shadow cast by the plant on the desert ground)(go to 16:23):

I don't think it is possible to compare two different scenes/games to gauge the ray-tracing implementation, we probably need to wait till the 3rd party test results based on the exact same scene/game. I can be wrong as I have not really followed the tech behind the scenes, my guess is DirectX ray-tracing should be a standard that ensures the "results of ray-tracing calculation" to be the (largely) the same, with the performance being the major difference.

 

[Jinxed]

I wouldn't say that is 100% correct. You can state that rasterization is straight forward as well but just as rasterization and RayTracing is you can do these in slightly different ways which will give you same outcome. You may use different hardware (we can see that with Nvidia vs AMD both graphics but not 1 to 1 same) I think, RT is just launching and starting to matter. I'm sure there will be more apps, API, hardware solutions, plugins etc. speeding things up and improve things. God knows what's in store for us in that matter. All in all it is interesting and I'm really curious how things will develop.

I agree. And it's not just different approaches. You can use the same approach, but decrease the amount of rays cast or bounces computed. You can also use low-poly models for the BHV computation instead of high-poly ones. You may not update reflections in every frame. It may produce vastly different quality. Remember the Crytek raytracing demo?

The video is here (screenshots taken from a 4K output at 0:32 and 1:19):

I don't think it is possible to compare two different scenes/games to gauge the ray-tracing implementation, we probably need to wait till the 3rd party test results based on the exact same scene/game. I can be wrong as I have not really followed the tech behind the scenes, my guess is DirectX ray-tracing should be a standard that ensures the "results of ray-tracing calculation" to be the (largely) the same, with the performance being the major difference.

Yes, DXR is the same. But developers optimize for different GPUs. If for example the AMD cards won't have enough raytracing muscle, the devs may descrease the amount of rays cast to give you playable framerates on an AMD card. I would not be suprised if that was the case in some of the games AMD was presenting recently, given the quality issues seen in some of them.

 

[Sandbo]

Yes, DXR is the same. But developers optimize for different GPUs. If for example the AMD cards won't have enough raytracing muscle, the devs may descrease the amount of rays cast to give you playable framerates on an AMD card. I would not be suprised if that was the case in some of the games AMD was presenting recently, given the quality issues seen in some of them.

I don't really know how the implementation will be like, but then the same idea could have applied to normal rasterization/rendering and we will never be able to compare performance between the two brands of GPUs; this didn't seem to be the case. I haven't read a lot, but I seldom see people having to compare "picture quality' between AMD/Nvidia.
If the card doesn't have sufficient processing power for ray-tracing, it should be up to the users to tune down the "quality" setting, but the same setting should give the same results.
At the end, there maybe difference, but the difference should be insignificant enough that it won't be seen in most scenes.

 

[ZoneDymo]

Yes, DXR is the same. But developers optimize for different GPUs. If for example the AMD cards won't have enough raytracing muscle, the devs may descrease the amount of rays cast to give you playable framerates on an AMD card. I would not be suprised if that was the case in some of the games AMD was presenting recently, given the quality issues seen in some of them.

That makes little sense, that is what graphics settings are for.
But instead of pointless speculation, lets wait for the cards to actually be released, im sure plenty of channels (Gamers Nexus, Global Foundry's) will find it interesting to compare every facet of ray tracing between the two brands (and eventually Intel if that ever comes to pass).

Should watch Global Foundry's recent vid on Watchdogs Legion, pretty deep dive in the RT it has to offer.

 

[Jinxed]

That makes little sense, that is what graphics settings are for.
But instead of pointless speculation, lets wait for the cards to actually be released, im sure plenty of channels (Gamers Nexus, Global Foundry's) will find it interesting to compare every facet of ray tracing between the two brands (and eventually Intel if that ever comes to pass).

Should watch Global Foundry's recent vid on Watchdogs Legion, pretty deep dive in the RT it has to offer.

There is a difference between graphics settings and actual visible graphics glitches due to shortcuts in the implementation. And there should be no issue in checking this in at least the first few graphics releases. It's kinda the point of independent reviews, to verify that the companies are not feeding us just with marketing crap.

 

[Punkenjoy]

The Ray Tracing implementation is not a fixed function but compute units that do a specific types of computations.

The rendering will be the same on AMD and Nvidia. Game Devs won't implement AMD or Nvidia ray tracing but Vulcan or DXR ray tracing. Both cards will perform better or worst depending on what the developers do their implementation. I suspect that they will have in the settings different option to set different Ray Tracing level. Shitty devs will do shitty implementation where some others devs will have a very polish implementation.

That is already the case right now with Offline renderer. Anyone who did some blender or others software know that it's easy to do a render, it's hard to make things look good.

As for how AMD and Nvidia will really perform in the end in future games, we will have to wait for benchmark.

 

[bug]

Thing is RT is much more straight forward than rasterization, there isn't much that can be optimized to improve performance, it's mostly about sheer throughput. So it's going to stick out when a game performs much worse than expected.

This is going to be complicated by games not actually doing RTRT, but rasterization+RTRT hybrid. But having a dedicated RTRT benchmark can't hurt solving that equation (baring the fact that 3DMark frequently uses implementations that are questioned by one side or the other).

 

[DrCR]

Unlikely, I have always preferred actual games instead of synthetic benchmarks

And that's appreciated and doesn't go unnoticed.