RX6800XT vs RTX3080 - Real-time Ray Tracing performance impacts compared in two AMD Sponsored titles

[wolf]

As the title implies, I'm aiming to measure the impact of enabling RTRT so we can get some insights into the performance cost on competing architectures. I've chosen Riftbreaker and RE Village as they're newer titles, and AMD-sponsored ones.

There was and is still some hoo-haa and reluctance around comparing the cards in Nvidia sponsored titles, and I can at least in part agree why, So I wanted to give AMD the chance to put their best foot forward here, given the games are optimized and effects (amount of effects, ray counts etc) are chosen and geared for AMD hardware capability.

These two games/sites where I got the figures from have testing methods that appear to give highly repeatable and accurate results, by way of built-in benchmark or highly scripted story events/runs.

Riftbreaker 1440P Ultra - results taken from Wccftech video

	Nn RT avg / 1% low	RT Ultra avg / 1% low	RT Ultra + RT AO avg / 1% low	RT Ultra perfomance as a % of rast fps	RT Ultra + RT AO perfomance as a % of rast fps	% faster at rendering RT Ultra effect	% faster at rendering RT Ultra + RT AO effects
RX6800XT fps	400 / 340	160 / 116	132 / 103	34-40%	30-33%
RX6800XT frametime ms	2.5 / 2.9 ms	6.3 / 8.6 ms	7.6 / 9.7 ms
RTX3080 fps	334 / 287	196 / 154	165 / 133	53-59%	46-49%
RTX3080 frametime ms	3 / 3.5 ms	5.1 / 6.5 ms	6.1 / 7.5 ms			80-90%	58-70%

Resident Evil Village 1440P max from Techpowerup! performance review

	Nn RT avg	RT on avg	RT on perfomance as a % of rast fps	% faster at rendering RT effect
RX6800XT fps	202.2 fps	95.5 fps	47%
RX6800XT frametime ms	4.9 ms	10.5ms
RTX3080 fps	175 fps	108.4 fps	62%
RTX3080 frametime ms	5.7 ms	9.2ms		60%

As you can see from the data, before we enable RT settings, the 6800XT is able to comfortably surpass the RTX3080 by 15-20% @ 1440p across both tested games. This isn't unexpected being AMD-sponsored titles, where the RDNA2 parts are generally able to outpace their normal Ampere competitor. And it must be said well done to AMD at this point, those are some serious optimizations that allow RDNA2 cards to effectively compete 1-2 tiers higher than their usual competitor.

It all turns around when we enable RT, however, and there are different ways to size that up.

In RE Village we see that the RTX3080 retains 62% of the rasterization only average FPS, where the 6800XT can only retain 47%, but the difference between those two numbers relatively speaking doesn't tell the entire story.

Lets look at FPS measured in frame times, the RTX3080 incurs an average 3.5ms render time penalty to render the RT effects, but the 6800XT incurs a 5.6ms render time penalty, leading us to believe that the 3080 can render the RT effects ~60% faster in this scenario.

The story gets even more interesting in Riftbreaker.

we see that the RTX3080 retains 53-59% of the rasterization only average FPS, where the 6800XT can only retain 34-40%. Again looking at frame times, the RTX3080 incurs a 2.1 - 3 ms render time penalty to render the RT effects, but the 6800XT incurs a 3.8 - 5.7 ms render time penalty, leading us to believe that the 3080 can render the RT effects ~80-90% faster than the 6800XT in this scenario.

But it doesn't stop there, when we add yet more RT load into the pipeline, the numbers shift and the gap narrows.

the RTX3080 incurs a 3.1 - 4 ms render time penalty to render both RT effects on top of rasterization only gameplay, but the 6800XT incurs a 4.9 - 6.8 ms penalty, leading us to believe that the 3080 can render both RT effects ~58-70% faster in this scenario.

What are some of the conclusions we can draw?

• For at least an RTX3080 vs 6800XT scenario, in these games, the 3080 appears to handle the additional RT workload considerably faster, in the order of 58-90% depending on the game and settings. The obvious caveat being the small sample size of games, however, my conclusions appear to closely match those found by DF when they compared these cards head to head seeking the same sort of answers and testing different games.
• While the Absolute performance with RT on can look very competitive for the 6800XT relative to the 3080 (in these titles and perhaps others too), that figure does not tell the story as well as showing how many extra milliseconds to your frame time RT can add, and how much rasterization only performance is lost when enabling RT.
• When increasing the number of effects applied, the lead in RT computation that the 3080 has starts to diminish. To what end? hard to tell, perhaps given enough RT load the 6800XT would eventually overtake the 3080, but both would likely be a slideshow at that point.

I'd like to hear any other conclusions/theories people might be willing to draw with this data or other similar comparisons, and if my testing is flawed or omits valuable data or points of interest.

 

[oxrufiioxo]

I think it's way too early to compare and at least to me both implementations seem drastically different. The Achilles heel of the 3080 and something that might become more noticeable as time goes by is its 10GB of vram causing issues with RT at 4k and possible 1440p. RT seems to be very vram heavy depending on how many effects are used. Only time will tell though.... My 3080 ti will be relegated to 1080p gaming whenever there is a gpu at least 40% faster to replace it with.


Right now Nvidia RT implementation is much stronger imo but AMDs is impressive enough in games coded for it for a 1st iteration.

 

[ZoneDymo]

its all pointless, if you really want RT you have to wait two or more generations honestly, all the DLSS mitigation just to deal with how hard it is on the gpu.
Saw a 3090 video of RT on in The Ascent which is not a particulairy intense looking game and the RT seemingly only does reflections and yet it cuts the fps by a like 2/3's.... 140 to about 50 ish, and that is just the reality of it.

 

[Audioave10]

Ray Tracing is not for me. I don't need 4K and an extra $400 - $1000 for a video card to get a tiny bit better lighting and shadows. I wonder if the lack
of high Frame Rates for games these days is because of poor on-line coding, bloatware, telemetry and an online store for buying the rest of the game.
I also see modern games as being cartoons and poor console ports. Many are Zombies and Battle Royale games so they don't even write AI anymore.
Sorry, I'm grumpy today.

 

[DemonicRyzen666]

Wolf,
My only suggestion would be to add the average cpu load for this

as I believe the cpu load will be higher on the RTX 3080 as it does not have a hardware scheduler like the 6800 XT does
Since the 6800 Xt does have a hardware scheduler you're limited to it's clock speed domain of what it's set to.
Where in the RTX series you limited by what the clock speed of the cpu is able to push out to the gpu.

4.6ghz-5.0ghz cpu scheduler vs 2.3ghz-2.7ghz GPU scheduler

I do realize that this is a 1st gen RT vs second gen RT too.

 

[wolf]

The intention here was not 'is RT even worth turning on in games' type of a question, no matter if you have RTX 20/30 or RDNA2, the answer to that will be extremely subjective to what you like in games. My intention was seeing what RDNA2 can do VS Ampere when AMD is heavily involved in development and optimisation, what can be achieved.

So, we're just taking a look at when AMD does their own first party games with RT, How does it perform and does it favour AMD hardware? I've been led to believe that RDNA2 can indeed match or exceed Ampere at RTRT,that there is no compelling evidence that RDNA2 is slower, in particular there is one suite of synthetic benchmarks on github people refer to. I think that if AMD could replicate those wins (pure RT benchmark, seemingly not really represtative of what we get in a shipping game) in games, that it would be AMD sponsored titles we'd be seeing it by now, or at least some signs it was possible, and so far at least (2 very recent AMD sponsored games), that doesn't seem to be the case. It still would seem that the hardware capability, when mated with a gaming scenario that involves RT, Ampere still has a comfortable RT lead.

 

[Audioave10]

wolf...Don't get me wrong, I do think this work has interesting results and is informative. The RT cost from Nvidia disturbs me however.

 

[wolf]

wolf...Don't get me wrong, I do think this work has interesting results and is informative. The RT cost from Nvidia disturbs me however.

Cost of purchasing the hardware, or 'cost' in terms of framerate hit to enable it?

 

[londiste]

This Digital Foundry piece is still the best source with details about what and why in RT performance:

tl;dw
- There is a fixed cost to enable RT.
- RDNA2 currently has less RT horsepower compared to comparable Ampere card, which leads to three scenarios:
1. As long as horsepower available is enough on both they get similar results.
2. When game demands more RT horsepower than RDNA2 has available but less than what Ampere has available, Ampere tends to pull ahead.
3. When game demands more than either has available, things even out again.

 

[Chomiq]

2nd gen RT implementation vs 1st gen RT implementation. No surprises.

 

[Audioave10]

Cost of purchasing the hardware, or 'cost' in terms of framerate hit to enable it?

When the RTX cards began, the Price & Performance "both" were terrible. They also have increased the power draw and HEAT more
than I've ever seen before. As you know, heat is the No.1 enemy.

 

[wolf]

When the RTX cards began, the Price & Performance "both" were terrible. They also have increased the power draw and HEAT more
than I've ever seen before. As you know, heat is the No.1 enemy.

Indeed, Turing was the first generation I skipped in a loooong time. I did get truly lucky with a launch day MSRP 3080 however, and running it undervolted and deshrouded it never really gets hot, but it of course still has 250-270w of heat to shed.

 

[wolf]

Honestly expected more discussion on this one, I could swear there are users here who insist that RDNA2 has no RT deficit against Ampere, Theres some github link of synthetic benchmarks around that seems to get referenced when it comes up. Be that as it may for Purely RT loads, in games that support RT so the GPU has to process other, non-RT elements too, RDNA2 seems to choke a lot sooner, even in AMD sponsored titles where if they could leverage that capability, surely that's where we'd see it.

 

[oxrufiioxo]

Honestly expected more discussion on this one, I could swear there are users here who insist that RDNA2 has no RT deficit against Ampere.

I just think there is not enough software especially software where you definitely should run raytracing vs leaving it off.

Until we get some AMD sponsored titles with as many effects as something like Cyberpunk/Control It's kinda hard to compare. RE8 is the only big game that seems to perform close enough from both vendors atm. I think the more interesting battle will be RDNA3 vs Nvidia 4000 series personally not enough people actually own Ampere/RDNA2 to probably care and whoever is buying either likely had no choice and just went with what they could get in stock/cheapest.

I think the more interesting comparison a couple years from now is how Nvidia 8/10/12GB cards are fairing in modern titles of that year vs RDNA2 cards with their 16GB of vram on the top 3 models. RT to me is still in it's infancy and most just don't care because they don't want to take a 40-60% hit to performance for better reflections/lighting/shadows.

 

[wolf]

I think the more interesting battle will be RDNA3 vs Nvidia 4000 series

That will definitely be one to watch, looks like AMD will make significant strides with RDNA3 in general, so I wonder if RT performance will get any priority over just supporting it like RDNA2. Likewise, I wonder if Nvidia can maintain their current lead or what their plan even is, given the current console generation being RDNA2 based, so in theory with any PC port with RT they will already comfortably be able to match or exceed RT performance with Ampere's RT capability, let alone if they down again. AMD certainly has them on notice with such a massive catch-up in rasterization, they'll need to push ahead in all manners to not get left behind.

 

[pavle]

As the title implies...

The situation is similar than with DLSS vs. FSR (Am I in hot water here? ); Nvidia does it on dedicated hardware, with much more processing than AMD while being only around 11% faster/better. AMD does it on general purpose processors/shaders and that's quite impressive.

 

[wolf]

I think the more interesting comparison a couple years from now is how Nvidia 8/10/12GB cards are fairing in modern titles of that year vs RDNA2 cards with their 16GB of vram on the top 3 models. RT to me is still in it's infancy and most just don't care because they don't want to take a 40-60% hit to performance for better reflections/lighting/shadows.

That will also be interesting, no doubt. I'm not really too fussed tbh, given generally speaking you can lower textures and free up gigabytes of VRAM for little-to-no visual difference. I've found RT in the right games utterly transformative, where turning textures down from Ultra to high, you be hard-pressed to tell, but that's personal and of course one could argue RT on can be hard to tell too.

Of course in a pure apples to apples test, if say the game demanded more than 8gb, I'd expect a 3070 to lose to a 6800 (lets say, by a bigger margin than expected), but if you owned a 3070, you'd just turn textures down till you were below 8gb and likely get on with your life without thinking about how you should have bought the unobtanium 6800 at the time.

Like you eluded too, these differences at a point are fairly academic, and people are just happy that they got what they got at all.

 

[oxrufiioxo]

Like you eluded too, these differences at a point are fairly academic, and people are just happy that they got what they got at all.

For sure, anyone who was able to grab a 3060ti-3090 or 6700XT-6900XT for non scalper pricing shouldn't think twice and be happy they have anything at all.

This stuff is all fun to compare as long as it doesn't turn into a fanboy flame war. One of my favorite things to watch on YouTube is how older hardware stacks up from sources I trust.

I was very tempted to pick up a 6900XT this round but for the model I wanted my 3080 ti was 468 usd cheaper pre tax so that was the deciding factor....

Raytracing was probably the 3-4 down the list of importance for me. I do like DLSS 2.0+ though much more so at 1440p than FSR.

I'm hopeful Witcher 3 with raytracing is pretty decent that games base lighting and reflections suck.

 

[wolf]

For sure, anyone who was able to grab a 3060ti-3090 or 6700XT-6900XT for non scalper pricing shouldn't think twice and be happy they have anything at all.

It's sad to have to think that way, but it's largely true, if you got any of those at, or even within spitting distance of MSRP, you're laughing.

This stuff is all fun to compare as long as it doesn't turn into a fanboy flame war. One of my favorite things to watch on YouTube is how older hardware stacks up from sources I trust.

If only it didn't devolve that way often, I'm not above that there are points of view and facts/nuances I haven't yet considered, so I am open to the discussion usually, just gets really hard with people that bury their head in the sand and stick to their guns no matter what. We should all be willing to learn and grow.

Raytracing was probably the 3-4 down the list of importance for me. I do like DLSS 2.0+ though much more so at 1440p than FSR.

May I ask what that list of importance looks like? Interesting to hear what people want in hardware and why.

 

[oxrufiioxo]

May I ask what that list of importance looks like? Interesting to hear what people want in hardware and why.

1. Rasterization performance at 1440p/4k
2. Price in comparison to similarly performing cards for example my ftw 3080 ti was $1399 the 3080 ti strix was $1999 which is a silly price difference for cards that perform very close to each other.
3. RTRT performance and extra features the gpu provides like DLSS vs FSR. Vram amount etc.
4. Does the card perform within 10% of the flagship gpu for example if the 3080 ti was 12% slower than a 3090 I would have just gotten the 3090.
5. Noise normalize thermals.

3 and 4 are really like 3A and 3B they are equally important to me.

I will also buy gpus if they interest me but not for either of my main 2 systems for example I picked up a sapphire nitro 5700XT back in late 2019 just to test if I could re-create the blackscreens people tended to have i couldn't in multiple systems.

I would say I'm probably not the norm though as I could upgrade yearly assuming the performance justifies it so 30-40% improvement.... But typically I try to stick with every other year assuming stock allows it.

 

[wolf]

1. Rasterization performance at 1440p/4k
2. Price in comparison to similarly performing cards for example my ftw 3080 ti was $1399 the 3080 ti strix was $1999 which is a silly price difference for cards that perform very close to each other.
3. RTRT performance and extra features the gpu provides like DLSS vs FSR. Vram amount etc.
4. Does the card perform within 10% of the flagship gpu for example if the 3080 ti was 12% slower than a 3090 I would have just gotten the 3090.
5. Noise normalize thermals.

3 and 4 are really like 3A and 3B they are equally important to me.

I am not too far off those points tbh.

1. Rast perf at my native res 3440x1440 targeting 100+ fps
2. Comparative upgrade over my existing hardware, valued case by case depending on many factors, this time was ~2x with GTX1080 to RTX3080
3. Bang for buck in the high end, generally one SKU down from the top one as this is often 50%+ more money for ~10% more performance.
4. Value add features, like RT performance and support, DLSS, NVENC etc

I figure the rest is all negotiable, like thermals, I'm happy to mod the card with pads, TIM, deshroud etc

 

[wolf]

The trend is still holding true in Farcry 6, same analysis methodology, the results used are from TechSpot (HUB).

1440P Ultra + high res textures, RTX3080 accelerates RT Reflections ~25% faster
2160P Ultra + high res textures, RTX3080 accelerates RT Reflections ~75% faster

I have a strong feeling that AMD's play, perhaps at least until we see RDNA3, and maybe beyond, is this;

Use extremely toned-down RT effects (if any at all) in sponsored titles.
Advertise that games need a lot of VRAM in sponsored titles.

This can give benefits to their marketing and mindshare twofold, make RT seem undesirable and a gimmick because the effects they choose and how they're employed can be, or outright are, extremely hard to spot. Yet, it paints their cards in a light where, by fine-tuning the specific load of the RT effects (and perhaps like in this title, giving users no way to change that), the absolute performance appears on paper to be very competitive with Ampere, leading to the illusion that RT-on performance is as good as Amperes is. A clever play really, and to some extent, it's clearly working.

Then leaning heavily on the VRAM advantage in the mid and high end, whenever possible, even if only on paper. Things like recommending 16GB VRAM, even when it isn't strictly needed, will help sell RDNA 2 cards. Shipping games with texture packs or blown-out VRAM requirements definitely plays to RDNA 2's mid/high-end strengths, even for the little-to-no visual difference.

RDNA2 v Ampere just keeps getting more and more interesting, both are really trying to push their own strengths, and highlight their competition's weaknesses.