I just assume the RT cores are so integrated in Ampere, the R&D cost of removing them would be almost like creating a new architecture. Even Orin kept the RT cores, even though there's probably not a lot of use for them in a self driving car.
-
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.
StarTopic Future Nintendo Hardware & Technology Speculation & Discussion |ST| (Read the staff posts before commenting!)
- Thread starter Dakhil
- Start date
They weren't really "added" though. They were there from the beginning in the Ampere architecture, which Nvidia and Nintendo decided to use for Drake.
LiC
Member
I think this is probably overstating things. Drake's SMs are not identical to Orin's, which are not identical to desktop Ampere's, and especially if Drake is on a different process node, the work is already being done to create a new physical design and you would have every opportunity to remove an unused (and largely independent) piece of circuitry from the SM.
But this discussion is completely moot anyway. There has never been a hypothetical chance that ray tracing might not be used in practice despite the accelerators being in the SoC, even before we heard about the ray tracing demo at Gamescom. Like, when we were going through the Nvidia leak, how did we find out about ray tracing capabilities? Did I use an electron microscope and count the RT cores inside the chip? No, I found a place where the driver sets up ray tracing and uses a definition of RT_CORE_COUNT = 12. If there were 12 RT cores on the chip but they weren't used, then that definition would say RT_CORE_COUNT = 0 and the ray tracing code wouldn't exist. Oh, and there wouldn't be a ray tracing API in NVN2 and a 4000-word section in the NVN2 documentation about how to use the ray tracing API and bunch of sample programs and automated tests for the ray tracing code.
ItWasMeantToBe19
Manakete
I wonder what is the maximum level of RT you could use that would saturate all 12 RT cores without touching other GPU resources.
(the CPU load due to generating BVHs would also be pretty intense and the denoiser load could be intense as well is an issue here too. Would have to also do very custom and simplified denoisers and BVH update systems)
(the CPU load due to generating BVHs would also be pretty intense and the denoiser load could be intense as well is an issue here too. Would have to also do very custom and simplified denoisers and BVH update systems)
Raccoon
Fox Brigade
- Pronouns
- He/Him
Nvidia has been pushing RT for a long time, even before RT cores and DX12 was a thing. finally having enough power in the low end to run RT, they're gonna push for it whether Nintendo wanted it or not. and it's not like Nintendo would say no
- Pronouns
- He/Him
I beg you to use your photon receptors that are attached and embedded into your skull along with your billions upon billions of neurons to read and analyze what is given to you carefully and analyze the data that has been given to you, numerous times, across the thread.
LiC
Member
Just want to be clear that the first line here is a big understatement of the point of my post. There aren't just uses for them. They are 100% confirmed to be used. So, the question of "would they keep them on die even if they weren't using them" is moot.
As for removing other hardware, they did. For example, they removed the NVJPEG, DLAs, and a bunch of camera and image processing hardware. Re: "integrated in the architecture," the OFAs are outside the GPC altogether, so they likely would have been as easy to remove as NVJPEG. Why didn't they remove them? Who knows. Maybe Nvidia was hoping they could get Nintendo to make use of them somehow.
Last edited:
ItWasMeantToBe19
Manakete
While custom hardware will (in future gens) make the CPU burden due to RT irrelevant, quality RT is actually pretty CPU expensive at this point.
It needs a lot more than just RT cores.
As Pokemon and Zelda are the games most discussed about future RT applications, this is especially relevant as Zelda (due to its ambition) and Pokemon (due to GF's fucking horrible programming) seem extremely CPU bound.
It needs a lot more than just RT cores.
As Pokemon and Zelda are the games most discussed about future RT applications, this is especially relevant as Zelda (due to its ambition) and Pokemon (due to GF's fucking horrible programming) seem extremely CPU bound.
Last edited:
- Pronouns
- He/him
My girlfriend isn't good at programming either
I thought AMD's solution for RT required additional CPU power (and cut into GPU power) because of something not present in their GPUs, but was present in Nvidia's GPUs? Maybe I'm not remembering correctly....
SoldierDelta
Designated Xenoblade Loremaster
- Pronouns
- he/him
Considering it's a good flex to have when compared to other hardware, I think Nintendo would hear Nvidia's elevator pitch and said "Oh my God!" before grabbing 50 chequebooks.
Jury is still out as to how much Nintendo would go into the feature, but there's some decent odds that they'll experiment with it. Hell, we might actually see Nintendo try and pull off a technical showcase with a game.
Teal'c
Shriekbat
Ah I understand, I thought it could be a possibility, never mind
ziggyrivers
thunder
- Pronouns
- He/him/his
Good points. The Switch isn’t just a hybrid because it delivers home consoles experiences on a portable form factor that can enable TV, it also factors mobile aspects such as software management, which is the reason as to why the console’s OS is so light: to prioritize software processing. There’s also how to imitate Apple in other aspects. If youre imitating something, and consumers notice, it’d be contradictory to not imitate the aspect of carrying everything over and ensuring compatibility
CalmMind
Unconventional wisdom
I rather be someone obnoxious than someone oblivious to the obvious.
LiC
Member
If you're just talking about hardware specifically inside the GPU block, I think NVJPEG is the only one I listed. In the case of the OFA, though, it's inside the GPU but it's not inside the GPC. It's probably true that some hardware is harder to remove than others, I just think OFA is probably easy to remove. And I'm just not sure of the relevance of the Ampere architecture making it out like RT cores are so integral to the entire GPU or SM that it would take re-architecting to get rid of them. Fusing them off, the alternative to physically getting rid of them, still means the GPU has to be able to operate as if they weren't there, so if you have to re-lay out the circuitry anyway, you might as well delete it. But, obviously this is just my uneducated opinion.
TheSentry
Piranha Plant
I don’t think anyone here is arguing that the Seitch 2 will do full path tracing like Cyperpunk 2077 Overdrive.
But I believe Nintendo will totally make use of the RT cores to do at the least some ray tracing features.
I’m sorry but if third-parties can get some RT running on the Series S, then Nintendo will be able to get some RT running on the Switch 2.
But I believe Nintendo will totally make use of the RT cores to do at the least some ray tracing features.
I’m sorry but if third-parties can get some RT running on the Series S, then Nintendo will be able to get some RT running on the Switch 2.
Raccoon
Fox Brigade
- Pronouns
- He/Him
so real
D
Deleted member 887
Guest
There are uses for the OFA, but I think the reason the OFA is there is because Nvidia is planning on adding more uses in the future. That’s why the OFA is in the hardware in the first place, and removing it eliminates Nintendo’s ability to benefit from those future improvements.
As long as Drake looks like a standard (but small) RTX 30 card, then Nintendo will benefit from every software improvement that Nvidia delivers to those cards, and every improvement made on Nintendo’s behalf can go back into the PC space. It’s a win-win for everybody.
The person you are replying to is on my ignore list, because they seem to value winning the argument over talking in good faith.
But you remember correctly. Not that Nvidia’s RT solution is cheap on the CPU. But AMD’s is heavier.
Do you remember the old meme of which you would rather fight - one horse sized duck or a hundred duck sized horses? That’s actually a half way decent way to think about these two architectures.
When it comes to AI and RT, Nvidia is the big duck, and AMD is the tiny horse. The big duck is the more powerful animal, but there is a line where it’s just overwhelmed by the sheer number of tiny horses.
The Series X and the PS5 are a shitload of horses. The Series S isn’t. This is why folks have trouble grasping what NG is potentially capable of. “The Series S is 4 horses! How will Switch, with a max of 3 ducks keep up?”
Sir, you clearly don’t understand how big these fucking ducks are. They eat horses, don’t they?
Conversely, there are a few Nvidia stans who don’t realize that, when it comes to “classic” graphics, the situation is reversed. An RDNA 2 FLOP is generally “better” than an Ampere FLOP when it comes to raw pixel pushing. Nvidia compensates by shoving more FLOPS (tiny horses) per dollar in their cards.
darthdiablo
+5 Death Stare
- Pronouns
- he/him
- Pronouns
- He/Him
I theorize a use for the OFA in something like Drake is that it would be for videos that are streamed or recorded stuff if it’s going to be used at all that is, since the OFA in Turing and Ampere were advertised for video before.
And it could be used for creating higher quality captures on the Drake Switch that you can share without necessarily taking up more than the needed amount of space for say… the RAM.
darthdiablo
+5 Death Stare
- Pronouns
- he/him
Codename_Steam
Cappy
- Pronouns
- He/Him
If the final spec for Drake were:
8XA78C @ 1.7 to 2.1GZ
Handheld Mode GPU: 1.8 - 2.0 Teraflops
Docked Mode GPU: 3.0-3.5 Teraflops
12GB of RAM @ 102GB/S (10GB for gaming, 2GB for the OS)
256GB or 512GB Storage @ 1GB/S
Would it be possible to run current gen games?
8XA78C @ 1.7 to 2.1GZ
Handheld Mode GPU: 1.8 - 2.0 Teraflops
Docked Mode GPU: 3.0-3.5 Teraflops
12GB of RAM @ 102GB/S (10GB for gaming, 2GB for the OS)
256GB or 512GB Storage @ 1GB/S
Would it be possible to run current gen games?
darthdiablo
+5 Death Stare
- Pronouns
- he/him
By "current gen" games do you mean PS5/XBX games?
All of them? Probably not.
At least a good number of them? Probably.
Has to be optimized for Switch 2 hardware tho, if the game is demanding in resources.
Absolutely, but it would also depend on the specific bottleneck on the specific game.
Adulfzen
Like Like
- Pronouns
- he/him
I think no matter what the specs look like on paper, there will be some surprising ports, some suprising absences and overall good third party support that would still be behind playstation and xbox but probably way better than previously by comparison of Switch 1.
I'd suspect a decent bit more Japanese studios support right off the bat, hopefully.
Also we kinda have to never underestimate the initial cost of porting a whole ass engine on a brand new platform like for example the Dragon Engine for Yakuza.
There are also some games that feels like big impossibilities such as Alan Wake 2 because of how far they're seemingly pushing their tech at Remedy if the pc requirement specs are anything to go by (tho we'll need to see how the console versions pan out).
Giancarlo
Nintendo connoiseur
PS5,Xbox Series X has reached a point you can do games that nearly real live, for now on, small details like better hair/water details will be the highlight of next consoles generation, dont expect a massive leap such as 2D to 3D, or SD to HD
Νεφέλωμα
Switch 3 enthusiast
- Pronouns
- He
I would expect early PS5/XSeries ports, like how we got early PS4/XOne games on the Switch. Maybe the Switch 2 gets 1 or 2 years of third party parity with other consoles. But after that, there will only exist miracles.
snoota
Rattata
- Pronouns
- He/Him
IMO the main bottleneck for future ports that Drake faces would be in the form of CPU intensive titles that prove to be a nightmare for even the home consoles. Even with a benefit of a node shrink & the A78s having the benefit of an IPC lead over Zen 2, the clocks required to keep all 8-cores (Or 7? Assuming they let the one OS core clock down to oblivion for efficiency) would be pretty limiting. Memory bandwidth is also a concern, but it shouldn't as absurdly limiting compared to what the TX1 dealt. From a design perspective, it really does feel like T239 was intended to squeeze out as much perf/watt as possible (especially in regards to GPU) without pushing the power budget far too past last gen, & these compromises I listed aren't as game breaking as my previous statements would suggest.
SiG
Chain Chomp
Likely it will be the same as before, with company politics (and moneyhatting) mostly determining releases.
darthdiablo
+5 Death Stare
- Pronouns
- he/him
I don't think Nintendo needs to be convinced of rendering features like ray tracing. though if they absolutely want 60fps, then it might be the first thing that's cut. but I'm betting the house on the next Zelda making extensive usage of it
The screen recording is handled by a separate block, NVENC. The current Switch already has that one.
Giancarlo
Nintendo connoiseur
Nintendo can say no to Nvidia, they are paying Nvidia to design the SOC of Switch sucessor,if Nintendo dont think Ray-Tracing is worthy for it next hardware, they will tell dont put this on my next hardware
ItWasMeantToBe19
Manakete
ItWasMeantToBe19:
People being so mad at me that they’re buying an obviously very false report (who is turning off RT in handheld mode only) is weird.
yes, they can, but why would they? RT cores don't cost a lot of silicon and Nvidia can show it can runs well in low power environments. so Nintendo would use it like they used every other modern rendering feature. all these other low-powered devices from Mediatek, Qualcomm, and Samsung show that Nvidia was right on the money
why is it obviously false, or rather, not a safe prediction?
ItWasMeantToBe19
Manakete
What Switch 1 games turned down or off visual features other than framerate and resolution btw, were there any?
- Pronouns
- He/Him
I know, the OFA is supposed to also be in that block.
Mortal Kombat 1 for a recent example. turned off damn near everything
TLZ
Like Like
There are a few posts I read, yes. 4k+RT this and that. All I'm saying is, don't overhype yourselves. It'll only bite you in the backside later. I'd rather temper my expectations. Have to remember it's a portable first and foremost. There's only so much it can do, and I'll be content with what the "little" thing can do.
ItWasMeantToBe19
Manakete
What is the developer logic to go
"I am going to use RT instead of baking on the Switch 2 to save time in development even though it will use up like >10% more of the Switch 2's GPU and some chunk of the CPU as well"
"And then in handheld mode, instead of turning the pixels down by 2.25x, I'm going to drop RT and bake the lighting for that version."
What resolution would they even be targeting in docked mode for this to make sense.
there is logic there when they need to claw back more performance. but as necrolipe mentions, those devs don't see the need to turn off RT to get their desired performance
TLZ
Like Like
Thanks for the unnecessary label
ItWasMeantToBe19
Manakete
I’m sorry, but if you’re building your game around RT for convenience and then have to end up baking the lighting in handheld mode, that doesn’t make a ton of sense. You’re just doing a lot of extra work for no reason. Just bake it from the beginning.
Who is throwing out tons of other effects from their PS5 version for RT, only to then also bake it in handheld mode.
Concernt
Optimism is non-negotiable
- Pronouns
- She/Her
4K with significant RT usage AFTER upscaling and reconstruction would not be out of the question for the device, but I don't expect many games to take the additional development/optimisation time and likely significant hit to image quality to achieve such.
Like with any other visual feature or target, it's a balancing act. On Nintendo Switch's successor, as with Nintendo Switch, it's going to an extremely TIGHT balance. Some, undoubtedly, will achieve it regardless.
LoneRanger
Chain Chomp
- Pronouns
- He/Him
I think lot of people is playing the low expectations here but I think this time if X PS5/XBSeries game doesn't come to Switch NG is a matter of politics between third party publisher and Nintendo.
It doesn't have to do with anything relate to HW.
It doesn't have to do with anything relate to HW.
yes, it doesn't make much sense. you're better off ditching the RT and leaving shit raw.
but none of this is what's being said, so I don't know why this matters
Last edited:
