viveks86
Member
Without OMM, SER etc? 540p 30 may be. Not to mention that code branch likely being Nvidia exclusive…
-
Hey Guest. Check out your NeoGAF Wrapped 2025 results here!
[Digital Foundry] Resident Evil Requiem - DF Analysis - Stunning on PS5 Pro + PS5/Xbox Series X|S Breakdown
- Thread starter Lunatic_Gamer
- Start date
- Analysis Review
Bojji
Member
9060XT has 322.3 GB/s of memory BW. It needs L3 cache to not shit the bed vs. GPUs that have more than 500GB/s.
Separate memory was always the thing in PC space so I don't get why you bring it now, DDR is also super slow compared to GDDR.
In raster games 6800 should offer very similar performance to PS5 Pro. Bit better thanks to L3 cache and higher power draw.
Note: Ai got RDNA3 wrong for PS5 Pro and dual issue (it's RDNA2 for raster).
Last edited:
PaintTinJr
Member
IIRC KeplerL2 said the other day that infinity cache effectively doubles or triples memory bandwidth on RDNA GPUs, and the 500GB/s for the Pro GPU is still interrupted in an interleaved way by the CPU because it is an APU, which ironically if it had an LLC of 30MB that would make a huge difference, as would an extra 50Gpixels.
Last edited:
Gaiff
SBI’s Resident Gaslighter
Yep, you're right. Path tracing is only available on NVIDIA cards in this game.
Watching a highly compressed YouTube video?
adamsapple
Or is it just one of Phil's balls in my throat?
hmm.
This sounds like another game DF tested a few months ago, I can't recall exactly which one, where they said that in the RT mode you can often see the lighting 'catch up' on camera cuts and visibly load in.
Between this and the RT flickering, it seems like John just wanted to gush on the game more.
Bojji
Member
It can't be that high. Ampere GPUs with more than 700GB/s memory speed were beating RDNA2 top dogs in 4k even if performance in 1440p was very similar or better:
Gaiff
SBI’s Resident Gaslighter
PaintTinJr
Member
But I suspect they were losing by fillrate, not memory bandwidth or there was a sponsorship deal for Nvidia or maybe he was describing the benefit on RDNA4. Either way an RX 9060XT discrete GPU with 30MB of L3 and on PCie5 mapping the GPU driver in RAM isn't at a memory bandwidth disadvantage to a PS5 Pro GPU that has no L3 cache at all.It can't be that high. Ampere GPUs with more than 700GB/s memory speed were beating RDNA2 top dogs in 4k even if performance in 1440p was very similar or better:
Skifi28
Member
RTGI taking a little time to process changes is a pretty normal optimization in most games that use it, nothing new and something you won't notice 95% of the time unless you really go looking for it. Low quality reflections with denoising issues is also quite common. Having both RT effects along with a huge increase in image quality while retaining smooth performance in a game with cutting edge visuals is still quite impressive though. After seeing so many recent games launch with 720p resolutions and god-awful RT you can barely make out, it's definitely worth gushing over even if it's not perfect
Last edited:
Giallo Corsa
Gold Member
Yep, crawling and sizzling is all over the place, it's not as bad as Alan Wake 2 or SH2Re but it's there, which brings me to : how the heck didn't DF/John notice it ?
Friggin' PSSR man...
Wut? It's not the PSSR the cause of such stuff
it's how raytracing work in RE engine
Last edited:
Bojji
Member
PS5 Pro provides roughly 78% more memory bandwidth than the 9060 XT
Now ask yourself, if 30MB of cache can make up for that difference. I think that there could be instances where PS5 Pro will win in memory BW vs this GPU. Like here:
Low resolution, Radeon wins:
Higher resolution, Pro wins:
I think it's just low quality RT, low ray counts and bad denoiser. Standard stuff for RE games.
I doubt PSSR (or whatever they use) is the issue here.
Gaiff
SBI’s Resident Gaslighter
Probably just low-quality RT. It has to run at 60fps after all, so there probably aren't that many rays being cast and the denoising part might be lacking.
I mean the latest Resident Evil which use RE engine.
No playing for 3 hours on my 55 OLED.
It reminds me of the poor Layers of fear remake which was one of the first games on UE5.
When you are playing Grace with the interruptors switching, it takes 2 seconds for the lighting to update, with huge artefacts. It's worse than SH2 Remake imho
Last edited:
It's something which was already visible in RE Village. In a minor way because if I'm not wrong raytracing was limited to the indirect light but this time is full GI so obviously quite more visible. Isn't it exactly the most precise raytracing used.
Last edited:
adamsapple
Or is it just one of Phil's balls in my throat?
Not feeling so bad about missing out on RTGI on base consoles now.
Still think they should have atleast tried to get RT reflections on them.
kevboard
Member
and that right there is why I am not a fan of RTGI.
the hardware is simply not there yet to make it look good.
in order for RTGI to be visually pleasing, it basically would need so many rays per pixel that a denoiser was almost not needed anymore... which we are a good way away from still... well, maybe a 5090 could do it lol
baked GI + RT reflections is the way to go for now. and developers should focus on that.
Last edited:
Skifi28
Member
adamsapple
Or is it just one of Phil's balls in my throat?
viveks86
Member
kevboard
Member
IT HAS DEADZONE SETTINGS AND A FUCKING REACTION CURVE SELECTION!!!!!!!!!!!!!
HALLELUJAH!!!!!
who cares about the graphics!?!? THAT was one of the worst parts of both RE8 and ESPECIALLY RE4 Remake!
after the absolute shit show that was Resident Evil 4, with the broken deadzone on Xbox, and the fucked up reaction curve they had, I guess this time they thought to prevent that shit from happening again and just go all out with their settings!
so as of now, Resident Evil 9 has better analog aim setting that fucking Halo Infinite and Destiny 2
HALLELUJAH!!!!!
who cares about the graphics!?!? THAT was one of the worst parts of both RE8 and ESPECIALLY RE4 Remake!
after the absolute shit show that was Resident Evil 4, with the broken deadzone on Xbox, and the fucked up reaction curve they had, I guess this time they thought to prevent that shit from happening again and just go all out with their settings!
so as of now, Resident Evil 9 has better analog aim setting that fucking Halo Infinite and Destiny 2
Last edited:
Gaiff
SBI’s Resident Gaslighter
Looks great in AC Shadows.
kevboard
Member
it's serviceable in some instances, but especially in horror games it's just a REALLY bad fit.
for example, in Silent Hill f, the game sometimes tries to do subtle handholding by having flickering lightbulbs guide the way... well, that flickering looks like someone installed a dimming switch and is sliding it up and down, because the RTGI reacts so slowly.
so anything where lights are moving fast, or where lights are flickering/are prone to go off and on in an instant, will make RT GI lool awful.
also really bad are dimmly lit environments with many light sources, as that tends to create insanse boiling issues. even in Doom TDA, which has generally really good RTGI, the GI breaks in such moments (which are thankfully rare enough to not get annoying)
Last edited:
Gaiff
SBI’s Resident Gaslighter
In the case of RE, it's true there is no real reason not to bake the GI. As far as I'm aware, there's no dynamic time of day, so it's not like they would have to redo a bunch of lightmaps for every instance.
viveks86
Member
Didn't you want PT on pro just a minute ago? A couple of flickering shadows and you want to go back to raster?
kevboard
Member
that's always so weird to me. baked GI is basically always superior (with current hardware) compared to RTGI if your environment is mostly static.
I understand why UE5 devs use it... it's a shortcut... but why Capcom added it here is kinda a mystery to me. and apparently you can't even toggle GI and Reflections separately on PC either... that's such a weird choice.
they take the time to bake the GI, then add inferior realtime GI, and then couple it with RT Reflections. so weird.
maybe a mod will decouple them lol
Gaiff
SBI’s Resident Gaslighter
Yeah, but upon learning that 1, PT is exclusive to NVIDIA, and 2, the RTGI is a bit slow due in no small part to a low amount of rays and we seem to have denoising issues, PT doesn't seem like a viable alternative for the Pro anymore. At least, not without a lot of flickering/noise.
Didn't you want PT on pro just a minute ago? A couple of flickering shadows and you want to go back to raster?
I was looking to see how much performance RDNA4 loses going from PT to RT, but it's not available there. We do have the numbers for a 5070 Ti and...
64% performance loss from RT lol. 1080p30 with PT wouldn't be feasible at all on the Pro.
Last edited:
kevboard
Member
Yeah, but upon learning that 1, PT is exclusive to NVIDIA, and 2, the RTGI is a bit slow due in no small part to a low amount of rays and we seem to have denoising issues, PT doesn't seem like a viable alternative for the Pro anymore. At least, not without a lot of flickering/noise.
I was looking to see how much performance RDNA4 loses going from PT to RT, but it's not available there. We do have the numbers for a 5070 Ti and...
64% performance loss from RT lol. 1080p30 with PT wouldn't be feasible at all on the Pro.
path tracing will use a shitload of rays probably.
the reflections look way cleaner with PT
Buggy Loop
Gold Member
AMD’s RDNA 2: Shooting For the Top
In 2019, AMD moved off their long-serving GCN architecture in favor of RDNA.
chipsandcheese.com
Evaluating the Infinity Cache in AMD Strix Halo
Strix Halo is the codename for AMD’s highest end mobile chip, which is used in the Ryzen AI MAX series.
chipsandcheese.com
This is well explained and microbenched, although they don't have infinity cache counters for the profiling tool.
Infinity cache is not gonna translate to equivalent VRAM GB/s memory speed it really doesn't work like that. It does alleviate their need of high bandwidth VRAM though. Infinity's cache main advantages are the cache bandwidth for medium loads, very low latency and has high cache hitrates, because L2 & L1 are probably never waiting/stalling with L3 (infinity cache) being so close.
"Infinity Cache bandwidth scales very well too, and actually closely matches RDNA 1's L2 bandwidth. It can't match L2 bandwidth on Nvidia's 3090, but it doesn't need to because the 4 MB L2 in front of it should catch a lot of accesses. So far, AMD looks pretty good in terms of cache bandwidth. VRAM however, is another story."
"Nvidia has a massive VRAM bandwidth advantage. With large workloads that don't fit in cache, Ampere is far less likely to run out of VRAM bandwidth. However, both RDNA generations are better at making use of the VRAM bandwidth they do have. They don't need as much work in flight to make good use of their available bandwidth."
So it seems Nvidia made it more general and would never choke on workloads that don't fit in cache while AMD was more fine tuned for for being really fast in certain ranges. At 4K with 128MB of cache they also lose roughly 20% hitrates from 1440p and if the cache overflows then I guess it starts to tank compared to Nvidia. RDNA 3 and 4 moved to higher bit rate and reduced cache so it might have been a better balance to have more VRAM bandwidth than relying too much on cache to save the day
Buggy Loop
Gold Member
This can be clearly seen here on PC too
Patch tracing likely using ReStir or NRC ? Nvidia has not promoted which method as of now but even ReSTIR with ray reconstruction was very fast in cyberpunk 2077. RT is just not keeping up. I think all RT solutions, be it from Nvidia or Lumen, suffer that kind of lag.
Corporal.Hicks
Banned
The RT in Resident Evil games was always low quality but at least it was extremely fast. On my PC I saw just 1fps difference in the RE4 remake between RT on and off.
However, the noise in the Digital Foundry Resident Evil Requiem video is definitely more intense, as John pointed out. That's probably because, unlike previous RE games, RE9 uses RT GI on a much bigger scale (not just to enhance raster lighting but to fully replace it). Today, I will test RE9 on PC to see how bad the RT GI noise is in this game. On consoles, it could be an issue simply running the game at an internal resolution of 1080p. RTGI games look fairly stable at 4K, but as the internal resolution is lowered, the lighting starts to boil and the reflections start to flicker.
This is especially noticeable in UE5 games. At 4K with DLAA, lumen looks fairly stable in most of the UE5 games I've played, and I'd have to nitpick to complain about anything, especially if the UE5 games use HW lumen and ray reconstruction. However at 4K with DLSS Ultra Performance (33% resolution scaling), the lumen boiling becomes very noticeable, and it's almost impossible to ignore even though DLSSUP reconstruct a fairly sharp and detailed image. That's why I dont want to play UE5 games with anything below DLSSQ.
As for Doom TDA, I saw lighting boiling with PT in dimly lit locations and especially around small light sources, but the hybrid RT GI was fully stable with DLAA and looked great. The same goes for Indiana Jones and lighter RTGI games like GTA5 EE and Metro Exodus. RTGI elevated the lighting fidelity to a new level in these games and looked amazing. I can easily see a difference between RT GI and raster even if game would use prebaked GI, becasue you cant prebake GI for dynamic objects like vegetation and character models. They will always stand out and look flat, so it's not true that prebaked lighting can look as good as RT GI if game has static TOD.
Last edited:
viveks86
Member
I would have have guessed ReSTIR PT as Nvidia has been pushing that quite aggressively for UE5, but given how heavy it is in this game (5090 can't even hit 60 fps on dlss performance), I'm going to wager it's something custom built for their engine that's more brute force with multiple rays per pixel, like AW2. That stuff will never make it to console. Not even next gen. ReSTIR or megalights are the type of approaches where more work/research needs to be done to make this all truly viable for console adoption.
kikii
Member
DigitalFegs ^^
roosnam1980
Gold Member
alan wake 2 on consoles is using some settings which are lower than low on PC , those settings are not available for PC versionPS5 Pro provides roughly 78% more memory bandwidth than the 9060 XT
Now ask yourself, if 30MB of cache can make up for that difference. I think that there could be instances where PS5 Pro will win in memory BW vs this GPU. Like here:
Low resolution, Radeon wins:
Higher resolution, Pro wins:
I think it's just low quality RT, low ray counts and bad denoiser. Standard stuff for RE games.
I doubt PSSR (or whatever they use) is the issue here.
Last edited:
PaintTinJr
Member
PS5 Pro provides roughly 78% more memory bandwidth than the 9060 XT
Now ask yourself, if 30MB of cache can make up for that difference. I think that there could be instances where PS5 Pro will win in memory BW vs this GPU. Like here:
Low resolution, Radeon wins:
Higher resolution, Pro wins:
I think it's just low quality RT, low ray counts and bad denoiser. Standard stuff for RE games.
I doubt PSSR (or whatever they use) is the issue here.
Actually Buggy Loop's image shows that it does double and triple the effectiveness of bandwidth on 6900XT compared to the other two if I'm not mistaken, and around notable workload data sizes (L1, L2, L3) does the relationship change, so the absence of the 30MB L3 on PS5 Pro compared to 9060XT is significantly different. And as I said, the CPU in Pro is interrupting the GPU in an interlaced fashion negatively impacting the L2 efficiency compared to how it would perform with a 30MB cache allowing it optimally reduce RAM access frequency by upto 7 times(30MB/4MB).
Even the 8MB CPU cache in the Pro tells us that even in an optimal situation at full workload the CPU is accessing the RAM and blocking the GPU's 500GB/s bandwidth at ~5000 per second (40(GB) 000/8(MB))
So that still looks like the PC with a mapped GPU driver in RAM and the ability to opportunistically pipeline data over PCIe5 into a large L3 does more than make up for the bandwidth deficiency, along with the disruptive L2 cache missing on the Pro - to then utilise VRAM bandwidth - and the XSS extra worth of fillrate on the 9060XT means even on VRAM bottlenecks it has a huge 33% capacity there to make up for even a 20% bottleneck buggy loop mentioned on VRAM - that the L3 can't get around.
Last edited:
CrustyBritches
Gold Member
The RT boiling you are seeing is probably from a lack of ray recon/denoiser outside of PT mode. I tried it with highest RT on both 9070 XT and 5060 Ti and regardless of res or upscaler quality settings it persists on both cards. When you enable PT it unlocks RR and looks leagues better. I didn't think to try it with PT without RR, but I'm too tired.
Last edited:
Bojji
Member
Actually Buggy Loop's image shows that it does double and triple the effectiveness of bandwidth on 6900XT compared to the other two if I'm not mistaken, and around notable workload data sizes (L1, L2, L3) does the relationship change, so the absence of the 30MB L3 on PS5 Pro compared to 9060XT is significantly different. And as I said, the CPU in Pro is interrupting the GPU in an interlaced fashion negatively impacting the L2 efficiency compared to how it would perform with a 30MB cache allowing it optimally reduce RAM access frequency by upto 7 times(30MB/4MB).
Even the 8MB CPU cache in the Pro tells us that even in an optimal situation at full workload the CPU is accessing the RAM and blocking the GPU's 500GB/s bandwidth at ~5000 per second (40(GB) 000/8(MB))
So that still looks like the PC with a mapped GPU driver in RAM and the ability to opportunistically pipeline data over PCIe5 into a large L3 does more than make up for the bandwidth deficiency, along with the disruptive L2 cache missing on the Pro - to then utilise VRAM bandwidth - and the XSS extra worth of fillrate on the 9060XT means even on VRAM bottlenecks it has a huge 33% capacity there to make up for even a 20% bottleneck buggy loop mentioned on VRAM - that the L3 can't get around.
But this is benchmark, I doubt games really represent synthetic tests. I already showed you that with increased resolution 6800XT started to lose advantage it had over 3080.
3080 has ~50% more memory BW, L3 cache is used to mitigate the difference. If L3 (or big L2) memory was so effective no GPU would have more than 512GB/s memory speed.
True, forgot about that. 9060XT is pushing more stuff in that benchmark vs. quality Pro version.
I would expect almost 18 hours after, DF confirming they don't know what tech is it, or confirming it's not PSSR2. But since they got independent... I wonder if they are using those rumors to get speak about them
saintjules
Gold Member
I mean he blames the ps5 pro hardware and the pssr but in the end it's how raytracing work in capcom engine. But I applaud the try.
Last edited:
Bojji
Member
I mean he blames the ps5 pro hardware and the pssr but in the end it's how raytracing work in capcom engine. But I applaud the try.
True.
I wonder if Sony will make their own ML denoiser for PS6 or just use AMD Ray Regeneration.