I wouldn't read much into PT numbers since how lopsided they have been not just for AMD vs. nvidia, but also for intel vs. nvidia.
The path tracing updates to Portal and Cyberpunk have quite poor numbers on AMD but also on intel. Arc770 goes from being ~50% faster than 2060 to 2060 being 25% faster when you change from RT Ultra to Overdrive. This despite the intel cards' RT hardware which is said to be much better than AMD if not at nvidia's level.
The later path tracing updates to classic games of Serious Sam and Doom had the 6900XT close to 3070 performance. Last year, I benched 6800XT vs 4090 in the old PT updated games and heavy RT games like updated Witcher3 and Cyberpunk, and 4090 was close to 3.5x of 6800XT.
With the next-gen of consoles likely using AMD GPUs that are good enough at path-tracing, we'd see game engines optimizing for AMD hardware for path-tracing. I expect many scenarios like the Starfield launch where nvidia cards were much slower despite showing 100% usage.
No doubt a lot of this discrepancy is, but there's no denying NVIDIA just has stronger HW (look at how 40-50 series destroys even 20-30 series in PT at iso raster), but clearly as you said they've optimised for their black box SDKs unlike everyone else.
AMD's Xbox PC strategy + sharing GPU dies across all looks like the perfect strategy to end NVIDIA's stranglehold on PC gaming and change the game to be on AMD's term. Add AMD finally taking PT seriously and doing a ton of work to make it run faster with RDNA5.
Sure also look at Doom TDA RT and Assasin Creed shadows performs unusually well on AMD hardware.
but there's no denying NVIDIA just has stronger HW
Of course, I also agree that nvidia have quite the lead. My example of Starfield was also meant to convey that nvidia have better hardware, but the game optimization is equalizing them with AMD.
What irates me is the thinking that since nvidia is better at RT/PT, then the performance relative to AMD should keep increasing as RT/PT workload keeps getting higher and higher to infinity. Which I think is quite absurd. Like when Portal RTX released, 6800XT got 1fps while 4090 was around 20, and people that was just normal.
look at how 40-50 series destroys even 20-30 series in PT at iso raster
I will look up those benchmarks, but I agree that nvidia have been improving their hardware for RT/PT. Even in these benchmarks that I linked before, the 2080Ti is way behind 3070 despite being quite close in raster.
Yep game optimization is king and here NVIDIA has the advantage.
Lol. No I don't think it's quite that big xD but would be interested in seeing someone compare the cards for who completes the offline renderer the fastest. That would probably be the most apples to apples comparison for the PT capabilities of each gen.
Yep AMD really didn't or probably even could optimise for that. NVIDIA Remix ReSTIR black box and AMD card just dies. Maybe it ran out of VRAM, IDK. RDNA2 had terrible RT stack, very inefficient in terms of VRAM usage. RDNA4 is much closer to NVIDIA here.
The best example I can think of where you can really see the difference is Indiana Jones and the great circle in the jungle section (beginning of game always used for benchmarking). IIRC there even the 4070 pulls ahead of the 3090 and 3090 TI. IIRC DF also did some testing back when OMM was added in Cyberpunk 2077 in the city park section and they got +30% gains on the 40 series cards.
Only expect this gap to widen. Who knows what new primitive and optimization tricks 60 series will bring, but sure hope AMD can light a fire under their arse (RDNA 5 RT HW is no joke if patents are any indication) and force them to be less complacent. Pretty much coasting 20 series and on. OMM+SER were low hanging fruits + RTX MG in HW is nice but really doesn't move the needle. LSS is another OMM or SER situation and very cool tech but again they haven't really pulled the HW lever since Turing unlike AMD when they doubled everything in RDNA 4 (RT intersections for boxes AND triangles).
If I were to guess I would expect 60 series to bring at least two new major HW level features like they did with 40 series OMM+SER but we'll see.
That 3070 vs 2080 TI gap at Phantom Liberty PT at 1080p is absolutely wild. Falls apart on highr res due to VRAM. Concurrent compute + RT was a big deal for 30 series. They also increased caches by 33% that prob helped significantly as well.
And to adress the last point in your prev comment abourt PT it also looks like we're getting path tracing on the PS5. Yep didn't think that was possible. Unreal Engine 5's Megalights is some insane tech (DF talked about it yesterday). I don't expect ReSTIR to be the standard moving forward, it's just too demanding.
PS5 PRO will get nerfed PS6 PT I would wager but if they can make megalights run on a PS5 then imagine what they'll be capable off in +2027 on a PS6 with proper optimization, not ReSTIR wasteful PT.
Maybe it ran out of VRAM, IDK. RDNA2 had terrible RT stack, very inefficient in terms of VRAM usage
Yep, AMD seem to have slacked off on VRAM optimization since their Fury 4GB HBM days. And it's not just RT, even in some normal games the German review sites, pcgameshardware and computerbase, found that nvidia cards do better in scenarios close to VRAM limitation.
As for Portal RTX, one of the users here profiled it on a RDNA2 card and it was twiddling its thumbs most of the time. I don't have any AMD card now, but my experience was the same with 6800XT using <200W power when it would easily coast over 300W for a heavy game.
Similarly for Starfield, 4090 would show 100% usage, but <300W power when it's the norm for it to go over 400W easily with heavier games. This one I didn't like as much since I had to play Starfield on 4090 while with Portal RTX, the 6800XT was in secondary rig.
it also looks like we're getting path tracing on the PS5
Will check that DF video out, I usually get their videos on my youtube feed. Just wish that they would be more critical for 1st order observations like LoD/texture pop-in issues as much as they focus on whether lightning in a scene is correct or not, which requires a non-layman PoV. For instance, on the Cyberpunk sub, you will see so many posts complaining about the abject texture popin it has, but if you'd watch DF videos you'd come away thinking it's the best looking game by a country mile.
Anyway, we are living in interesting times with RT solutions being so new and so black-boxy. I'm just hoping that I get a "I told you so!' moment for when some AMD optimized path-traced console port ends up running abysmally on nvidia. :)
I think it is very interesting that with RT they're essentially tied considering they're different tiers. The PT shows weakness, but is 30 fps really all that good? Is it really something to be concerned about if a $1000+ can barely do it?
I think Nvidia is miles ahead, but PT just looks like it is a tech demo right now.
I say this as a 4090 owner. I don't think PT has moved on from something I turn on for the wow factor and promptly turn off afterwards.
I personally disagree that it’s wishful thinking, udna is going to be a insane generation based off of the rumors and leaks, this generation will sell to both consumers and data centers, and because of that in my opinion render servers. I think amd sees that this is a area Nvidias is struggling in and can capitalize off of the opportunity, and again it’ll let them sell the product as a high end renderer
Yeah but it’s not like it’s unreasonable due to the unifying of architectures. If there’s a time it would be now. I’m most worried about the price not being the leaked price than the specs being off
Just like old gen RT. Nvidia leads an early adopter tax strategy. While AMD pointing out when its ready. They are ready to sell it to masses. Because 30fps promise on a 1000 dollar card is just sooo ridiclous. When it will be obsolete next gen anyway. Just like how ampere cards suffocated on rt demanding benchmarks vs rdna2.
The lower end you go, the more nuanced the feature gets. It's not an instant on for anyone. Not even owners of high-end cards. I owned a 3080 for years and never enabled it and now I own a 4090 and use it on control and cyberpunk, but I don't enable path tracing at all.
It's important to note that this is Q upscaling at 4k. Given that the game has DLSS4, you can easily go to P at 4k with only minor visual impact, which will substantially improve throughput.
(Less noteworthy but still relevant: the review is comparing DLSS4 performance numbers with FSR 3.1 performance numbers; Obviously the former is far higher quality, but also slightly more expensive. A more apples-to-apples DLSS4 vs FSR4 comparison would most likely eliminate the lead of the 9070XT in the non-PT scenario)
: the review is comparing DLSS4 FPS numbers with FSR 3.1 FPS numbers; obviously the former is far higher quality, but also slightly more expensive. A more apples-to-apples DLSS4 vs FSR4 comparison would most likely eliminate the lead of the 9070XT in the non-PT scenario
isn't DLSS4 slower than DLSS3.1 by about the same ratio?
The 5080 is only 20mm² bigger (350mm² vs 370mm²). They're a very similar sized chip, what's impressive is how much of the die Nvidia is spending on AI and RT acceleration yet still managing to provide very nice raster performance. Also this game is heavily optimized for AMD hw across multiple devices that's how lower end RDNA2 can run the game so well on PC, hell the Series S can run the game with RT GI at 60fps!
It still uses substantially faster memory, which impacts cost and die size because now AMD uses more cache to compensate. I don't disagree that the 5080 is impressive compared to RDNA4. Nvidia has for a long time had a leaner architecture comparably.
In any case, it doesn't change the fact that it costs substantially more. So it is a different class of product even if technically they should be comparable.
If anything, the poor equation of performance loss X quality gain against RT shows how good its regular RT is. And I think that's in part because of the “forced RT”. Without a Raster mode every line of code, every piece of level design, mesh, texture — everything is built and tested and optimized for RT first & RT only, and it shows.
I am convinced this is a new trend: we will have more “RT-only” games hitting a great sweet spot of visual fidelity + performance and making the Path Tracing option look silly— until new GPUs are fast enough in PT.
The things is do any gpus perform ray tracing well without being a space heater? Energy Is so damned expensive lately that every hour the computer is blasting at 600W is an hour of the ac blasting at 200W.
And I don’t have an ac unit right in the bedroom so these AAA games are like running a v8 engine in the room while I wait for the air to sort of dissipate as the vernal air blows everywhere instead of in the space that needs it.
Lately I’ve been frame limiting my games and turning graphics down because I’m not an off worlder with infinite money to spend. Do people just let their computers sauna them into heat exhaustion? I’m not exaggerating, a AAA game heats my room to 27C.
I think it's neat to include the tech for scalability for the future and developer training. It'll be nice to turn it on in 5-10 years when average people can get a GPU that'll run it well.
MrMPFR@reddit (OP)
TL;DR:
At 4K+Q upscale at RT Ultra Nightmare 9070XT wins over RTX 5080 by 2%
\^ vs PT 4K+Q upscale results:
- 5080 PT -56%, or RT Ultra Nightmare 2.27X/127% faster.
- 9070XT PT = -76% or RT Ultra Nightmare 4.17X/317% faster.
9070XT performance in Abyssal forest completely craters (-82%). RT max vs PT think it's roughly:
- 5080 82-83FPS vs 34FPS
- 9070XT 82FPS vs 15FPS
Tied vs 2.27x NVIDIA advantage.
Normalised to 5080 RT = 100 the averages are:
9070XT RT = 102
5080 RT = 100
9070XT PT = 24.5
5080 PT = 44.4
Tied vs 1.81x NVIDIA advantage.
Conclusion: AMD has a long way to go before matching NVIDIA Blackwell. Total frame time is more than just PT ms, so difference larger there. Chips and cheese Cyberpunk 2077 PT sample frame was 78.5% PT and 21.5% other. https://chipsandcheese.com/p/shader-execution-reordering-nvidia-tackles-di
The rumoured AMD RDNA5 has a mountain to climb if hypothetically they're serious about PT with nextgen consoles. 50 series isn't even there yet.
bctoy@reddit
I wouldn't read much into PT numbers since how lopsided they have been not just for AMD vs. nvidia, but also for intel vs. nvidia.
The path tracing updates to Portal and Cyberpunk have quite poor numbers on AMD but also on intel. Arc770 goes from being ~50% faster than 2060 to 2060 being 25% faster when you change from RT Ultra to Overdrive. This despite the intel cards' RT hardware which is said to be much better than AMD if not at nvidia's level.
https://www.techpowerup.com/review/cyberpunk-2077-phantom-liberty-benchmark-test-performance-analysis/6.html
The later path tracing updates to classic games of Serious Sam and Doom had the 6900XT close to 3070 performance. Last year, I benched 6800XT vs 4090 in the old PT updated games and heavy RT games like updated Witcher3 and Cyberpunk, and 4090 was close to 3.5x of 6800XT.
https://www.pcgameshardware.de/Serious-Sam-The-First-Encounter-Spiel-32399/Specials/SeSam-Ray-Traced-Benchmark-Test-1396778/2/#a1
With the next-gen of consoles likely using AMD GPUs that are good enough at path-tracing, we'd see game engines optimizing for AMD hardware for path-tracing. I expect many scenarios like the Starfield launch where nvidia cards were much slower despite showing 100% usage.
MrMPFR@reddit (OP)
No doubt a lot of this discrepancy is, but there's no denying NVIDIA just has stronger HW (look at how 40-50 series destroys even 20-30 series in PT at iso raster), but clearly as you said they've optimised for their black box SDKs unlike everyone else.
AMD's Xbox PC strategy + sharing GPU dies across all looks like the perfect strategy to end NVIDIA's stranglehold on PC gaming and change the game to be on AMD's term. Add AMD finally taking PT seriously and doing a ton of work to make it run faster with RDNA5.
Sure also look at Doom TDA RT and Assasin Creed shadows performs unusually well on AMD hardware.
bctoy@reddit
Of course, I also agree that nvidia have quite the lead. My example of Starfield was also meant to convey that nvidia have better hardware, but the game optimization is equalizing them with AMD.
What irates me is the thinking that since nvidia is better at RT/PT, then the performance relative to AMD should keep increasing as RT/PT workload keeps getting higher and higher to infinity. Which I think is quite absurd. Like when Portal RTX released, 6800XT got 1fps while 4090 was around 20, and people that was just normal.
I will look up those benchmarks, but I agree that nvidia have been improving their hardware for RT/PT. Even in these benchmarks that I linked before, the 2080Ti is way behind 3070 despite being quite close in raster.
https://www.pcgameshardware.de/Serious-Sam-The-First-Encounter-Spiel-32399/Specials/SeSam-Ray-Traced-Benchmark-Test-1396778/2/#a1
MrMPFR@reddit (OP)
Yep game optimization is king and here NVIDIA has the advantage.
Lol. No I don't think it's quite that big xD but would be interested in seeing someone compare the cards for who completes the offline renderer the fastest. That would probably be the most apples to apples comparison for the PT capabilities of each gen.
Yep AMD really didn't or probably even could optimise for that. NVIDIA Remix ReSTIR black box and AMD card just dies. Maybe it ran out of VRAM, IDK. RDNA2 had terrible RT stack, very inefficient in terms of VRAM usage. RDNA4 is much closer to NVIDIA here.
The best example I can think of where you can really see the difference is Indiana Jones and the great circle in the jungle section (beginning of game always used for benchmarking). IIRC there even the 4070 pulls ahead of the 3090 and 3090 TI. IIRC DF also did some testing back when OMM was added in Cyberpunk 2077 in the city park section and they got +30% gains on the 40 series cards.
Only expect this gap to widen. Who knows what new primitive and optimization tricks 60 series will bring, but sure hope AMD can light a fire under their arse (RDNA 5 RT HW is no joke if patents are any indication) and force them to be less complacent. Pretty much coasting 20 series and on. OMM+SER were low hanging fruits + RTX MG in HW is nice but really doesn't move the needle. LSS is another OMM or SER situation and very cool tech but again they haven't really pulled the HW lever since Turing unlike AMD when they doubled everything in RDNA 4 (RT intersections for boxes AND triangles).
If I were to guess I would expect 60 series to bring at least two new major HW level features like they did with 40 series OMM+SER but we'll see.
That 3070 vs 2080 TI gap at Phantom Liberty PT at 1080p is absolutely wild. Falls apart on highr res due to VRAM. Concurrent compute + RT was a big deal for 30 series. They also increased caches by 33% that prob helped significantly as well.
And to adress the last point in your prev comment abourt PT it also looks like we're getting path tracing on the PS5. Yep didn't think that was possible. Unreal Engine 5's Megalights is some insane tech (DF talked about it yesterday). I don't expect ReSTIR to be the standard moving forward, it's just too demanding.
PS5 PRO will get nerfed PS6 PT I would wager but if they can make megalights run on a PS5 then imagine what they'll be capable off in +2027 on a PS6 with proper optimization, not ReSTIR wasteful PT.
bctoy@reddit
Yep, AMD seem to have slacked off on VRAM optimization since their Fury 4GB HBM days. And it's not just RT, even in some normal games the German review sites, pcgameshardware and computerbase, found that nvidia cards do better in scenarios close to VRAM limitation.
As for Portal RTX, one of the users here profiled it on a RDNA2 card and it was twiddling its thumbs most of the time. I don't have any AMD card now, but my experience was the same with 6800XT using <200W power when it would easily coast over 300W for a heavy game.
Similarly for Starfield, 4090 would show 100% usage, but <300W power when it's the norm for it to go over 400W easily with heavier games. This one I didn't like as much since I had to play Starfield on 4090 while with Portal RTX, the 6800XT was in secondary rig.
Will check that DF video out, I usually get their videos on my youtube feed. Just wish that they would be more critical for 1st order observations like LoD/texture pop-in issues as much as they focus on whether lightning in a scene is correct or not, which requires a non-layman PoV. For instance, on the Cyberpunk sub, you will see so many posts complaining about the abject texture popin it has, but if you'd watch DF videos you'd come away thinking it's the best looking game by a country mile.
Anyway, we are living in interesting times with RT solutions being so new and so black-boxy. I'm just hoping that I get a "I told you so!' moment for when some AMD optimized path-traced console port ends up running abysmally on nvidia. :)
the_dude_that_faps@reddit
I think it is very interesting that with RT they're essentially tied considering they're different tiers. The PT shows weakness, but is 30 fps really all that good? Is it really something to be concerned about if a $1000+ can barely do it?
I think Nvidia is miles ahead, but PT just looks like it is a tech demo right now.
I say this as a 4090 owner. I don't think PT has moved on from something I turn on for the wow factor and promptly turn off afterwards.
SharkBaitDLS@reddit
PT is now where RT was on the 30-series cards. I wouldn’t be surprised if the 60-series brings enough improvements on PT to make it worth it.
why_is_this_username@reddit
Honestly I think amd is gonna go hard into path tracing and force NVIDIA to play catch up for once.
soru_baddogai@reddit
I wish that would happen but knowing AMD's Radeon division this is wishful thinking at best.
why_is_this_username@reddit
I personally disagree that it’s wishful thinking, udna is going to be a insane generation based off of the rumors and leaks, this generation will sell to both consumers and data centers, and because of that in my opinion render servers. I think amd sees that this is a area Nvidias is struggling in and can capitalize off of the opportunity, and again it’ll let them sell the product as a high end renderer
gartenriese@reddit
This has been said for how many years now? Don't fall for the hype.
why_is_this_username@reddit
Yeah but it’s not like it’s unreasonable due to the unifying of architectures. If there’s a time it would be now. I’m most worried about the price not being the leaked price than the specs being off
Sevastous-of-Caria@reddit
Just like old gen RT. Nvidia leads an early adopter tax strategy. While AMD pointing out when its ready. They are ready to sell it to masses. Because 30fps promise on a 1000 dollar card is just sooo ridiclous. When it will be obsolete next gen anyway. Just like how ampere cards suffocated on rt demanding benchmarks vs rdna2.
ResponsibleJudge3172@reddit
Again, its not like the RT core gains are exclusive to lower spec Nvidia cards. People have been using RT on 2060 till 5060
the_dude_that_faps@reddit
The lower end you go, the more nuanced the feature gets. It's not an instant on for anyone. Not even owners of high-end cards. I owned a 3080 for years and never enabled it and now I own a 4090 and use it on control and cyberpunk, but I don't enable path tracing at all.
Vb_33@reddit
Speak for yourself it's an instant on for me on AAA games.
the_dude_that_faps@reddit
Exactly my point. It's not an instant on for everyone. Which is why it is nuanced.
DuranteA@reddit
It's important to note that this is Q upscaling at 4k. Given that the game has DLSS4, you can easily go to P at 4k with only minor visual impact, which will substantially improve throughput.
(Less noteworthy but still relevant: the review is comparing DLSS4 performance numbers with FSR 3.1 performance numbers; Obviously the former is far higher quality, but also slightly more expensive. A more apples-to-apples DLSS4 vs FSR4 comparison would most likely eliminate the lead of the 9070XT in the non-PT scenario)
noiserr@reddit
isn't DLSS4 slower than DLSS3.1 by about the same ratio?
DuranteA@reddit
Yes. Which is why a DLSS4 to FSR3 performance comparison, as in this article, is inappropriate, which is my point.
Vb_33@reddit
The 5080 is only 20mm² bigger (350mm² vs 370mm²). They're a very similar sized chip, what's impressive is how much of the die Nvidia is spending on AI and RT acceleration yet still managing to provide very nice raster performance. Also this game is heavily optimized for AMD hw across multiple devices that's how lower end RDNA2 can run the game so well on PC, hell the Series S can run the game with RT GI at 60fps!
the_dude_that_faps@reddit
It still uses substantially faster memory, which impacts cost and die size because now AMD uses more cache to compensate. I don't disagree that the 5080 is impressive compared to RDNA4. Nvidia has for a long time had a leaner architecture comparably.
In any case, it doesn't change the fact that it costs substantially more. So it is a different class of product even if technically they should be comparable.
Kryohi@reddit
Couldn't agree more tbh.
Strazdas1@reddit
but it wont run max settings on my 210GTX therefore lazy devs unoptimized slop.
ThinVast@reddit
when ps7 comes, we'll have forced path tracing and then people will complain why can't we run games with ray tracing only?
UnexpectedFisting@reddit
The Reddit hive mind will be screaming about this in due time
Vb_33@reddit
Bet you they'll get this game running on Switch 2
kwirky88@reddit
The things is do any gpus perform ray tracing well without being a space heater? Energy Is so damned expensive lately that every hour the computer is blasting at 600W is an hour of the ac blasting at 200W.
And I don’t have an ac unit right in the bedroom so these AAA games are like running a v8 engine in the room while I wait for the air to sort of dissipate as the vernal air blows everywhere instead of in the space that needs it.
Lately I’ve been frame limiting my games and turning graphics down because I’m not an off worlder with infinite money to spend. Do people just let their computers sauna them into heat exhaustion? I’m not exaggerating, a AAA game heats my room to 27C.
StevannFr@reddit
This game is too greedy
dztruthseek@reddit
Who's bech are we marking?
letsgoiowa@reddit
I think it's neat to include the tech for scalability for the future and developer training. It'll be nice to turn it on in 5-10 years when average people can get a GPU that'll run it well.
OutlandishnessOk11@reddit
Last time I profile this game, the ray tracing portion occupied less than 5% of the frame.