darthdiablo
+5 Death Stare
- Pronouns
- he/him
Just to give us some ideas of what's possible on a next-gen Switch 2
Last edited:
-
Hey everyone, staff have documented a list of banned content and subject matter that we feel are not consistent with site values, and don't make sense to host discussion of on Famiboards. This list (and the relevant reasoning per item) is viewable here.
-
Do you have audio editing experience and want to help out with the Famiboards Discussion Club Podcast? If so, we're looking for help and would love to have you on the team! Just let us know in the Podcast Thread if you are interested!
StarTopic Future Nintendo Hardware & Technology Speculation & Discussion |ST| (Read the staff posts before commenting!)
- Thread starter Dakhil
- Start date
Serif
𝕽𝖊𝖓𝖊𝖌𝖆𝖉𝖊 𝕬𝖓𝖌𝖊𝖑
Correct. Unless you want to include the Analogue Pocket.
'Docking' with the Steam Deck has been an ordeal.
TSR3
Pilotwings
Agreed. Orin launched Q4 2022. If T239 was coming out on Samsung 8nm then it would already be in production, the Switch 2 would have already been announced and we would be mere weeks away from the launch and hyping ourselves silly. As we do not live in that alternative timeline, I think we can treat SEC8N as a very low probability.
node doesn't dictate release timing like that
darthdiablo
+5 Death Stare
- Pronouns
- he/him
That's a bingo.
There's also another reason: the bottom line. Samsung 8N would have been more expensive to go with, when we look at per wafer cost. A major point that I think ItWasMeantToBe19 is conveniently glossing over.
On a per wafer cost, TSMC 4N is more expensive, but more denser which gives TSMC 4N "more bang for the buck". And it is an option that was available for Nintendo/nvidia at the time.
nvidia products are at present predominantly made on either SEC8N or TSMC 4N. Older ones are made on SEC8N, newer TSMC 4N.
Last edited:
TSR3
Pilotwings
The point is that SEC8N is quite an old node now. There's arguments for T239 production on TSMC 4N being held back until needed, but I'd expect any T239 production on SEC8N to start as soon as RTX 30x0 series production dropped off.
Angel Whispers
Follow me at @angelpagecdlmhq
Some thoughts… A prospective successor will be around for 7-8 years, possibly more. 7 has been the standard for HD consoles. Games are escalating in their scale and increasingly expensive to develop. Also, part of the “uncharted territory” is selling as much as they are in the 7th year. On a successor, said territory is no longer uncharted. That’s just as true for their partners. 7 years or more was also common for their portables - DS was around for 7 years, and in production for about 9. 3DS was around for 7 years, and in production for about 8. For all sorts of logical and practical reasons, there won’t be a “Pro” for their successors, and there never will. You can thank me for telling you this so that you don’t get sucked into rumour mill grifts driven by certain Youtubers and gaming publications. Developers don’t want it, especially smaller studios which target multiple platforms, and where an extra performance profile would stretch resources thin/to a point where port costs aren’t worthwhile.
8nm is highly unlikely because it’s an end-of-the-line process, which alludes to further shrinkages being near impossible - That’s something they’ll need to consider when doing a Lite/hardware revision. Also, it’s clear that this SoC has been developed alongside Nvidia’s Lovelace products, and there were whispers about a future Nintendo console having a Lovelace GPU in it as far back as 2021 - Some conflated this with the idea of a “Pro” system, but it was always the successor, and that’s been why I’ve always said that it’s better to exercise more patience, if it means we get a higher performance in the next console. Lovelace starts on Nvidia’s custom 4nm/5nm process, and it offers the best value, as others have demonstrated here - So, before “Nintendo is cheap” trolls get in their drive-by car, you’ll have to decide if that “cheapness” means they’ll apply it to get the better process, or pay more for a worse one because of a foundationless, perceived “aversion to the idea of more performance”. Take your pick. It can’t be both. I have thoughts on whether it’s full Ampere, or full Lovelace, or elements of both, but I won’t go into more details in this post (I refuse to call it Lovelamp, so, let’s go with Amperada…).
But let’s also speak to the possibility of the 8nm process - If they decide to go with that, it’s because they’ve done enough due diligence to give it the green light. So, I’m not seeing it right now for the panic, it feels a little overcooked on here. The Orin products and others made on this process have featured automotive elements, and it’s entirely possible that this was a factor in the SoC performance, heating issues, and clock speed limitations. Also, the CPUs were multi-clustered. Furthermore, the Orin products haven’t been built with RT or DLSS in mind. The truth is that none of us have enough to determine the extent to which these factors were a hindrance, whether it was the sum of those elements rather than the process itself. However, Nintendo’s SoC won’t have any of these concerns I listed, so, the Orin tools posted much earlier in this thread are extremely flawed for that reason, and they shouldn’t be taken as an indicator of the power consumption or performance. We can also concentrate on what’s being reported, where it can be traced to the horse’s mouth and developers. If, between now, reveal, and launch, developers express dissent or dissatisfaction with the successor’s performance, then, and only then will there be a reason to be concerned. I don’t believe we’re anywhere near close to that point at all, and we can continue daring to expect, based on the latest reports.
8nm is highly unlikely because it’s an end-of-the-line process, which alludes to further shrinkages being near impossible - That’s something they’ll need to consider when doing a Lite/hardware revision. Also, it’s clear that this SoC has been developed alongside Nvidia’s Lovelace products, and there were whispers about a future Nintendo console having a Lovelace GPU in it as far back as 2021 - Some conflated this with the idea of a “Pro” system, but it was always the successor, and that’s been why I’ve always said that it’s better to exercise more patience, if it means we get a higher performance in the next console. Lovelace starts on Nvidia’s custom 4nm/5nm process, and it offers the best value, as others have demonstrated here - So, before “Nintendo is cheap” trolls get in their drive-by car, you’ll have to decide if that “cheapness” means they’ll apply it to get the better process, or pay more for a worse one because of a foundationless, perceived “aversion to the idea of more performance”. Take your pick. It can’t be both. I have thoughts on whether it’s full Ampere, or full Lovelace, or elements of both, but I won’t go into more details in this post (I refuse to call it Lovelamp, so, let’s go with Amperada…).
But let’s also speak to the possibility of the 8nm process - If they decide to go with that, it’s because they’ve done enough due diligence to give it the green light. So, I’m not seeing it right now for the panic, it feels a little overcooked on here. The Orin products and others made on this process have featured automotive elements, and it’s entirely possible that this was a factor in the SoC performance, heating issues, and clock speed limitations. Also, the CPUs were multi-clustered. Furthermore, the Orin products haven’t been built with RT or DLSS in mind. The truth is that none of us have enough to determine the extent to which these factors were a hindrance, whether it was the sum of those elements rather than the process itself. However, Nintendo’s SoC won’t have any of these concerns I listed, so, the Orin tools posted much earlier in this thread are extremely flawed for that reason, and they shouldn’t be taken as an indicator of the power consumption or performance. We can also concentrate on what’s being reported, where it can be traced to the horse’s mouth and developers. If, between now, reveal, and launch, developers express dissent or dissatisfaction with the successor’s performance, then, and only then will there be a reason to be concerned. I don’t believe we’re anywhere near close to that point at all, and we can continue daring to expect, based on the latest reports.
darthdiablo
+5 Death Stare
- Pronouns
- he/him
Last edited:
Giancarlo
Nintendo connoiseur
Nintendo will choose the process node at giveSome thoughts… A prospective successor will be around for 7-8 years, possibly more. 7 has been the standard for HD consoles. Games are escalating in their scale and increasingly expensive to develop. Also, part of the “uncharted territory” is selling as much as they are in the 7th year. On a successor, said territory is no longer uncharted. That’s just as true for their partners. 7 years or more was also common for their portables - DS was around for 7 years, and in production for about 9. 3DS was around for 7 years, and in production for about 8. For all sorts of logical and practical reasons, there won’t be a “Pro” for their successors, and there never will. You can thank me for telling you this so that you don’t get sucked into rumour mill grifts driven by certain Youtubers and gaming publications. Developers don’t want it, especially smaller studios which target multiple platforms, and where an extra performance profile would stretch resources thin/to a point where port costs aren’t worthwhile.
8nm is highly unlikely because it’s an end-of-the-line process, which alludes to further shrinkages being near impossible - That’s something they’ll need to consider when doing a Lite/hardware revision. Also, it’s clear that this SoC has been developed alongside Nvidia’s Lovelace products, and there were whispers about a future Nintendo console having a Lovelace GPU in it as far back as 2021 - Some conflated this with the idea of a “Pro” system, but it was always the successor, and that’s been why I’ve always said that it’s better to exercise more patience, if it means we get a higher performance in the next console. Lovelace starts on Nvidia’s custom 4nm/5nm process, and it offers the best value, as others have demonstrated here - So, before “Nintendo is cheap” trolls get in their drive-by car, you’ll have to decide if that “cheapness” means they’ll apply it to get the better process, or pay more for a worse one because of a foundationless, perceived “aversion to the idea of more performance”. Take your pick. It can’t be both. I have thoughts on whether it’s full Ampere, or full Lovelace, or elements of both, but I won’t go into more details in this post (I refuse to call it Lovelamp, so, let’s go with Amperada…).
But let’s also speak to the possibility of the 8nm process - If they decide to go with that, it’s because they’ve done enough due diligence to give it the green light. So, I’m not seeing it right now for the panic, it feels a little overcooked on here. The Orin products and others made on this process have featured automotive elements, and it’s entirely possible that this was a factor in the SoC performance, heating issues, and clock speed limitations. Also, the CPUs were multi-clustered. Furthermore, the Orin products haven’t been built with RT or DLSS in mind. The truth is that none of us have enough to determine the extent to which these factors were a hindrance, whether it was the sum of those elements rather than the process itself. However, Nintendo’s SoC won’t have any of these concerns I listed, so, the Orin tools posted much earlier in this thread are extremely flawed for that reason, and they shouldn’t be taken as an indicator of the power consumption or performance. We can also concentrate on what’s being reported, where it can be traced to the horse’s mouth and developers. If, between now, reveal, and launch, developers express dissent or dissatisfaction with the successor’s performance, then, and only then will there be a reason to be concerned. I don’t believe we’re anywhere near close to that point at all, and we can continue daring to expect, based on the latest reports.
them the best battery life and afordable price, if 8nm can give Nintendo that option, is obvious they would choose that
MSnap
Kremling
- Pronouns
- He/Him
When it works, it’s really cool. But what’s not cool is when the menus stop working and the audio cuts out. Switch does not have that problem at all.
Evil Lucario
Star Successor
Very late to this but this means very little if we're going with the colored buttons theory for Switch 2 specifically. Bayonetta 1 on Wii U had the Super Famicom color layout, except for green being missing due to having no real symbol for guns. Bayonetta 3 adds green back in for Y, but the seeds of that were already planted back in the Wii U.
Also, A wasn't for punches, it was for kicks.
If anything, it just shows that Kamiya and his team cares about the unique aesthetics of each console manufacturer, because the original Xbox 360 version matched those controller buttons too.
ItWasMeantToBe19
Manakete
This seems to ignore the possibility that NVIDIA was extremely high on their own supply (and projected pandemic era stuff to last forever) and assumed that there would be massive demand for 4N based chips for years and years to come and said that they would be able to produce very few Switch 2s on 4N in 2024-2025.
And Nintendo just decided to go with a bulky monster instead in 2021.
is that a Unity game? cause it's not the first time I've heard of that problem with Unity
Dekuman
Kremling
- Pronouns
- He/Him/His
I don't think Nintendo wants the mobile version or it would have been on Switch a long time ago.
darthdiablo
+5 Death Stare
- Pronouns
- he/him
Was wondering if you going to respond to the point I brought up in the other comment: The fact that TSMC 4N would be cheaper compared to SEC8N on per wafer basis.
"and said that they would be able to produce very few Switch 2s on 4N in 2024-2025." Who said that?
the mobile version is still in Beta. it's meant to keep up with Warzone on other systems, though I don't know if they envision crossplay for it eventually
that's not how these things work. if Nvidia could hit Nintendo's price and performance goals on 8N, then it would be on there. if it needed 4N, it'd be on there. that's the beginning and end of it
SirArthur
Kremling
- Pronouns
- He/Him
Oh I know. My comments were made in jest. I just found it funny the whole conversation around colored button happened and suddenly Kamiya got a snes controller lol.
YolkFolk
Tingle
I don’t think that’s because of the hardware.
lemonfresh
#Team2024
- Pronouns
- He/Him
Yeah give us switch 2 nintendo, we need our frames!
Dekuman
Kremling
- Pronouns
- He/Him/His
Sorry, misread the tweet which is specifically talking about Warzone. My comment was more about CoD mobile generally. It's arguable, but my feeling is Nintendo prefers to be lumped with the consoles than be seen only getting mobile ports of games.
As for X'play it is probably something they will look at. Fortnite kind of set the stage for PC/console/mobile crossplay, and I know Warframe will add a mobile version next year which will have crossplay and cross save with all platforms as well.
D
Deleted member 887
Guest
Yeah,I’m now seeing different benchmarks. I imagine it’s rig dependent - possibly a CPU component
Concernt
Optimism is non-negotiable
- Pronouns
- She/Her
I'd also like to remind people that T239 had validation testing alongside Lovelace products, not alongside Ampere products. Sharing a similar production and design timeline with Lovelace, it would definitely make sense from that angle that it would be on Lovelace's node- 4N.
Furthermore, there's a very obvious win on the table for Nintendo and Nvidia. T239 on 4N isn't Lovelace, but it has almost every Lovelace feature relevant to a low power device. A majority of the Lovelace efficiency gains were from node shrink, while the CUDA version in T239 is one above Ampere. It's not Lovelace, no, but if you made a mobile Lovelace chip from scratch, there wouldn't be many (if any) benefits to it compared to T239.
darthdiablo
+5 Death Stare
- Pronouns
- he/him
So in your (revised, it sounds like now) thinking, do you think Switch 2 will also see similar behavior? (Slightly better FPS for SR+RR, over SR alone).
Can you expand on what you mean by "CPU component", which one?
ItWasMeantToBe19
Manakete
I'm sorry, but isn't the "price" part of this in large part determined... by NVIDIA?
NVIDIA: "If you want more than M (small number) of T239s produced before 2026 on 4N, we're going to have to charge you a huge amount. This is because all of our GPUs at this time are selling out so all of our 4N based stuff is going to sell out at any price so we would be incurring a huge opportunity cost if we sold these chips to you at a reasonable price."
Nintendo: "Okay, we'll go with the node that you're willing to use for much cheaper chips because you believe it has significantly lower opportunity cost."
Multiverse
Moblin
That does not look good. The 3D models look like paper cutouts, which is very much a mobile game thing. I expect the new Switch games to look way better than that.
This looks worse than current Switch games.
Last edited:
forgenator
Rattata
After reading a bit about the what nintendo would choose as the fab of choice. The biggest problem with SEC8N for me is that they would have to clock the switch2 lower than on 4N. And that would make it perform worse for the entirety of its life.
So even if they would go SEC 6/5 or whatever they have later, they are still bound with the original specs and every game has to run on that system.
Yes, they can release a PRO version later, but still, games NEED to work on the lowest end system nintendo has released.
If they truly want a long life for switch2, i think choosing the most performant fab at the beginning is a big thing, for the consumer.
I know it's not that cut and dry, but for me, i would be disapointed about a switch2 which isnt highperformant just for nintendo to release a pro version later? I want one system that works a long time and then a totally new system. If ive understood, ps4pro didnt receive that much support because you still need to make everything work for the weakest configuration.
Just my 2c and im hoping nintendo does right for everyone, even themselves and goes for 4N even if its a tad more expensive now.
So even if they would go SEC 6/5 or whatever they have later, they are still bound with the original specs and every game has to run on that system.
Yes, they can release a PRO version later, but still, games NEED to work on the lowest end system nintendo has released.
If they truly want a long life for switch2, i think choosing the most performant fab at the beginning is a big thing, for the consumer.
I know it's not that cut and dry, but for me, i would be disapointed about a switch2 which isnt highperformant just for nintendo to release a pro version later? I want one system that works a long time and then a totally new system. If ive understood, ps4pro didnt receive that much support because you still need to make everything work for the weakest configuration.
Just my 2c and im hoping nintendo does right for everyone, even themselves and goes for 4N even if its a tad more expensive now.
Koren Lesthe
Turrican, Terranigma and ESO lover
- Pronouns
- He/Him
.
Last edited:
darthdiablo
+5 Death Stare
- Pronouns
- he/him
Did you not see what I shared with you in my other replies to you: TSMC 4N will be cheaper than SEC8N on per wafer basis. You cannot say SEC8N chips are cheaper, because they're not.
You keep glossing over those. And nvidia doesn't determine the cost, that's TSMC. TSMC = Taiwan Semiconductor Manufacturing Company.
You also haven't responded to this question of mine
Where is this "small number on 4N" coming from? First I heard of it and I'm not sure it's a thing either.
darthdiablo
+5 Death Stare
- Pronouns
- he/him
Yes, Switch 2 will look better than that on docked at least.
What are your expectations on handheld performance? Keep in mind the video is a small screen (iPhone) zoomed in for our computer screen. When we look at the smaller screen in normal (not zoomed in), it's going to look better overall.
ItWasMeantToBe19
Manakete
TSMC sold the wafers to... NVIDIA for their usage.
NVIDIA could have bid outrageous prices for any 4N chips for their vendors for years and years (until they projected to move onto to a better node) because they projected to use all of their 4N wafers that they bought from TSMC.
darthdiablo
+5 Death Stare
- Pronouns
- he/him
So nvidia doesn't want to use those wafers for Nintendo.. because reasons? You need to start using sources/citations if you're making statements like those.
At least I can source the bit where I said TSMC 4N is cheaper compared to SEC8N on per wafer basis.. it's coming from SemiAnalysis. SemiAnalysis's primary function is to do analysis on the semiconductor industry.
ItWasMeantToBe19
Manakete
I'm sorry, how is this argument hard to follow at all.
Step 1: NVIDIA buys M amount of 4N wafers from TSMC.
Step 2: NVIDIA projects that they will be able to use all of those wafers to sell extremely high priced chips that all sell out.
Step 3: Therefore, NVIDIA puts a very high price on chips made on 4N that they would sell to their vendors (because otherwise they would be losing money relative to their projections)
Step 4: Nintendo does not decide to enter a contract with NVIDIA for production of a 4N chip because NVIDIA would charge them a huge amount.
Step 5: Nintendo instead enters a contract with NVIDIA to produce a chip using wafers with an extremely low opportunity cost as NVIDIA would charge them much less.
Prices are set based on economic cost, not production cost.
boomersooner2000
Cappy
Good day, I'm relatively new to this forum and I haven't seen much discussion on the rumor A78C CPU for the NG Switch? @oldpuck @Z0m3le How does the A78C compared to the Zen 2, can it keep up with PS5/XBS cpus? The only thing I know about the A78C is that its only configured on big cores, instead of the big, little core configuration that is on other mobile apus. Hopefully the L3 cache is high enough so it won't have to rely on a lot of memory bandwidth from the LPDDR5 modules
Last edited:
- Pronouns
- He/him
I thought the general consensus was that 8nm would be more inefficient and therefore battery life would be negatively affected?
Yep. 4nm gives them about double performance per watt compared to 8nm.
darthdiablo
+5 Death Stare
- Pronouns
- he/him
citation needed
citation needed
Even if what you said is true, it's a bold assumption that Nintendo would think it's more cost effective to go with a more unwieldy device that drains battery power significantly more - not to mention image and public perception problem - and think "yes, totally worth it".
ItWasMeantToBe19
Manakete
Yes, 8nm is extremely power hungry and would cause the device to be very big (due to the chip being chipper and due to requiring a much bigger battery).
The only reason for Nintendo to use Samsung 8nm vs. TSMC 4N is if NVIDIA was charging massively more for TSMC 4N chips.
Sol
Moblin
"all of our gpus are selling out"
except they're not. they're halting production because of overabundance of desktop gpus.
"use 8nm because it's cheaper"
except Nvidia has stopped production on Ampere because it's old so they don't have any or very little 8nm capacity
in fact, Nintendo is their only GUARANTEED customer, just like Sony and MS is for AMD. every wafer will be sold instead of sitting around waiting for material (their HPC chips) or waiting for buyers (their dgpus)
ItWasMeantToBe19
Manakete
Citation needed? I'm detailing my argument for how 8nm may have come about if it is 8nm. The Switch 2 was designed when all of NVIDIA's products were selling out at ridiculous prices. It's possible they expected this to continue and this charged a huge premium for their cutting edge node due to the projected opportunity cost.
darthdiablo
+5 Death Stare
- Pronouns
- he/him
For sure.
ItWasMeantToBe19
Manakete
When was the Switch 2's chip designed.
Obviously NVIDIA's GPUs are selling like shit now.
I could just see a future where NVIDIA made terrible projections for future demand of their 4N products and the Switch 2 ends up being extremely thicccc due to that.
They got some, for Orin. But I guess that's their only current 8nm product.
2019 to 2022 I believe.
- Pronouns
- He/him
If I may add a new point to this current discussion. I think a lot of people are obsessing with Nintendo going for something that would save them money because "they never sell their consoles at a loss."
This is simply untrue. The Wii U initially sold at a loss and it was widely publicized, so I don't get where the amnesia of that precedent being set is coming from. It made more business sense prior to the Switch for them to sell their systems at a profit because they weren't backed by large conglomerates like PlayStation and Xbox were, but the Switch has been their most successful platform ever and they're richer now than they've ever been. They also have a more diverse revenue stream than they did 10 years ago (Switch Online subscriptions, mobile game transactions, other endeavors such as their partnership with Universal) and that's important to keep in mind as well.
Sony and Microsoft sell their consoles at a loss at launch because they're expecting software sales and their online subscription to pick up the slack while they wait for manufacturing costs to drop below MSRP. Nintendo, now more than ever, can comfortably afford to do this (obviously within reason, I'm not insinuating they'd be okay with taking a sizable loss on consoles at launch) and so I don't think it's unrealistic to expect them to go against their instinct and pay a little more for something more premium.
This is simply untrue. The Wii U initially sold at a loss and it was widely publicized, so I don't get where the amnesia of that precedent being set is coming from. It made more business sense prior to the Switch for them to sell their systems at a profit because they weren't backed by large conglomerates like PlayStation and Xbox were, but the Switch has been their most successful platform ever and they're richer now than they've ever been. They also have a more diverse revenue stream than they did 10 years ago (Switch Online subscriptions, mobile game transactions, other endeavors such as their partnership with Universal) and that's important to keep in mind as well.
Sony and Microsoft sell their consoles at a loss at launch because they're expecting software sales and their online subscription to pick up the slack while they wait for manufacturing costs to drop below MSRP. Nintendo, now more than ever, can comfortably afford to do this (obviously within reason, I'm not insinuating they'd be okay with taking a sizable loss on consoles at launch) and so I don't think it's unrealistic to expect them to go against their instinct and pay a little more for something more premium.
darthdiablo
+5 Death Stare
- Pronouns
- he/him
Sounds about right. Kopite tweeted out T239 mid-2021.
So the discussion about the next Nintendo SOC would have had to begin at least a year or two earlier.
TSMC 5nm process entered HVM (high volume manufacturing) 2Q of 2020, experiencing strong ramp up in 2H of 2020.
The first nvidia products using 5nm process released was 2H 2022, I believe (RTX 40 series)
Last edited:
Multiverse
Moblin
That iPhone game still looks hollow, washed out, and has lower polygons which phone games tend to be while being sharper than Switch. It’s sharper, but the fidelity doesn’t look right. This is quite old, but this is what I mean:
My expectation for handheld?:
Last edited:
Thraktor
"[✄]. [✄]. [✄]. [✄]." -Microsoft
- Pronouns
- He/Him
The 2.0 update overall is more CPU intensive, so performance may drop on some systems compared to the previous version. I've tested it on my system (Ryzen 5900X & RTX 3070, so it shouldn't be CPU-limited), and I get around a 5% increase in performance from using RR vs the standard denoiser, both with otherwise identical settings on 2.0.
I had a brief play around with it the other day. The RTX 3070 is on the lower end of capability for the RT overdrive mode, so I can get a stable enough 30fps at 720p internal resolution, but that's about it. I mostly tested with DLSS ultra-performance mode with 4K output, for curiosity if nothing else, and it's just about usable, and better in many ways than the standard denoiser, but still has quite a few noticeable artefacts and issues.
One thing probably worth noting is that they've changed the standard denoiser since the first release of the RT overdrive mode. I played around with it back then, and honestly it was pretty rough, with lots of fireflies on specular highlights, issues with fine detail (eg wire fences blurring into the background behind them) and noticeable boiling when you pan the camera. A lot of these problems have been reduced, but it does seem the denoiser is a lot more aggressive now. Perhaps this is just from comparison to RR, but the denoiser seems to be removing a lot of detail, even at higher internal resolutions. Of course this is inherently a trade-off in denoising between preserving detail and reducing noise and related artefacts.
Using ultra-performance DLSS without RR, the results are unsurprisingly blurry. The game looks like it's running at a very low resolution, which of course it is, but DLSS doesn't seem to be able to do much about it. With RR, detail and sharpness increase considerably. You're not going to mistake it for a native 4K rendering, even in stills, but it's resolving much more detail than with the standard denoiser. It does feel like it errs a bit too far into over-sharpening, though, with some edges appearing a bit unnaturally defined, and haloing occasionally visible. There's also some ghosting and trailing noise behind moving objects and characters, particularly noticeable in high-contrast scenarios.
It does handle lighting in and around moving objects a lot better than the standard denoiser, though. One example is indirectly lit moving cars, where they don't have a direct sun shadow, but have a dark area underneath, which is the result of occluded indirect illumination. With the standard denoiser, this shadow trails behind the car as it moves, as the denoiser over-weights data from previous frames. With RR, the shadow tracks the car almost perfectly, as it seems to do a better job of throwing out old data when it realises the lighting has changed.
Overall it's a huge improvement over the previous solution, although I wouldn't quite recommend using it with ultra-performance mode. Alex at Digital Foundry posted a technical review of it which is well worth watching:
He's running on a more powerful system, with a much higher internal resolution (and better framerate, meaning better temporal data for DLSS to work with). Overall I'd agree with all his comments, only that they're much more pronounced when you push down the internal res. The improvement in detail is much more noticeable at a lower internal res, but so are the artefacts.
It's also the first version of a new technique, so I'd imagine it will improve quite a lot over subsequent versions, much like DLSS 2.0 did. One thing I've noticed is that it does a very good job handling how light and shadow respond to moving objects or moving lights, but falls over a bit on the moving objects themselves, with occasionally noticeable trails, or the phantasmal objects shifting in and out of existence in Alex's video. I feel like they really focussed on the former when creating and training the model, because it was one of the main problems DLSS RR was designed to solve, but may have taken the latter for granted a bit.
Last edited:
