Quake 2 RTX v1.4 adds Vulkan / potential AMD GPU support, improved performance, and more.

Deleted member 79451 · Dec 15, 2020

This update has provided a large performance increase on my 3080 when using the Vulkan extensions instead of Nvidia's and/or using the new TAAU (probably mainly the latter?). Latest gen AMD GPU users should be able to play it now, although the Steam overlay may be causing them problems?

Hey, everyone! Today we're releasing v1.4.0, featuring added support for final Vulkan Ray Tracing API, and enabling dynamic selection between the pre-existing NVIDIA VKRay and the new Khronos extension backends.

New Features:

Added support for final Vulkan Ray Tracing API. The game can now run on any GPU supporting `VK_KHR_ray_tracing_pipeline` extension

Added temporal upscaling, or TAAU, for improved image quality at lower resolution scales.

Fixed Issues:

Fixed a crash that happened when there are no available sound devices.

Fixed a few issues with the tone mapper and the profiler for AMD GPU compatibility.

Fixed a server crash: https://github.com/NVIDIA/Q2RTX/issues/86

Fixed black materials and some light leaks: https://github.com/NVIDIA/Q2RTX/issues/55

Fixed missing railgun lights in photo mode: https://github.com/NVIDIA/Q2RTX/issues/75

Fixed missing sun light on geometry with invalid clusters.

Fixed the CFLAGS for MinSizeRel and RelWithDebInfo builds to generate correct debug symbols.

Fixed the game stuttering on Linux: https://github.com/NVIDIA/Q2RTX/issues/62

Fixed the issue with all models being missing or corrupted on some maps during network play.

Fixed the nearest filter when DRS was enabled and then disabled.

Fixed building the game with GCC10 on Linux: https://github.com/NVIDIA/Q2RTX/issues/80

Denoiser Improvements:

Implemented a new gradient estimation algorithm that makes the image more stable in reflections and refractions.

Implemented sampling across checkerboard fields in the temporal filter to reduce blurring.

Improved motion vectors for multiple refraction, in particular when thick glass is enabled.

Improved the temporal filter to avoid smearing on surfaces that appear at small glancing angles, e.g. on the floor when going up the stairs.

Improved the temporal filter to make lighting more stable on high-detail surfaces.

Misc Improvements:

Added git branch name to the game version info.

Improved the console log to get more information in case of game crashes.

Increased precision of printed FPS when running timedemos.

Reduced the amount of stutter that happened when new geometry is loaded, like on weapon pickup.

Replaced the Vulkan headers stored in the repository with a submodule pointing to https://github.com/KhronosGroup/Vulkan-Headers

Static resolution scale can now be set to as low as 25%.

Vulkan validation layer can now be enabled through the `vk_validation` cvar.

Updated SDL2 version to changeset 13784.

Source

Gitaroo · Dec 15, 2020

No dlss support is disappointing

ILikeFeet · Dec 15, 2020

surprised they didn't add DLSS to this game. thought it would have been one of the first to receive it

pksu · Dec 15, 2020

The problem with DLSS is that Nvidia is using open source Quake release licensed under GPL and DLSS implementation is closed source. One would think it would be possible to write some sort of dynamic module loader capable of loading and using a closed-source DLSS library but that would be also against GPL too as long as the components share "complex data structures" IIRC.

mordecaii83 · Dec 15, 2020

I'm looking forward to seeing benchmarks for Nvidia and AMD cards with this new update!

ILikeFeet · Dec 15, 2020

~45fps at 1440p with no resolution scaling

Crowboy · Dec 15, 2020

My Geforce Experience just notified me that there is a new driver update for this too.

FliX · Dec 15, 2020

ILikeFeet said:
~45fps at 1440p with no resolution scaling

It never fails to impress me.

Dictator · Dec 15, 2020

Yeah the taau and all the denoising improvement are great - Looks excellent im this New patch!

maximumzero · Dec 15, 2020

Shame there isn't some way to enable software ray-tracing, even if it's at a poor framerate. I'd like to see what the hubbub is about.

neoak · Dec 15, 2020

maximumzero said:
Shame there isn't some way to enable software ray-tracing, even if it's at a poor framerate. I'd like to see what the hubbub is about.

Open a new PowerPoint and paste screencaps of the videos. Then present the PowerPoint in full screen, navigate the slides.

There, Full software experience.

ILikeFeet · Dec 15, 2020

maximumzero said:
Shame there isn't some way to enable software ray-tracing, even if it's at a poor framerate. I'd like to see what the hubbub is about.

AMD should have still allowed a RT renderer for their 5000 series like Nvidia has

Eblo · Dec 15, 2020

maximumzero said:
Shame there isn't some way to enable software ray-tracing, even if it's at a poor framerate. I'd like to see what the hubbub is about.

You can play Quake II RTX on GTX 1000 series cards at least. I think those do it via software? I tried on my 1070. All minimum settings netted me about 15 FPS lol

Tora · Dec 15, 2020

No idea how it used to run, but at 1440p I'm getting 140fps with a 3080

copelandmaster · Dec 15, 2020

Does the new Vulkan support in 460.89 change behavior/performance for pascal cards, or does the 1.4 update to the game do that on its own? The nvidia subreddit says there is a flickering problem on 1080 Ti's with this new driver. I'm still on 455.71, and am considering rolling back to 446.14 to remove the SteamVR stutter issue.

EDIT: Just rolled back and now I can pick 10 bit color at 4k in my control panel again. Hmm.

BAW · Dec 15, 2020

I don't get it. Why Quake 2? Why not other classic FPSes?

orava · Dec 15, 2020

BAW said:
I don't get it. Why Quake 2? Why not other classic FPSes?

Quake 2 is fully open sourced and easily available. Not many other older games are.

Deleted member 79451 · Dec 15, 2020

BAW said:
I don't get it. Why Quake 2? Why not other classic FPSes?

Mainly because a few years back someone wrote a software raytraced solution for Quake 2, then someone wrote a CUDA accelerated solution based on the first. Then Nvidia built this upon those previous works.

Why the first guy did it? I think the environments are well set up for real-world lightning. There are actual physical light sources (that is, 3D meshes that textures for lights were based upon) for the environments. That's something the original Quake only has some of (there's a lot of fake lights without any artistic sources placed frequently in Quake 1).

Without that critical piece a developer would have to go in and do ART (decide the lighting) to have raytracing work.

That said, someone is working on Quake 1 path tracing right now: https://twitter.com/lh0xfb

brain_stew · Dec 15, 2020

BAW said:
I don't get it. Why Quake 2? Why not other classic FPSes?

It's open source.

scitek · Dec 15, 2020

BAW said:
I don't get it. Why Quake 2? Why not other classic FPSes?

Aren't they working on other games, too?

ILikeFeet · Dec 15, 2020

AMD RX 6800 XT runs at almost half the frame rate of Nvidia's RTX 3080 in Vulkan Ray Tracing tests

Things can only get better, right?

www.pcgamer.com

Gitaroo · Dec 15, 2020

ILikeFeet said:
AMD RX 6800 XT runs at almost half the frame rate of Nvidia's RTX 3080 in Vulkan Ray Tracing tests

Things can only get better, right?

www.pcgamer.com

So about Turing performance? Honestly not bad for first gen though.

ILikeFeet · Dec 15, 2020

Gitaroo said:
So about Turing performance? Honestly not bad for first gen though.

from the initial release of the renderer

Quake 2 RTX Is Out And What Kind Of Performance You Can Expect

Quake 2 may be a 22-year-old game and run on just about anything with an electrical current, but that's not what Quake 2 RTX is about. What started out as a mod to Quake 2, by Christoph Schied, known as Q2VKPT (Quake 2 Vulkan Path Tracing) is now a full standalone release thanks to the team at...

wccftech.com

given the price, I wouldn't really call that "not bad for first gen". it just shows a possible flaw in AMD's design. there's probably benefits to their design, but it's currently not showing in RT games. AMD's probably banking on games not using RT much if at all

Akronis · Dec 15, 2020

Gitaroo said:
So about Turing performance? Honestly not bad for first gen though.

it's honestly pretty terrible if they can only match last gen's RT solution for almost the same price as NVIDIA's current gen

Deleted member 79451 · Dec 15, 2020

Akronis said:
it's honestly pretty terrible if they can only match last gen's RT solution for almost the same price as NVIDIA's current gen

Not to pile on AMD but they're also on a legit 7nm node while Nvidia was on 12nm with Turing.

jwhit28 · Dec 15, 2020

Eblo said:
You can play Quake II RTX on GTX 1000 series cards at least. I think those do it via software? I tried on my 1070. All minimum settings netted me about 15 FPS lol

15fps is how I experienced the original game anyway.

SapientWolf · Dec 15, 2020

Also has native Vulkan multi GPU support for some reason.

Looking at the charts, I don't think balls out RT is going to be a thing on the consoles this gen. RDNA2 just doesn't seem cut out for it. That could end up creating a pretty wide gulf between PC and console lighting quality eventually.

The 3080 doesn't run out of poop until 4k but I think the smart money is on waiting one more GPU gen if you're all on board the RT train. AMD's RT performance should be good and Nvidia's should be great.

Akronis · Dec 15, 2020

Jigen said:
Not to pile on AMD but they're also on a legit 7nm node while Nvidia was on 12nm with Turing.

I'm not even that matters in this case, they just don't have the fully dedicated hardware on the board for RT. They tried to find a cheaper alternative to NVIDIA's ~~tensor~~ RT cores by just straight up not having them.

Despite paying that early adopter fee with the 2080 Ti, I feel like NVIDIA's approach with dedicated ~~tensor~~ RT cores was absolutely the right move.

Veliladon · Dec 15, 2020

ILikeFeet said:
given the price, I wouldn't really call that "not bad for first gen". it just shows a possible flaw in AMD's design. there's probably benefits to their design, but it's currently not showing in RT games. AMD's probably banking on games not using RT much if at all

Possible flaw? They're doing BVH on the shader cores instead of in fixed function hardware like Nvidia. Which is basically where the problem is. Turn on RTX on an Nvidia card and you're limited to the throughput of the RT cores. Turning on ray tracing on an AMD card not only limits performance to whatever TFLOPs the ray accelerators can put out but it also takes FP32 throughput away from everywhere else at the same time.

Honestly, we should just be thankful that RTG actually made a product that matches Nvidia in raster performance for once. After all, RTX only came out *checks notes* two years ago.

ILikeFeet · Dec 15, 2020

Veliladon said:
Possible flaw? They're doing BVH on the shader cores instead of in fixed function hardware like Nvidia. Which is basically where the problem is. Turn on RTX on an Nvidia card and you're limited to the throughput of the RT cores. Turning on ray tracing on an AMD card not only limits performance to whatever TFLOPs the ray accelerators can put out but it also takes FP32 throughput away from everywhere else at the same time.

Honestly, we should just be thankful that RTG actually made a product that matches Nvidia in raster performance for once. After all, RTX only came out *checks notes* two years ago.

"flaw" is the wrong word to use, rather it might have been a bad call, in hindsight. it's not like AMD was in the dark when RT was brought up when Microsoft was laying the groundwork for DX12. they probably thought it just wasn't gonna take off beyond single effects like shadows or AO (and they might still be right) at 30fps. maybe it's a case of just not having Nvidia's R&D budget given where AMD was prior to Navi and Zen.

here's a comparison of the NV extension and the standard

brain_stew · Dec 16, 2020

ILikeFeet said:
AMD RX 6800 XT runs at almost half the frame rate of Nvidia's RTX 3080 in Vulkan Ray Tracing tests

Things can only get better, right?

www.pcgamer.com

That's actually better than I thought from RDNA2 in a path traced title. Still utterly miserable performance considering the 6800XT is more expensive than a 3080 currently but better than I thought it would be based on Minecraft and Control.

GameAddict411 · Dec 16, 2020

brain_stew said:
That's actually better than I thought from RDNA2 in a path traced title. Still utterly miserable performance considering the 6800XT is more expensive than a 3080 currently but better than I thought it would be based on Minecraft and Control.

Dude, that's noticably worse then a 2080 ti. If dlss gets added to this game, it would be amazing.

BAW · Dec 16, 2020

Jigen said:
Mainly because a few years back someone wrote a software raytraced solution for Quake 2, then someone wrote a CUDA accelerated solution based on the first. Then Nvidia built this upon those previous works.

Why the first guy did it? I think the environments are well set up for real-world lightning. There are actual physical light sources (that is, 3D meshes that textures for lights were based upon) for the environments. That's something the original Quake only has some of (there's a lot of fake lights without any artistic sources placed frequently in Quake 1).

Without that critical piece a developer would have to go in and do ART (decide the lighting) to have raytracing work.

That said, someone is working on Quake 1 path tracing right now: https://twitter.com/lh0xfb

Wow, great explanation, thanks!
It's kind of impressive of how they keep working on it and improving it instead of treating it like a one-off proof of concept.
I can confirm it both looks better and runs faster on my 3080. Previously 1080p upscaled to 2160p was blur city, not anymore. I don't know what temporal upscaling is but it works.

Deleted member 30077 · Dec 16, 2020

Gitaroo said:
So about Turing performance? Honestly not bad for first gen though.

It's not even close to Turing here. There isn't a big gap in performance for raytracing in games between Ampere and Turing. Ampere is only slightly better in raytracing than Turing in most games. The more raytracing the better it becomes but not by a huge amount.

For example the 3070 is only 7% faster at 1440p than the 2080Ti, which is its raster equivalent Turing card, in Quake 2.

dgrdsv · Dec 16, 2020

Akronis said:
I'm not even that matters in this case, they just don't have the fully dedicated hardware on the board for RT. They tried to find a cheaper alternative to NVIDIA's tensor cores by just straight up not having them.

Tensor cores have nothing to do with ray tracing.

Veliladon said:
They're doing BVH on the shader cores instead of in fixed function hardware like Nvidia.

BVH is built on CPU and shader cores in both cases (at least I think so, it's possible that AMD is building it solely on CPU).
BVH testing is done on fixed function units in both cases.
Intersection evaluation is done on dedicated MIMD processors by NV and on shader cores by AMD.
The latter is the most likely reason for a comparative performance deficit. But it's also possible that other parts of the pipeline are running faster on NV h/w.

Gitaroo said:
So about Turing performance? Honestly not bad for first gen though.

Turing launch: "Wtf is this shit, how can RT be so slow, why is it so much slower than rasterization?"
RDNA2 launch two years later: "So it's about Turing performance with RT, not bad, despite it being twice slower than Turing!"

Bluelote · Dec 16, 2020

SapientWolf said:
Also has native Vulkan multi GPU support for some reason.

Looking at the charts, I don't think balls out RT is going to be a thing on the consoles this gen. RDNA2 just doesn't seem cut out for it. That could end up creating a pretty wide gulf between PC and console lighting quality eventually.

The 3080 doesn't run out of poop until 4k but I think the smart money is on waiting one more GPU gen if you're all on board the RT train. AMD's RT performance should be good and Nvidia's should be great.

check the digital foundry video on the watchdogs ray tracing on consoles, there are many ways to cut back the effects while still retaining a lot of it,
also Quake 2 rtx is heavily optimized for Nvidia, not really Optimized for AMD I would think,
no doubt Nvidia is ahead in this approach they pioneered not long ago, but it surely can be explored on consoles, and there is always 30FPS.

6800xt is clearly delivering some playable performance on this, despite not being great at this sort of thing.

dodo021 · Dec 16, 2020

Does AMD use a denoiser for the RT part like Nvidia or AMD calculate RT at native resolution without denoising ?

Pottuvoi · Dec 16, 2020

I wonder how BVH and actual tracing part would change if developer would build this from ground up for AMD hardware.

dodo021 said:
Does AMD use a denoiser for the RT part like Nvidia or AMD calculate RT at native resolution without denoising ?

Both use same denoiser using compute.
Denoising really doesn't have anything to do with resolution.

Pipyakas · Dec 16, 2020

AMD not having a software RT fallback in their driver on both Windows and Linux is killing me here. I want to murder my 580 just to experiment with RT god damn it

On another note, seems like 6800XT is getting roughly 50% performance of the 3080 in a fully path traced game, ouch

brain_stew · Dec 16, 2020

GameAddict411 said:
Dude, that's noticably worse then a 2080 ti. If dlss gets added to this game, it would be amazing.

The 6800XT is actually slower than a 2070 there. A 2080Ti isn't even a ballpark comparison.

I've seen a lot of talk about "1st generation" RT from AMD, but it is far slower than Nvidia's 1st generation RT solution. It's just slow due to AMD accelerating less of the RT pipeline in hardware than Nvidia. This can't be hand waived away at just 1st generation hardware.

dgrdsv · Dec 16, 2020

Pottuvoi said:
I wonder how BVH and actual tracing part would change if developer would build this from ground up for AMD hardware.

Would probably change back into Quake 2 running under OpenGL without RT.
Unless someone would figure out how to fit the whole BVH into 128MB IC and how to do full on path tracing with zero ray incoherence.

ILikeFeet · Dec 16, 2020

from Beyond3D

Q2 RTX, Denoising takes 4.23ms on my RTX 2060, 19.8% of a frame time.

wonder if denoising takes up a longer amount of time on AMD

Akronis · Dec 16, 2020

dgrdsv said:
Tensor cores have nothing to do with ray tracing.

Are they just for accelerating ML loads for things like DLSS then? I had thought they benefited RT in a more direct way beyond that.

EDIT: I'm dumb, I was combining the RT cores and tensor cores together in my brain

7thFloor · Dec 16, 2020

scitek said:
Aren't they working on other games, too?

supposedly

Veliladon · Dec 16, 2020

dgrdsv said:
BVH testing is done on fixed function units in both cases.
Intersection evaluation is done on dedicated MIMD processors by NV and on shader cores by AMD.

You sure? This was AMD's presentation:

Building the BVH set would have to happen on the CPU, yes but traversal of the BVH is handled with shader code. Ray accelerator does intersections.

SapientWolf · Dec 16, 2020

Bluelote said:
check the digital foundry video on the watchdogs ray tracing on consoles, there are many ways to cut back the effects while still retaining a lot of it,
also Quake 2 rtx is heavily optimized for Nvidia, not really Optimized for AMD I would think,
no doubt Nvidia is ahead in this approach they pioneered not long ago, but it surely can be explored on consoles, and there is always 30FPS.

6800xt is clearly delivering some playable performance on this, despite not being great at this sort of thing.

I was thinking more along the lines of using RT for shadows, reflections, global illumination, and ambient occlusion at the same time, which is a basically a no go on PC without DLSS right now. I predict that most console games will pick and choose a few effects rather than implement it across the board. I personally think AO and reflections are where you get the most bang for your buck.

Spoit · Dec 16, 2020

napata said:
It's not even close to Turing here. There isn't a big gap in performance for raytracing in games between Ampere and Turing. Ampere is only slightly better in raytracing than Turing in most games. The more raytracing the better it becomes but not by a huge amount.

For example the 3070 is only 7% faster at 1440p than the 2080Ti, which is its raster equivalent Turing card, in Quake 2.

This, along with minecraft and control, is one of the handful of games where the performance is not raster limited (and the AMD cards actually have competititive raster performance, so that arguably makes the difference even worse)

dgrdsv · Dec 16, 2020

Akronis said:
Are they just for accelerating ML loads for things like DLSS then?

Yeah. They also handle FP16 shading on Turing and Ampere.

Akronis said:
I had thought they benefited RT in a more direct way beyond that.

There was some (misleading?) talk about using TCs for RT denoising but that hasn't really happened as no game with RT released at the moment are using them for that.

Veliladon said:
Building the BVH set would have to happen on the CPU, yes but traversal of the BVH is handled with shader code. Ray accelerator does intersections.

Correct, although if AMD builds BVH completely on CPU it's another weak point of their RT approach - NV is building/updating BVH's BLAS on GPU, only TLAS are built on CPU.
Intersection testing is "BVH testing", same thing. Both NV and AMD are doing this on FF units.
Hit (intersection) evaluation is done on shading SIMD by AMD and dedicated MIMD inside RT cores by NV.

Quake 2 RTX v1.4 adds Vulkan / potential AMD GPU support, improved performance, and more.

DF Deet Master

DF Deet Master

Master of the Reality Stone

Digital Foundry

DF Deet Master

The Enlightened Wise Ones

Alt Account

DF Deet Master

DF Deet Master

Prophet of Regret - Lizard Daddy

Prophet of Regret - Lizard Daddy

DF Deet Master

DF Deet Master

Prophet of Regret - Lizard Daddy