[HN Gopher] Cortex A57, Nintendo Switch's CPU ___________________________________________________________________ Cortex A57, Nintendo Switch's CPU Author : rbanffy Score : 165 points Date : 2023-12-13 15:11 UTC (7 hours ago) (HTM) web link (chipsandcheese.com) (TXT) w3m dump (chipsandcheese.com) | tiahura wrote: | As my 6 year old makes his way through Fortnite Season 5, I'm | impressed that his Switch still handles his no-scope sniper | odyssey on these 8 year old Tegra X1s. | gjsman-1000 wrote: | It's impressive, but it's also somewhat amazing to consider | that the Tegra X1 is still about 3.5-4x as powerful as the GPU | in the Raspberry Pi 5 - even if any iPhone chip since the A11 | Bionic would beat the X1. | MBCook wrote: | We'll never see it, but I've wondered what a Switch would | look like with an Apple chip. | | Take something relatively modern but no longer too expensive, | say an A15 like the Apple TV has, bump the screen to 1080p, | and I bet it would scream. Possibly with better battery life. | | Nintendo would never do that. I doubt Apple would either. But | it would be a very interesting test. | drewzero1 wrote: | I've been playing Skyrim and Portal 2 on the Switch Lite and it | keeps amazing me that games that my PS3 struggled with are now | perfectly playable on a handheld device that has pretty good | battery life and doesn't burn my hand. A device like this would | have blown my mind 10 or 15 years ago (even more than the | original PSP did the first time I saw one). | fishtacos wrote: | The PSP was amazing. The PS Vita - on a whole 'nother level, | both performance-wise and controller improvement. It wasn't | until the Switch came around in 2017 (purchased in 2018) that | we realized HD (720p) gaming on a handheld machine. The fact | that so many AAA games have been ported down to it is a | testimony to its capabilities. Not a fan of nVIDIA in | general, but they did an amazing job with this unit. If one | wants real PC gameplay without porting down, the Steam Deck | starts at $400, add a MicroSD card for more storage. | hypercube33 wrote: | What really blows me away is stuff like the z1 extreme and | a laptop I have that is two generations behind that but has | a 130w power envelope pushing 60-100fps AAA titles at 1440p | - runs super hot but in a few years we may have handhelds | doing something similar which I'm super excited for. | fishtacos wrote: | There is nothing that can be encompassed by a 220-300+ | power envelope into a handheld for the next several | years. Still, downgrading visual FX is one doable. Steam | Deck 2 and Switch 2 are in the horizon. AMD v. Intel, as | far as I'm concerned. | zeusk wrote: | Wait till you see what Apple's Vision group has in store. | graphe wrote: | The vita sucked with no homebrew and the switch is another | psp. You could run games off the pro duo or a microsd | adapter but vita was locked down, slow and a bunch of lame | remakes and psp titles. The PSP emulated up to the PS1. | | By the time the switch was around people had been emulating | on Android for a long time. They can do better graphics but | there hasn't been a game that made me feel like I needed 4K | on handheld. The switch really is another golden age for | gaming with the switch being hacked so quickly and having | such good homebrew. | fishtacos wrote: | The PS Vita predated the Switch by 6+ years... by the | time emulation and CPU speed to match its requirements, | it was already discontinued by Sony. It fulfilled its | purpose. | graphe wrote: | The vita was a direct downgrade from a PSP if you valued | fast loading times, emulation, homebrew and piracy. Even | if you didn't pirate you gained the ability to rebuy your | PSP games, and new expensive games with an OLED that had | less battery life. | fishtacos wrote: | A downgrade from UMDs? Perhaps you're referring to the | density of textures and newer, more demanding game | engines, because the UMDs were such a pain. Trying hard | not to get hyperbolic here. | danhor wrote: | At least regarding the homebrew, I disagree completely. | While it took a while for the vita to be thoroughly | hacked, it has been thoroughly hacked. And the benefits | are numerous: Using normal SD cards, expanding the the | integrated psp hardware into a virtual psp, reformatting | of the internal storage, extending it by replacing the 3G | modem. | | Meanwhile the switch had the big bootrom usb stack | exploit, but everything apart from the original SoC | doesn't have a publicly known easy exploit (there are mod | chips, but nothing like the 3DS/PS3/Wii U/Vita/PSP/...). | There also wasn't that as much "cool" stuff to do as on | older consoles with homebrew due to the hardware simply | being an android tablet with controllers (which doesn't | make a difference as a console, but makes it more boring | homebrew wise). So there is the usual stuff (savegames, | different controllers, piracy, themes, overclocking), but | nothing unique to the switch. | graphe wrote: | By the time it was thoroughly hacked, it was too late. | Much better hardware was around. People could do most of | what the vita was eventually able to do with a phone. | | Besides piracy what do you expect from the switch? The | PSP had an ebook reader, movie player and could play mp3s | as well as other cool old games very early in it's | release. It could play media and play games up to the | ps1. Modding a PSP vita today is like maxing out a citron | 2cv instead of buying an e scooter. When the PSP came out | it was amazing. Now it's yesterday's news. I'm impressed | by the hackers that did it but the vita just doesn't | impress even with hombrew today for the capabilities. | | The PSP had amazing battery life and felt like a better | Gameboy advance at the time. | fishtacos wrote: | I wish the person I sold my fully-loaded 1/2 TB MicroSD | Vita to a grand adventure with a pocketable | gaming/homebrew/emulation/piracy machine. Nothing matches | it in tis market given its capibilities. The Switch lite | tried, but I ain't touching that crap without HDMI output | and no mods available. | goosedragons wrote: | The Vita had a ton of amazing JRPGs, niche weeb stuff, | visual novels and indie games. Bit of a weird lineup but | for many years it was my most used system. These days the | Vita has been cracked wide open and there's loads of | homebrew. | | Right now it's only the early Switch units that are | hackable. | graphe wrote: | Did you have a PSP? The vita was such a downgrade from it | and the vita like the PSP mostly had remakes, except the | vita had psp remakes too. It had no killer game or | multimedia capabilities by the time it released. I can't | name an exclusive on there at all, I don't even know if | it had one. | goosedragons wrote: | Yes lol. I had a PSP shortly after launch. I still have | one. And yes, the Vita had games and exclusives. Most of | its launch lineup was exclusives, Uncharted Golden Abyss, | Wipeout 2048, Hot Shots Golf World Invitational, etc. | Even for just playing PSP games, the Vita is better. OLED | display+ the ability to remap the right stick to the | D-pad or buttons. Makes games like MH: Freedom Unite | waaay better. | | And a lot of indie games, JRPGs, Visual Novels just fit | the handheld form factor better than PC or PS4 even if | they weren't exclusive. | Grazester wrote: | Those games ported to the switch are severely downgraded | graphically compared to their other consoles counter parts. | fishtacos wrote: | My laptop runs Kbby Lake and Intel HD 620 and can't match | the Switch's 30 fps perf. It's incredible what it can do. | The 3070 that runs 220+W on my desktop is hard to replace | ore minituarize. | Grazester wrote: | Ehh one is a gaming machine the other a general purpose | machine with a really weak GPU. | bigstrat2003 wrote: | Frankly, we still have nothing providing HD gaming on a | handheld. The switch is way too big to be considered a | handheld. I can't slip one into my pocket and go, like I | could with a DS or 3DS. It's frustrating, because I love my | 3DS for gaming on the go but there is still nothing which | can replace it. Nintendo just gave up on the handheld | market. | glhaynes wrote: | A Switch Micro would be pretty cool. | fishtacos wrote: | HD at 720p is more than enough for a 6" screen. The | Switch is literally the only capable handheld that could | do that and still does. It doesn't offer excellence, but | it offers good enough. As already written, the Steam Deck | replaced it a half a decade later. Not sure what your | point is. Everything grows generatially in capability and | capacity, including handhelds. I would not compare the | Gameboy to the Gameboy Advance under such limitations. | Teever wrote: | > The switch is way too big to be considered a handheld. | | This seems like a pretty subjective judgment. Was the | Sega Game Gear not a handheld console? | hbn wrote: | The true testament to what can be handled on a mobile device | these days is showcased on iOS devices. The iPhone 15 Pros | can run a port of Resident Evil 7 at supposedly a pretty | stable 30fps. | | Unfortunately all of Apple's chips are stuck in Apple | devices, where they're still struggling to incentivize | developers to do ports as standard. So they're pumping out | beautiful graphics to beautiful displays for games with | shitty touch screen controls that are riddled with ads and/or | are just glorified virtual casinos. | sp332 wrote: | Sure but the iPhone 15 Pro costs 3x as much and came out | years later. | BD103 wrote: | Other AAA games like Tears of the Kingdom also having stunning | graphics. It's quite impressive, though only rendered in 720p. | MBCook wrote: | Isn't it 1080p when docked? | monocasa wrote: | It's a dynamic resolution in both docked and undocked modes | since it appears to be mainly memory bandwidth limited. It | does max out at 900p when docked though. | DeathArrow wrote: | > I wonder if similar optimization efforts could be carried out | to make modern games accessible to a wider audience. | | I wonder, too. | monocasa wrote: | > The SoC also contains a cluster of four A53 cores for power | efficient processing, but Nintendo has chosen not to use them. | | The rumor I've heard is that there's a bug in the system crossbar | which makes which core CCX you enable after reset the only choice | you can make until the system is fully reset. That is, if you | enable the A57 CCX, later enabling the A53 CCX triggers the bug | and vis versa, even with the first CCX disabled when enabling the | second. | moffkalast wrote: | Really weird for Nintendo to have picked such a completely | broken SoC. | monocasa wrote: | Probably got a really good deal on it. | gjsman-1000 wrote: | The Tegra X1 was also originally, rumoredly, not meant for | Nintendo. | | https://nvidianews.nvidia.com/news/nvidia-launches- | tegra-x1-... | | It was going to be for "Deep Learning," "Computer Vision | Applications," "NVIDIA DRIVE car computers," and robots. As | we know, outside of some Tesla models, that didn't really | happen. | | Rumoredly, according to people who have read the leaked | Nintendo documents (Modern Vintage Gamer has implied it in | replies to comments on his videos), NVIDIA had found the | bug with the recovery mode before the Switch's launch; but | Nintendo couldn't just move the announced Switch launch | date to wait for a chip revision, especially after the Wii | U financial performance. So, off it went and they just had | to cross fingers and hope nobody found it. Nintendo | probably got a good discount for that mistake too. | parl_match wrote: | > As we know, outside of some Tesla models, that didn't | really happen. | | As YOU know. | | The Tegra X1 (which the switch used) was never used in | any production automotive application, correct. But you | mentioned Tesla, so let's talk about other Tegra | generations. | | Other Tegra generations were used in Teslas in varying | quantities (Tegra 2, 3, and K1). Mercedes has been | shipping Tegra in their "MBUX" cars for a few years now | as well. A couple of Chinese companies are shipping Tegra | via NVD. Volvo, Land Rover, and Jaguar are also going to | be shipping it shortly as well. | gjsman-1000 wrote: | I don't know why you accuse me as though I said something | wrong, and then immediately admit that, yes, the Tegra X1 | was never used in any production automotive application. | That's what I was talking about - not Tegra as a whole. | The only mistake I made was that the Tegra X1 wasn't used | in any Tesla models. | mbf1 wrote: | This is news for nerds. Precision is important! So they | fixed your statement. I found it interesting. | extraduder_ire wrote: | A whole lot of the switch's design was about driving down | costs as much as they could. That's why it has no mic, | camera, and didn't get bluetooth headphone support for | years after launch. Plus chipmakers _really_ want those | console contracts, even if they make a loss on the first | few hundred thousand sold, they 're assured sales as long | as the console sticks around. AMD made out like bandits | with the ps4/xbone. | m-p-3 wrote: | And they're likely making a killing with the Steam Deck. | entropicdrifter wrote: | and soon will be with the PS5 and XBS(S/X) as well, if | they aren't already | Rapzid wrote: | Bluetooth audio is kinda ass for gaming anyway unless you | have low latency codec support on your source and | consumer which is almost never the case. | | Sony has their own such codec which is why the audio jack | on the PS controllers works so well.. | rkangel wrote: | They were probably in on it when it was pre-production. This | is something that you often want to do as a large customer so | that you are as close to the state-of-the-art as possible, | but it comes with downsides! I am doing a similar thing on a | project at work at the moment. | phonon wrote: | The chip shipped 2 years earlier than the Switch in the | https://en.wikipedia.org/wiki/Nvidia_Shield_TV | | More likely Nintendo got a really good deal on it. | rkangel wrote: | Yeah, but how long was the Nintendo Switch in | development? And bear in mind that we're comparing it to | a much simpler bit of Nvidia electronics. | | It is entirely possible that it was just price based and | they didn't care about the chip bug, but given the | timings I still think they would have selected the chip | before it was complete and in consumer products. | gjsman-1000 wrote: | Well, I think it's important to consider the competition | in 2017 was not great. | | The comparable competitive Android chip, also launched in | late 2015, was the Qualcomm Snapdragon 820. That chip was | widely known as being one of the worst Qualcomm chips | ever made; for mediocre power efficiency, lots of heat | generation, thermal throttling, and a buggy first attempt | at 64-bit instructions. All that for a GPU that is, on | paper, significantly weaker [1] (though, maybe the | Switch's cooling could've helped close that gap a bit). | But even then, you're dealing with Qualcomm, and everyone | knows they are just the worst. | | First, because Qualcomm loves royalties based on the | device's MSRP, rather than a flat charge per chip. | Nintendo probably wouldn't like that. Secondly, while | NVIDIA GPU drivers are a proprietary blob, that's of | little concern to Nintendo, and that blob can be easily | adapted to run on any OS under the sun, including their | own. Qualcomm - enjoy a hackneyed Linux fork, that's the | best you'll get. From our perspective they're both pretty | bad, but from Nintendo's perspective trying to add | support to their custom microkernel Switch OS, one's | clearly garbage. | | Outside of Qualcomm... what else do you have for 2017? | Exynos and MediaTek? I think it goes without saying... | there are no upsides to passing on the Tegra X1 for a | MediaTek from that era. | | [1] Edit: I previously said 50% and 100% weaker, but | that's very grammatically ambiguous; and FLOPs are a very | bad metric of performance, because there are 3 different | kinds of FLOP metrics floating around that aren't | comparable (due to different levels of precision). | Combined with the Tegra being designed for cooling and | the Qualcomm designed for no cooling, it's hard to tell | specifically how large the gap is, even though a gap is | almost certainly there. I think my point still stands. | my123 wrote: | That was the SD810. The SD820 was a lot better | my123 wrote: | By 2014, the Tegra X1 was already picked as the Switch | SoC. | | From digging at history threads: The alternative SoC | option they had was a quad-A53 SoC with Decaf (a Wii U | GPU cut in half with Wii backwards compat gone) co- | designed with STMicro. | IntelMiner wrote: | Is there anywhere one can read about those in more | detail? | monocasa wrote: | I'm not sure about reading, but this was an interesting | video on the topic linked elsewhere in this thread. | | https://www.youtube.com/watch?v=MZzXidHVvJU | CopperWing wrote: | Nintendo chose Nvidia Tegra X1 when its previous design | based on another custom ARM-based SoC failed: | https://www.youtube.com/watch?v=MZzXidHVvJU | hypercube33 wrote: | Rumor was nVidia was selling them at a discounted price since | the only other thing (I am aware of) is the Nvidia shield | using these chips so they went with them on that factor. | polpo wrote: | Gunpei Yokoi's ethos of "lateral thinking with withered | technology" [1] has been an guiding principle within Nintendo | since the Game & Watch days: to use "seasoned" or otherwise | imperfect technology in creative ways. Using a broken SoC | that they probably got a great deal on that still fits their | needs fits perfectly in that mindset. | | [1] https://medium.com/@adamagb/nintendo-s-little-known- | product-... | gjsman-1000 wrote: | This is additionally odd because according to Wikipedia, citing | the Technical manual (wherever you can find that), the A53 | cores were so borked that later versions of the manual removed | all references of their existence. The Tegra X1+ shipping in | all Switches since 2018-2019 might, possibly, not even have the | cores. | colejohnson66 wrote: | Does anyone know what this "bug" manifests as? Your first | sentence implies it's a lockout created at reset, but your | second implies it manifests post-reset through user/programmer | action. | toast0 wrote: | I think they're saying you get to run one CCX at a time; | either the A57 cores or the A53 cores and you can only change | that once after a reset. Presumably the boot time CCX is | selected by selection pins. Depending on exactly what reset | means, this might be something like the i286 that couldn't | leave protected mode without a reset, so systems were built | for OSes that entered protected mode where they could set up | a reset vector and then cause a reset and jump back to their | real mode kernel. Or, it might be a very intrusive reset that | results in memory contents being reset --- that's not going | to be something worth engineering around. Either way, it's | easier to just only use the A57 cores to begin with, and | especially if switching between the two is problematic. | colejohnson66 wrote: | _Assuming_ it is some "at reset" selection, Nvidia | advertising it as an eight-core chip would be deceptive, so | I'm reluctant to believe such a theory. I wouldn't put it | past marketing to do such a thing, but later revisions of | the manual don't mention the A53 cores, so I'm inclined to | believe it's a hardware bug. | | Unfortunately, without someone from inside Nvidia telling | us, all we have are rumors and no evidence. | toast0 wrote: | ARM big.LITTLE systems started out as use one or the | other, but not both simultaneously. Advertising those as | 8 core when it's really big 4 or LITTLE 4, or later | incantations where you could use the big or LITTLE of | each of the four paired processors, is sketchy, but was | common. | | If the plan was to allow big 4 or LITTLE 4, and then a | hardware bug became apparent that you could switch to | LITTLE cores but not back to big cores, well you notify | customers and stop advertising the LITTLE cores. | archive.org has them mentioned Jun 1 2016 [1], and then | removed Jun 14, 2016 [2]; the A53 cores aren't mentioned | but are shown in the die map on the current page [3]. | | [1] https://web.archive.org/web/20160601063237/http://www | .nvidia... | | [2] https://web.archive.org/web/20160614200203/http://www | .nvidia... | | [3] https://developer.nvidia.com/content/tegra-x1 | monocasa wrote: | Yeah, I'm saying that I've heard that there's some bug | that's only hit by running one CCX after you've run the | other one. | | > Presumably the boot time CCX is selected by selection | pins. | | There's actually another core, an ARM 7 referred to as the | BPMP (Bootstrap and Power Management Processor IIRC), that | handles main CCX bring up. | SillyUsername wrote: | Can anybody answer why the author wrote "indirect branches tend | to show up in object oriented languages"? | | Given that branches are, well, just branches in any language, | what makes OO so special? | | Also further down the author states "while FP registers have to | be wider to handle vector execution" | | Again I'm pretty certain FP registers are larger owing to the | greater precision they have, not specifically because they're | designed for vector ops... please somebody explain why my | understanding is wrong? | NovemberWhiskey wrote: | > _Given that branches are, well, just branches in any | language, what makes OO so special?_ | | I suppose the assumption is that you get lots of virtual | functions. | dwaite wrote: | > Can anybody answer why the author wrote "indirect branches | tend to show up in object oriented languages"? | | While useful for function pointers and jump tables, I suspect | they are speaking to polymorphism and vtables/witness tables. | kevingadd wrote: | Indirect branches are different from regular branches. A | regular branch is 'goto label', while an indirect branch is | like calling a function pointer or calling a virtual method | jchw wrote: | > Given that branches are, well, just branches in any language, | what makes OO so special? | | _Indirect_ branches. This is a result of vtable indirections. | | You are much more likely to encounter vtables in an object | oriented language. Obviously, you can still have the same basic | thing in a C program, e.g. SDL RWops, but in C++ for example, | it's going to show up all over the place. | llm_nerd wrote: | Virtual functions yield a lot of indirect branches. Virtual | functions are a foundational part of object oriented designs. | | FP registers have gotten much larger than the normal types that | people store. e.g. 128, 256, 512-bit registers. A normal double | floating point (pretty much the largest normally used floating | point representation) occupies 64-bits, while a normal int64 | occupies, unsurprisingly, the same 64-bits. But we're getting | the mega registers specifically because there are a lot of | multiple-four-singles at once, and so on, SIMD functions. | monocasa wrote: | > Also further down the author states "while FP registers have | to be wider to handle vector execution" | | > Again I'm pretty certain FP registers are larger owing to the | greater precision they have, not specifically because they're | designed for vector ops... please somebody explain why my | understanding is wrong? | | ARM Neon is both 128bit SIMD and the FPU for the system. | There's not a separate FPU from the SIMD. | fulafel wrote: | Indirect branches are common in many OO languages because | calling object.method(arg) essentially does | object.class.method(object, arg) or | object.prototype.method(object, arg) - the address of method is | loaded indirectly through the object's "class" field as it may | be inherited or not. | | (In some cases the compiler may statically know the class of an | object, if it's not allowing for subclassing and a potentially | overridden method) | crtified wrote: | It's interesting that, on long running and well regarded (but | obviously, not authoritative) site "HG101's Top 47k Games of All | Time" [0], the top 50 ranked titles includes virtually no games - | 1? 2? 3? arguably - of a technically more advanced pedigree (than | the Switch's capabilities), in terms of applied processing grunt, | 'graphical fidelity', etc. | | Expand that analysis further, to the top 100, and it remains | true. | | 'Regard accumulated over time' would clearly be a factor in that | bias, but not a definitive one. Relatively modest processors like | the A57, and (vastly) weaker, are still the home of the majority | of human video gaming enjoyment. | | [0] http://www.hardcoregaming101.net/hg101s-top-47k-games-of- | all... | danbolt wrote: | Broad acclaim requires a broad minspec. | sylware wrote: | Next nintendo console on RISC-V!!! | MBCook wrote: | That would toss away any shot of backwards compatibility | without a second processor or translation layer. | | I can't see how that would be beneficial. | paoda wrote: | A second processor isn't that crazy of an idea for Nintendo | given the 3DS had an ARM11, ARM9, and an ARM7! | | Famously, the 3DS has complete GBA hardware inside of it that | never got used outside of the Ambassador Program. | Rapzid wrote: | It's a well known "secret" at this point what the next chip | will be and it's a new Tegra. Nintendo is all in on DLSS 3.5 | 29athrowaway wrote: | Just a reminder you can take a Nintendo Switch Pro controller and | pair it to Linux via Bluetooth. ___________________________________________________________________ (page generated 2023-12-13 23:00 UTC)