Codename_Steam
Cappy
- Pronouns
- He/Him
If Drake uses Samsung 8nm, how high can it clock?
-
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
HankyPanky
Like Like
- Pronouns
- Funcle
considering we've had exactly 0 new information for a while I'd say any third party titles launching within a potential launch window next year are grounds for rampant speculation!
Concernt
Optimism is non-negotiable
- Pronouns
- She/Her
I think the doomering about Samsung 8nm is unwarranted.
One, Drake has certain minimum clocks where efficiency gets WORSE below them. Whatever that is, plus maybe 5-10% buffer, will be handheld mode's basic clocks. Think Shovel Knight. But that's still tonnes of power, on par with the most powerful mobile phone chips.
Two, these are a significantly smaller concern in TV mode, where it can gobble up plenty of power, so I'd expect raw performance numbers closer to PS4 even there, before DLSS, plus efficiency gains from the Ampere architecture compared to PS4's GCN.
Really I think it'll come down more to battery life- which with battery energy density improved since 2017, shouldn't be a problem. Worst case scenario would be launch Switch battery life for Drake exclusives.
I'm not talking in specific frequencies because I'm not sure how helpful it is, and rather I'm trying to talk in terms of capabilities. I hope that doesn't seem like dodging your question.
8nm Samsung is the worst case scenario, and it's a pretty good case even then. Of course better process nodes would give us slightly higher docked clocks and better (maybe a lot better) undocked battery life, but they're not do or die.
Last edited:
They're definitely pretty big on RISC-V for microcontrollers, but they're sort of actively resisting it elsewhere, with the most notable example being the refusal to provide any RISC-V builds of their GPU drivers. It's worth remembering that they did attempt to acquire ARM recently, and have some pretty hefty licenses as a result of that. They'll pivot away eventually if things move in that direction, but probably not before whatever they're cooking wrt CPUs right now runs its course.
- Pronouns
- He/Him
Short answer: we don't know. Without the specifics of its power consumption compared to that of Orin we can't answer that.
I'll let someone else give the long answer.
- Pronouns
- He/Him
Usually it would be due to size and cost, but we will have to actually see what the reason is with the final thing.
D
Deleted member 887
Guest
Drake's L2 cache is not unusually small. TPC's access the L2 cache, the amount of L2 cache per TPC is well within the bounds of what Ampere usually ships.
If you divide the number of memory controllers by the amount of L2 cache, yes, it looks like the amount of L2 cache is small, but that's not because the L2 cache is small, it's because the number of memory controllers is large. Drake has a wider memory bus relative to the size of the GPU than any other Ampere card.
The most likely reason for the wide bus is just that Drake isn't a GPU, it is a complete SOC. There is a single shared pool of RAM for the CPU and the GPU that services both, but the CPU has its own L2 cache distinct from the GPU's pool.
The other possible reason is simply that Drake is smaller than a single, full GPC, and some parts of the design simply don't scale down any further.
Look over there
Bob-omb
Well, that's stagnation.
Frederick Chen on LinkedIn: At IEDM this year, TSMC's SRAM cell is only 5% smaller at 3nm than 5nm… | 19 comments
At IEDM this year, TSMC's SRAM cell is only 5% smaller at 3nm than 5nm (0.0210 => 0.0199 um2)! | 19 comments on LinkedInwww.linkedin.com
For a bit more context... from this page:
Hmm, IIRC, according to one of the tables in the latest IEEE roadmap, we shouldn't really expect significant gains with sram until stacking gets figured out...
Actually, looking at table MM-10 on page 14 (although it's the 20th page in my reader) of the More Moore pdf, there's one more significant shrink in sram expected in the late 2020's, and then after that one, things come down to stacking.
So, currently I'm not expecting significant gains with sram through the mid 2020's.
strokes chin
This is veering away from Nintendo, as I think what's more impacted by this particular topic in this time frame is actually GPUs.
As of Samsung's roadmap given this past October, GDDR7 should probably start being made in 2024. Prior to that, there is a stopgap solution of stacking two GDDR6 chips at a time.
GDDR7 should offer roughly double the bandwidth of regular GDDR6. As the numbers go, Samsung currently states 36 GT/s for GDDR7. Regular GDDR6 typically varies from 14 to 18 GT/s, although wikipedia says that the RX 7900 cards are using 20 GT/s, huh. Anyway, that doubling of bandwidth should be sufficient for the 2024-2025 generation of consumer GPUs (potential hike in prices from just memory alone aside). The question is, what's the solution for the bandwidth needs of the 2026-2027 generation? There is HBM4 due by 2026, but whether that's economically doable...
Edit: Ooh yea, for the readers, I should probably explicitly state the connection between sram and GDDR & GPUs here.
SRAM = cache
Cache = a way to try to cut down reliance on sheer memory bandwidth to keep your cores fed with data and instructions
So, the question is, what do you do if you hit a wall in terms of cranking up memory bandwidth and a wall in expanding cache, but your needs are still growing?
Last edited:
Concernt
Optimism is non-negotiable
- Pronouns
- She/Her
While not a perfect solution, the PS5 approach to this was to throw a blazing fast SSD in it. And blazing fast it is, the PS5's SSD has over half the speed of the Switch's entire bus at a theoretical max in the 15-16GB/s range to Switch's bandwidth of 25.Well, that's stagnation.
For a bit more context... from this page:
Hmm, IIRC, according to one of the tables in the latest IEEE roadmap, we shouldn't really expect significant gains with sram until stacking gets figured out...
Actually, looking at table MM-10 on page 14 (although it's the 20th page in my reader) of the More Moore pdf, there's one more significant shrink in sram expected in the late 2020's, and then after that one, things come down to stacking.
So, currently I'm not expecting significant gains with sram through the mid 2020's.
strokes chin
This is veering away from Nintendo, as I think what's more impacted by this particular topic in this time frame is actually GPUs.
As of Samsung's roadmap given this past October, GDDR7 should probably start being made in 2024. Prior to that, there is a stopgap solution of stacking two GDDR6 chips at a time.
GDDR7 should offer roughly double the bandwidth of regular GDDR6. As the numbers go, Samsung currently states 36 GT/s for GDDR7. Regular GDDR6 typically varies from 14 to 18 GT/s, although wikipedia says that the RX 7900 cards are using 20 GT/s, huh. Anyway, that doubling of bandwidth should be sufficient for the 2024-2025 generation of consumer GPUs (potential hike in prices from just memory alone aside). The question is, what's the solution for the bandwidth needs of the 2026-2027 generation? There is HBM4 due by 2026, but whether that's economically doable...
Edit: Ooh yea, for the readers, I should probably explicitly state the connection between sram and GDDR & GPUs here.
SRAM = cache
Cache = a way to try to cut down reliance on sheer memory bandwidth to keep your cores fed with data and instructions
So, the question is, what do you do if you hit a wall in terms of cranking up memory bandwidth and a wall in expanding cache, but your needs are still growing?
- Pronouns
- He/Him
This is… incorrect.
Look over there
Bob-omb
That approach is more about addressing a different issue, I think. That issue is more of a 'unsatisfactory amount of RAM' thing.
The way I see that, it's like, oh, there's not enough RAM to hold everything I'll be needing soon. So instead, I'll just replace what I'm holding in RAM extra quickly using this fast storage.
Now, you're not necessarily wrong to mention the PS5; the slowdown in improvements in memory density is another issue. It's just that what I was talking about in my previous post is more about the transfer of data from RAM to cores. Storage is not involved in that part.
And quick correction: PS5's SSD theoretical effective max speed refers to getting data from storage to RAM. The Switch's 25.6 GB/s bandwidth number refers to RAM <> CPU/GPU. So, the numbers are referring to two different things.
Back to bandwidth concerns, it did occur to me that yes, if there's a stopgap solution of stacking two GDDR6 chips at a time, then why not stack GDDR7 chips. That'll probably happen and I'm not going to be happy about the money side of that.
Forgive me if this feels like I'm repeating someone else, but I feel that if TotK was to release with no Drake in sight, they would have probably advertised it more to get more sales of Switch hardware, especially during this holiday season. But if Drake were coming soon, I'd imagine that they would advertise TotK running on it with a big trailer. That creates a problem because we are in the holiday season, and why would Nintendo want to ruin existing Switch sale potential by advertising that now? It was no issue showing off the Switch back in Oct 2016 because they really didn't have much else. Wii U was already dead, and 3DS was sort of limping along.
cw_sasuke
Octorok
The more i think about it the more i feel not releasing Drake in the first half of 23 would be a big mistake in terms of mindshare and momentum. There are just too many big games releasing that could potentially be on a Nintendo system as well - why push your audience towards other devices and platforms when you have a system capable playing these games ?
This would at least be my opinion if DevKits has been around for 18-24 months and Nintendo could secure that mayor games like Diablo 4, Street Fighter 6, MK 12 and co. would launch alongside the other versions.
If the launch is mostly gonna be accompanied by older games like RDR2 or previous RE Engine games, i guess the timing wouldnt be as essential which is why im hoping that Nintendo can be part of the next-gen transition that is gonna start next year. Timing would be perfect.
This would at least be my opinion if DevKits has been around for 18-24 months and Nintendo could secure that mayor games like Diablo 4, Street Fighter 6, MK 12 and co. would launch alongside the other versions.
If the launch is mostly gonna be accompanied by older games like RDR2 or previous RE Engine games, i guess the timing wouldnt be as essential which is why im hoping that Nintendo can be part of the next-gen transition that is gonna start next year. Timing would be perfect.
Everything seems to be perfect for a release in the first half of 2023
this is why Nintendo won't release anything until 2050
Kenka
Like Like
- Pronouns
- He/Him
It's a good point indeed.
D
Deleted member 645
Guest
Anything new in here? Man, the wait for this new thing has been crazy.
Other than speculation? Nope.
You can check the OP, any updates will be added to it.
- Pronouns
- He/Him
The timing can be a sweet as honey, smooth as a hot knife through butter, a wire through styrofoam, a katana slicing through a hair strand lengthwise.
Nintendo on the idea of launching it then:
launch is not as ideal, it is imperfect, they get a Wii U/early 3DS and then end up with this face:
Third Parties when they see that they didn’t actually take their concerns into serious consideration, non tech Nintendo fans when Nintendo does whatever they want, again:
Me, disappointed or dissatisfied but not shocked at all because they are slow to react, very slow and the outcome is the result of their own deeds and actions, again:
Nintendo, switch collapsed due to people trying to move on to the 2, but also the switch 2 doing worse than expected, software ain’t as strong either:
Did I miss anything
Oh right, people will defend it yada yada, “go play on another system this isn’t the system for you clearly” or some garbage like that, someone actually said something akin to that in this thread. That people here should move on from Nintendo.
Last edited:
poochy
Big Wario Fan
- Pronouns
- she/her
is this a veiled cry for helpThe timing can be a sweet as honey, smooth as a hot knife through butter, a wire through styrofoam, a katana slicing through a hair strand lengthwise.
Nintendo on the idea of launching it then:
launch is not as ideal, it is imperfect, they get a Wii U/early 3DS and then end up with this face:
Third Parties when they see that they didn’t actually take their concerns into serious consideration, non tech Nintendo fans when Nintendo does whatever they want, again:
Me, disappointed or dissatisfied but not shocked at all because they are slow to react, very slow and the outcome is the result of their own deeds and actions, again:
Nintendo, switch collapsed due to people trying to move on to the 2, but also the switch 2 doing worse than expected, software ain’t as strong either:
Did I miss anything
Oh right, people will defend it yada yada, “go play on another system this isn’t the system for you clearly” or some garbage like that, someone actually said something akin to that in this thread. That people here should move on from Nintendo.
- Pronouns
- He/Him
Girl go somewhere else, why are you doing this
^^^^^^ this is spot on whenever Nintendo launches something new.The timing can be a sweet as honey, smooth as a hot knife through butter, a wire through styrofoam, a katana slicing through a hair strand lengthwise.
Nintendo on the idea of launching it then:
launch is not as ideal, it is imperfect, they get a Wii U/early 3DS and then end up with this face:
Third Parties when they see that they didn’t actually take their concerns into serious consideration, non tech Nintendo fans when Nintendo does whatever they want, again:
Me, disappointed or dissatisfied but not shocked at all because they are slow to react, very slow and the outcome is the result of their own deeds and actions, again:
Nintendo, switch collapsed due to people trying to move on to the 2, but also the switch 2 doing worse than expected, software ain’t as strong either:
Did I miss anything
Oh right, people will defend it yada yada, “go play on another system this isn’t the system for you clearly” or some garbage like that, someone actually said something akin to that in this thread. That people here should move on from Nintendo.
I also love the nintendoom and they should just go third party and make games for sony and microsoft like sega...
Last edited:
OP
OP
- Pronouns
- He/Him/His
(this is for the lurkers who don't know what RISC-V is and are too afraid to ask)
Despite my attempt at parsing the language used in the Wikipedia article trying to sus out the importance/relevance of RISC-V (of which, I only understand it's pronounced 'risk five'), would anyone be able to ELI5 what RISC-V actually is, and the comparison benefits to ARM?
The only things I can seem to understand with its relevance in this thread is that RISC-V is a type of computer processing chip, and that it's open source software (making it easier not having to worry about licensing fees). Would this be the kind of processor people are gauging their clockspeeds on based on whatever the #nm SoC gets used? Does this affect backwards compatibility? Will it provide smoother gameplay? Would this be paired with the tensor cores to enable better DLSS implementation?
(praying I used even half the terminology correctly
)
Despite my attempt at parsing the language used in the Wikipedia article trying to sus out the importance/relevance of RISC-V (of which, I only understand it's pronounced 'risk five'), would anyone be able to ELI5 what RISC-V actually is, and the comparison benefits to ARM?
The only things I can seem to understand with its relevance in this thread is that RISC-V is a type of computer processing chip, and that it's open source software (making it easier not having to worry about licensing fees). Would this be the kind of processor people are gauging their clockspeeds on based on whatever the #nm SoC gets used? Does this affect backwards compatibility? Will it provide smoother gameplay? Would this be paired with the tensor cores to enable better DLSS implementation?
(praying I used even half the terminology correctly
)
OP
OP
Frederick Chen on LinkedIn: At IEDM this year, TSMC's SRAM cell is only 5% smaller at 3nm than 5nm… | 19 comments
At IEDM this year, TSMC's SRAM cell is only 5% smaller at 3nm than 5nm (0.0210 => 0.0199 um2)! | 19 comments on LinkedIn
IEDM 2022: Did We Just Witness The Death Of SRAM?
IEDM 2022: Did we just witness the death of SRAM? While foundries continue to show strong logic transistor scaling, SRAM scaling has completely collapsed.
fuse.wikichip.org
1, Yes, a process node is basically how small you can make transistors. A ton more to it than that, but it affects performance no matter what tech you are using.Would this be the kind of processor people are gauging their clockspeeds on based on whatever the #nm SoC gets used?
Does this affect backwards compatibility?
Will it provide smoother gameplay?
Would this be paired with the tensor cores to enable better DLSS implementation?
(praying I used even half the terminology correctly
2.
Most likely yes.
3.
Would depend on the individual processor.
4.
No, it would only affect cpu, not gpu.
- Pronouns
- He/Him/His
Thank you kindly! So, in short the more transistors you can cram onto a process node, the better the clocks/performance/battery life/combination or all of the above?
Regarding backwards compatibility, is it a good 'most likely yes' or bad 'most likely yes'?
Anatole
Octorok
- Pronouns
- He/Him/His
RISC-V is an instruction set architecture (ISA), like ARM or x86. An ISA is an abstraction that describes what a hardware implementation of that ISA needs to include and how the programmable interface for that hardware should behave. You’ve identified the main point of interest compared to ARM, which is that RISC-V is open source. The specification is maintained by the RISC-V Foundation, of which many companies are members.(this is for the lurkers who don't know what RISC-V is and are too afraid to ask)
Despite my attempt at parsing the language used in the Wikipedia article trying to sus out the importance/relevance of RISC-V (of which, I only understand it's pronounced 'risk five'), would anyone be able to ELI5 what RISC-V actually is, and the comparison benefits to ARM?
The only things I can seem to understand with its relevance in this thread is that RISC-V is a type of computer processing chip, and that it's open source software (making it easier not having to worry about licensing fees). Would this be the kind of processor people are gauging their clockspeeds on based on whatever the #nm SoC gets used? Does this affect backwards compatibility? Will it provide smoother gameplay? Would this be paired with the tensor cores to enable better DLSS implementation?
(praying I used even half the terminology correctly
There’s a really good talk by professors David Patterson and John Hennessy from Berkeley and Stanford when they won the Turing award together for their work on RISC together. They talk about some of the history of RISC in general, and the RISC-V part of the presentation starts about 46 minutes in:
- Pronouns
- He/Him/His
Appreciate the clarification, thank you!
Is there any consensus yet in this thread on whether the benefits of RISC-V outweigh keeping ARM for this next Switch console, or are we all still waiting for more details?
Thank you for this, I'll find the time to watch this later this week!
- Pronouns
- he/him
I cannot imagine them even considering releasing a console that isn’t backwards compatible
Look over there
Bob-omb
Wait, that's even worse.IEDM 2022: Did We Just Witness The Death Of SRAM?
IEDM 2022: Did we just witness the death of SRAM? While foundries continue to show strong logic transistor scaling, SRAM scaling has completely collapsed.
Quick reminder: N3 (or also now known as N3B), which started volume production this year, is a dead end; there is no direct migration to N3E. Furthermore, according to wikichip, TSMC has stated before that N3E, which starts production next year, will be the basis for all other refinements/variants of the N3 family.
Alright, so N3E's SRAM area is the same as N5. Consequences, consequences...
It shouldn't really affect Nintendo just yet; I don't think that we're expecting above average amounts of cache.
That said, I probably should not expect ARM to update the standard maximum L3 cache again for the DSU anytime soon.
It's not good for the price tag of X3D cpu skus whenever AMD gets around to N3E. Maybe Zen 5c and/or Zen 6.
The monolithic and chiplet based skus will continue to not have L3 cache parity at this rate
For future RDNA cards, there's not much incentive to ever move the MCD down to N3E, huh?
SirShorse
Shorserer Supreme
- Pronouns
- He/Him
I feel like the one thing Drake needs to have is backwards compatibility
D
Deleted member 887
Guest
Short term, we are pretty damn sure that Drake uses ARM. The question is what happens after. Nintendo has had an ARM chip in every handheld since the GBA - they'll move on only if ARM-The-Company puts them in a position where they feel they have to, or if RISC-V makes huge leaps. And RISC-V is most likely to make huge leaps if ARM, Holdings screws a lot of people over, which it would very much like to do.
- Pronouns
- he / him
At some point, though, it can become a Wii U situation where being too tied to the hardware of the past holds everything back. It's way beyond my knowledge to know what kind of middle ground there might be.
- Pronouns
- he/him
I get what you’re saying, but I feel like, in this day and age, back compat is expected. Sony and Microsoft both have basically 100% compatibility on their current machines, and I feel like Nintendo won’t get away with not having it.
I think their best option would be some sort of compatibility layer, not unlike Apple’s Rosetta 2. This is the extent of my technical knowledge on this topic, however.
Concernt
Optimism is non-negotiable
- Pronouns
- She/Her
Veiled?
Concernt
Optimism is non-negotiable
- Pronouns
- She/Her
Not exactly. Virtualization has come a long way since then. JIT, VM, emulation, there are plenty of ways to jump architecture and keep compatibility. Xbox One has Xbox 360 and Xbox (original) emulation after all.
As long as they build the backwards compatibility software, be it JIT compiling, emulation or what have you, as they go, and the hardware is capable enough, there shouldn't be any issues other than one or two games not working properly. Think more Xbox Series X compatibility with Xbox One games rather than the weird world of PS2 on PS3.
yes, but Erista > Drake is so far from that point
boodream
Paratroopa
If it is possible to download patched and improved versions of some major games published by Nintendo that you already had on the original Switch, this will make one more reason for the strong installed player base to "switch" to the new generation.
Being able to play Zelda, Xenoblade or Bayonetta in better conditions can't hurt. from a marketing point of view.
This time around, Nintendo won't be able to rely as much on Wii U games that a lot of people haven't played to fill in the gaps in the schedule.
I wonder if at some point they will include Gamecube and Wii games in NSO but only on Drake. That would suck, but unfortunately it seems plausible.
Being able to play Zelda, Xenoblade or Bayonetta in better conditions can't hurt. from a marketing point of view.
This time around, Nintendo won't be able to rely as much on Wii U games that a lot of people haven't played to fill in the gaps in the schedule.
I wonder if at some point they will include Gamecube and Wii games in NSO but only on Drake. That would suck, but unfortunately it seems plausible.
- Pronouns
- he / him
I'm thinking more of systems for 2029, 2035, whatever. At some point "We must do this forever because we made what was the right decision for 2015" is a problem. But it becomes much less a problem if there are other feasible ways to maintain compatibility.
I'm not sure Xbox is a winning example of working around big differences between two neighboring generations, though. Xbox One can't play most Xbox 360 games, and Xbox 360 can't play most Xbox games.
Concernt
Optimism is non-negotiable
- Pronouns
- She/Her
Why shouldn't a Nintendo console in 2039 be able to play Breath of the Wild?
My laptop can play Commander Keen. Planet X3. Attack of the PETSCII Robots.
Compatibility doesn't mean doing the same thing over and over again. It just means... Compatibility.
As for the Xbox Series X, its coverage of Xbox One games is over 99%. As of a recent update it doesn't even have to check for patches anymore. It can just run them. Not because it's the same console, but because they have a software solution for backwards compatibility that works well. Why shouldn't we expect AT LEAST that standard from consoles going forward?
- Pronouns
- He/Him
That’s not really a comparable point though… PC isn’t the same as consoles in this regard.
RennanNT
Bob-omb
ISAs are kinda like the human language. ARM is one language, x86 another and RISC-V is yet another.
A big reason XSS and PS5 have 99+% of compatibility is that they have the same language (x86) as their predecessor. Drake is the same language (ARM) as OG Switch, so we should be good there as well. And as long as x86 and ARM are great choices for CPUs, BC for future console should be good as well.
If the companies which are funding ARM R&D move to RISCV and ARM CPUs become outclassed though, Nintendo may have to move as well or get an expensive weak chip (like the Wii U CPU, which was using PowerPC, a dead "language" in 2012, to achieve BC with the Wii, which used it to achieve BC with GC).
Getting a RISCV console be 99+% BC with an ARM console should be possible (specially with a huge gap in power), but it's much harder to achieve than if both were the same language, so it's not really fair to say "but MS did it" just like it's not fair to say "but the Wii U got 100%" when comparing to the partial BC the X1 got.
Last edited:
Concernt
Optimism is non-negotiable
- Pronouns
- She/Her
While an imperfect comparison, it remains that Xbox:
Got backwards compatibility through virtualization working on a new architecture.
For backwards compatibility through virtualization working with 99+% accuracy.
Even if those two things weren't the same generation, the point is these aren't insurmountable tasks, and as pointed out, there's been and continues to be a concerted effort by the industry at launch to ensure ARM software works on RISC-V systems.
I think the first true litmus test of this will be a major manufacturer releasing a RISC-V Android phone. If apps "just work" without updates almost universally, we can somewhat breathe a sigh of relief.
That said I think the RISC-V transition for Nintendo is a ways off. I just don't think it will be a major challenge for them, technologically, to keep the Switch library working on it when it does come.
Looking towards the immediate future, Drake should have very very good compatibility and like most I expect it to be a mostly software solution, like how Xbox Series X handles Xbox One.
Changing CPU ISA doesn't necessarily mean a break in compatibility, it's just an additional obstacle that needs to be overcome. Certainly a large obstacle, but not an insurmountable one. Historically, consoles have dealt with this mainly by including extra processors for compatibility, though more software driven emulation is certainly an option on the table.
That said, it should be understood that all the RISC-V talk happening right now is really more of a long term topic. Drake isn't going to have any RISC-V CPU cores in it (at least directly exposed to games, recent Nvidia GPUs use them for internal purposes), and it's still highly uncertain that this issue would be directly raised for even the system after that. It's mostly just a reaction to recent ARM controversies and the partially related building momentum behind RISC-V.
Android is a bad comparison to consoles in this respect for several reasons, but I can guarantee you that, unless something unusual happens, large swaths of apps (including most games) aren't going to run on RISC-V without being updated.
Last edited:
