Andrei Frumusanu from Anandtech also mentions that big.LITTLE is not as important for laptops as it is for smartphones.
So I don't think the Cortex-A78C was designed for Nvidia in mind.
Based on how tech has progressed since the Switch, how powerful could a Switch Pro in early 2021 actually be?
- Thread starter ItWasMeantToBe19
- Start date
- Status
So I don't think the Cortex-A78C was designed for Nvidia in mind.
The lines between console and mobile game have been blurred so much that there's not much of a difference now. GI is proof of that as it doesn't look out of place on PS4. "AAA" games on mobile will just be AAA games on consoles with settings set to low at 720p
Well an Orin SoC seems like it would benefit more from A78C cores vs regular A78. It also seems like they would scale down better to the cut down versions of Orin.
This thread has been moving fast lately, some leaks going around or just speculation going in to hyperdrive?
I see a lot of confidence in posts here, that's encouraing :>
I see a lot of confidence in posts here, that's encouraing :>
I wonder if Nvidia's space budget would be less restrictive if Nvidia chose Samsung's 7LPP node or Samsung's 6LPP node?
But speaking about Samsung's failures to have a competitive process node, I hope Samsung's 3GAE node and Samsung's 3GAP node are actually competitive with TSMC's N3 node.
The actual Orin SoC uses the Cortex-A78AE. But a scaled down Orin SoC (e.g. Orin S or Orin ADAS) could indeed benefit from using the Cortex-A78C, especially for purpose outside of automotive vehicles. My point is that Arm didn't design the Cortex-A78C, so that Nvidia would be the only company to use the Cortex-A78C. I'm sure Qualcomm or maybe Samsung or maybe Mediatek would benefit from using the Cortex-A78C as well.
Mostly speculation.
Last edited:
Yeah, I've long thought that Nintendo would be wise to try out an Oculus Questesque standalone headset that doubled as a TV Mode Switch. I've basically been obsessed with the idea since the murmurs about Nintendo and Microsoft becoming friendly with each other started a few years back. The thought was, Nintendo could pull a bit of an R&D shortcut by co-opting some tech from Microsoft's stagnant Windows Mixed Reality line of headsets. Most importantly, the inside-out tracking bits as I understand that to be one of the big R&D hurdles.
Take the tracking stuff, maybe add a camera or two and fine tune it (it's 3.5 yr old tech at this point), shove the next gen Nvidia SoC (hopefully DLSS capable) and a battery in the headset, include some wands that joycon can slide into for tracking and a "dock/dongle" that facilitates TV mode and allows screen casting for asymmetric multiplayer. Release that at $399-$449 in 2020 then shift hardware focus toward developing the next flagship hybrid tablet that will include that same next gen SoC to replace the current model at $299 within the next year or so.
Like you said, there's not a company on Earth that's better positioned to fill out a VR platform with content from popular IP than Nintendo. Their whole ecosystem is built on mobile hardware and they'll probably keep the OG Switch as the target platform for at least a few more years. They could relatively easily supplement the occasional VR exclusive with VR-enabled enhancements like a 3D stereoscopic effect (ala 3DS) and/or hand-tracked pointer controls (ala Wii). They could really showcase a full spectrum of VR experiences from fully-immersive VR exclusives for the hardcore, to fitness/rhythm/arcady titles with broad appeal, asymmetric multiplayer party games like Nintendo Land and traditional flat games augmented with immersion only possible in VR.
There's something about the notion of a "premium" Switch that retails $100 more than the OG being sold alongside today's $299 model, while retaining the flagship form factor and just offering an uptick in performance, that doesn't feel right. But I'd also be worried about them failing to maintain their 20+ million annual sales pace if they kick the OG model to the curb and try to keep up with demand by leaning on the new SoC and replacing Mariko fully this year. A solution, in my mind, is to maintain the current SKU's through this year. Try to sell through that inventory while avoiding a price drop is long as possible. And roll out the next gen SoC in a "premium" SKU with an experimental form factor that would probably be expected to sell at more of a "niche" level than the flagship hybrid. Once the supply chain for that new SoC is ready to keep up with flagship demand, then switch on over to a replacement of the OG at $299. Also, a sharp departure from existing form factors might motivate more consumers to hold on to their original for the benefit of handheld mode while upgrading to Switch VR, thus playing into their long-stated goal of selling multiple Switches per household.
Hmm, 8 A78C would be a possible option for a 8cx successor. But would such a laptop have a power budget small enough to the point that they can't do something like 2 clusters of 1 X1 and 3 A78's each? Or a 2 X1/2 A78 cluster + a 4 A78 cluster? Or 1 X1/3 A78 + 4 A78?
Nevermind just read that the A78AE is automotive enhanced, but I can definitely see the Switch and Nvidia shield version of this SoC using A78C. Whether that be 4, 6 or 8 cores that have various clock configurations between them, the potential for sustained performance seems more likely.
My main question is do we know where the various A78 configurations fall performance wise relative to Desktop CPUs
Like, would a 8 core high clocked A78C be akin to a Ryzen 7 1700 or 2700 for example.
If they can even reach Ryzen 5 1600x or 2600x performance with a 6 core config that would be massive.
rumors through reputable sources
- 4K
- miniLED
- DLSS
- everything else
We currently don't have any devices that are using the new A78C cores to compare to, but we know that they have more cache than the standard A78. So my interest is in how will these cores compare to the ARM X1 cores performance and efficiency wise, since many of the current mobile SoC's are using at least one of these X1's as the performance core.
I take DLSS and I will run with it.
Everything else is just gravy..
There's only a hexa-core (6) configuration or an octa-core (8) configuration of the Cortex-A78C. A quad-core (4) configuration of the Cortex-A78C doesn't exist.
The nice thing about DLSS being rumored by a reputable source (Nate for instance) is that means it has to be a new and relatively strong chip for DLSS to be possible.
Yeah that's kind of low key the most significant detail there. Really narrows down the possibilities.
That's also my takeaway; and assuming that DLSS is also planned for handheld mode, it would mean a surprisingly powerful machine when docked.
One company that could aid Nintendo in developing a VR solution is Nvidia. They are already the tech partner for the SoC, but mostly they would benefit a lot if VR gets more popular. A larger "low-end" market would likely translate into more software being made and by extension a into larger "hi-end" market, meaming more demand for high performance GPUs on PCs. Also form what I understand they are already investing into VR specific R&D.
Yeah, thus where the "Between Xbox One S and PS4 Performance when docked" for native resolution estimation comes from.
With DLSS however, the range could be anywhere from slightly above the PS4 Pro to keeping up with the PS5 even depending on the Memory, Memory Bandwith, number and strength of the RT cores, and developer optimization.
Honestly, we could see some "PS5 Quality" ports from some devs, but PS4-Pro level ports from others depending on the engine/optimization/how well it uses DLSS.
It would be 1 X1 + 3*A78 + 4*A55 or 6*A78 or 8*A78.
X1 is only available in a standard 1+3+4 configuration unless ARM announces a new configuration. The only new configuration recently announced by ARM is actually A78C 6 and 8 cores configuration.
Big.LITTLE only really works for smartphones because, most of the time, they don't rely on sustained performances. That's why all these talks about S888/E2100 being weaker than S865 are pretty pointless. They all compare the sustained performances of these chips while a cortex X1 will probably be open an app 30~40% faster than an A78.
Yeah, I recall Nvidia doing demos of foveated rendering tech way back in 2016. They also confirmed that DLSS 2.1 supports VR and dynamic resolution, a technique frequently deployed on Switch, several months back.
Nintendo beating the whole industry to eye-tracked foveated rendering is probably more outlandish than anything predicted in this thread, but it is an example of how long Nvidia has been dabbling with VR solutions. Even if Switch VR hardware isn't imminent, I'd be shocked if Nintendo and Nvidia haven't had the conversation.
The thing about releases like this is the release date is set years in advance.
So even if there were supply issues, they can't change the date as retailers.etc plan things all around those.
If it is coming out this year they have to start fabricating chips shortly if they haven't already. The thing should have much more concrete details by march or april or it becomes hard to imagine them getting it out by september with literally not even the slightest peep from a development or manufacturing source.
I think shortages are going to be inevitable if the demand for the next Nintendo Switch model is at least half as strong as the Nintendo Switch.
If they end up releasing the revision in Holidays of 2021, I feel like it's going to make it a touch impossible to buy it for a few months. A global chip shortage doesn't sound like something that can easily be fixed especially with rabid fans causing it to be out of stock more often than not, not even touching on auto makers who seemingly have priority to computer chips for their new cars.
Please correct me if I'm wrong, but from what I think I learned in this thread is that production of silicon chips has a certain failure rate based on fabrication, and this is what determines how many are viable to sell in a given batch. If Nintendo uses chips that are going to be intentionally underclocked, then the chips fabricated that might otherwise be failed could be used for a Switch revision, ergo there might be a few more consoles made in the wild despite a global shortage? Is that a very rudimentary layman's understanding of the process?
Please correct me if I'm wrong, but from what I think I learned in this thread is that production of silicon chips has a certain failure rate based on fabrication, and this is what determines how many are viable to sell in a given batch. If Nintendo uses chips that are going to be intentionally underclocked, then the chips fabricated that might otherwise be failed could be used for a Switch revision, ergo there might be a few more consoles made in the wild despite a global shortage? Is that a very rudimentary layman's understanding of the process?
I feel like we'd have heard leaks about that (and that would factor into their reported inventory levels) so they probably haven't fabbed many if any yet.
yea more or less. smaller chips and less power hungry chips means better yield. also means cheaper chips as the more of the wafer you use, the more spread out the costs are. Bloomberg suggests this issue is hitting the PS5, hence some folks ire when Mochizuki is mentioned
Ahh, I see, I think that's the missing puzzle piece Ive been missing. Whenever I try to bring up discussion of a Switch revision with my friends, they keep immediately refuting it by pointing to the global chip shortage, but if this is primarily a PS5 (and I suppose to a lesser extent XSX) issue, then there's no reason it would be a problem for mobile tech like Nintendo then, eh?
Actually that's incorrect, it's effecting everything from RAM chips, to GPUs, to phones, etc.
However since the PS5 and XSX dies are quite large the issue is made worse.
Also the 8nm node is practically fully booked out with other Nvidia GPUs.
everything is effect. even the parts no one even heard of are hard to get. shit's crazy right now. one benefit of starting production early is that NIntendo can stockpile more parts for longer in order to have a modicum of better supply. maybe
It's not necessarily just a PS5 and XSX issue, but all of these companies such as AMD and Nvidia who are making these large monolithic GPU's running at max clocks are experiencing yield issues. People have to understand that before these new GPU's and next-gen consoles were on 7nm and 8nm, these manufacturing processes were used for mobile architecture primarily and we didn't have 250-350w devices using such nodes.
Ah, thank you both for the clarification. Is there any kind of long-term solution for fixing this global shortage crisis (hopefully beyond further stripmining the Earth of resources that continues to accelerate climate change)?
So it's not a matter of mining materials faster but production issues that need to be solved? I guess it's just with the times of wanting the latest and greatest on smaller devices, but it seems curious to me with full sized cars running computers that arguably have a little more space to use bigger chips that aren't so reliant on mobile technology keeping up with production
Last edited:
I'd be blown away if a new Switch Is able to play games on par with Xbox One S in portable mode and base PS4 while docked. Done thanks to a combination of better raw specs than OG Switch coupled with Nvidia's increasing architectural and efficiency advancements they've made in the years after Maxwell and Tegra X1/Erista.
that doesn't make any sense. people really exaggerate the gap between the One and PS4
This pandemic has highlighted a number of flaws especially when it comes to the tech world, such as needing to balance out chip manufacturing lines across the world. The bulk of all tech products are manufactured and then shipped from southeast Asia, so something like COVID-19 will effect everything in between because the logistics of moving parts around becomes so much more difficult.
The talks from companies like Foxconn, Samsung and TSMC opening up future plants in western countries could greatly alleviate the demand from being all on one specific territory of the world.
It will more likely be
- Portable
- Native Performance: Somewhere around 2.5x the power of the OG Switch's Portable mode
- With DLSS: PS4 Level locking it at 1080p
- Docked
- Native Performance: PS4 Level (Leveraging the Better CPU and a One S level GPU when docked)
- With DLSS: Anywhere from PS4 Pro with a locked 4k to PS5 with locked 4k level (Varies depending on Optimization/Developer's use of DLSS)
Honestly that would be very good. I believe everything here is possible/likely with the exception of any talk of PS5-level comparisons as the end, docked result in games.
PS4 Pro level seems more reasonable,
The main reason is DLSS creates a gamut of comparability depending on how the developers use it.
If they just stick with DLSS performance mode and go for IQ, more often than not it will likely be PS4 Pro level.
If they want to push it to the limit at the cost of IQ, they can use DRS+DLSS with 720p as the floor in order to squeeze in more graphical effects.etc to make it closer to the PS5, and for certain beyond the Series S.
A mobile SoC should be relatively high volume in output due to its size and the resulting number of working chips from a platter.
It would also avoid the main reason for current moment GPU shortages which is crypto miners.
Then again a successful Switch Pro/2 launch would likely require a lot of supply to satisfy demand.
So the question is had NV and/or Nintendo booked enough capacity beforehand at Samsung or whereever they are planning to produce the SoC.
The majority of components (things like card readers, wifi antennae, joycon rails, etc.) they won't need to change in terms of production, so it really comes down to SoC, RAM and storage I guess (also screen if they are indeed using a new screen technology). Assuming they use new RAM at least I can see that being the primary production bottleneck.
SoC shouldn't be the production bottleneck in any case.
SoC shouldn't be the production bottleneck in any case.
Speaking of RAM, there was that power outage at a Micron fab a couple of months ago.
And while looking this up, huh, I didn't know that Samsung started up 6400 mb/s LPDDR5 production on their 1z process back in August. I think that the last time we talked about LPDDR5 prices here, we were looking at the 5500 mb/s offerings?
And while looking this up, huh, I didn't know that Samsung started up 6400 mb/s LPDDR5 production on their 1z process back in August. I think that the last time we talked about LPDDR5 prices here, we were looking at the 5500 mb/s offerings?
Let say the screen is 720p and of course it has DLSS, what's more battery efficient for handheld mode: Running games natively 720p, or using DLSS to upscale it to 720p?
That depends.
In at least one game (I forget which! Sorry) the RTX 3090 has actually been benchmarked as having worse performance @1080p with DLSS versus just native rendering, because in some specific games it has such a good framerate already that the DLSS resolve time was higher than the saving in conventional rendering time. This varies game to game, and was a weird edge case.
At 720p I imagine you'd run into situations like that sometimes, depending on the complexity of the game. For battery? I would assume even if the render time was increased slightly you might see battery savings by doing DLSS, since the rest of the GPU is not doing work while the tensor cores do. But that's just my speculation.
I'm more interested in like, DLSS to 1080p/1440p (@ 648p or 720p or thereabouts) then donwsampled to the 720p screen. Would look crisp af, yummy.
Oh yeah, that was something I was wondering.
How do you compare the speeds of normal DDR/LPDDR to GDDR?
The main thing is the Xbox One S used normal DDR3 with a cache of 32mb of eSRAM while the PS4 used 8 GBs GDDR5 yet the main thing that weakened the One S was the raw GPU cores relative to the PS4 iirc.
Asking as it could be a big factor in comparing a Switch 2021 using LPDDR5 VS the PS4 using GDDR5 and the PS5/Xbox Series S | X using GDDR6.
If they were using DLSS to reconstruct to 720p, it would likely be running at 540p in order to keep highest IQ (DLSS Quality). Battery life and efficiency wise, imagine cutting the workload by anywhere from 25%-33.33%
It probably depends on how the clock profiles are set up. If they can afford to clock it a lower profile and still get 720p with DLSS that would probably be more energy efficient.
Running DLSS upscaled to 720p will be more efficient for battery life.
Edge of Eternity. I brought it up with Dictator and he said an Nvidia engineer told him that rare instances would crop up like that. don't think it's really applicable for the Switch Pro. EoE was already running at 200fps before DLSS. the tensor cores probably just couldn't keep up
- Status