StarTopic Future Nintendo Hardware & Technology Speculation & Discussion |ST| (Read the staff posts before commenting!)

[Z0m3le]

Yeah, it puts Samsung's 5nm processes at around the same place as TSMC's 7nm family, which is generally in line with expectations.

Regarding the use of original Switch clocks on the new model, although I don't think that specific rumour about using Samsung 5LPP has any weight to it, it's helpful in highlighting one of the reasons why I think the size of T239's GPU is indicative of a more advanced process, and actually higher clocks than the original Switch.

By poking around at the Jetson Power Tool, we can find that the power curve used for the Orin GPU fits very closely to the following equation:

P = N x 0.4132e^2.01C

Where N is the number of TPCs (or 2x the number of SMs), C is the clock speed measured in GHz, and P is the power consumption measured in Watts. If we take the 27% reduction for 5LPP as a flat value, we can just multiply the equation by 0.73, giving us a hypothetical 5LPP power curve of:

P = N x 0.3016e^2.01C

So, for a 6 TPC design like T239, we would get 3.92W for the GPU at 384MHz and 8.47W for it at 768MHz. Both of these are within the ballpark of what we'd expect for a new Switch, but still don't really explain why they would use such a large GPU. If we invert the equation, we can calculate the clock speed that can be achieved at a given power consumption for a given number of TPCs:

8nm: C = ln( P / (N x 0.4132) ) / 2.01
5nm: C = ln( P / (N x 0.3016) ) / 2.01

Let's assume that Nintendo were choosing between an 8 SM design and a 12 SM design, both on 5LPP with this hypothetical power curve. If their goal was 8.75W for the GPU in docked mode, then they could either have an 8 SM design clocked at 970MHz, providing 1,986 Gflops, or a 12 SM design clocked at 768MHz, providing 2,359 Gflops. Effectively, they're increasing their GPU size by 50%, but only achieving a 19% performance increase out of it. It's not zero return on investment, but it's not great.

Portable mode makes less sense, though. An 8 SM GPU within a 3.92W limit could clock to 586MHz, which would give 1,201 Gflops. A 12 SM design clocked at 384MHz consumes the same amount of power, and hits 1,178 Gflops. That is, they're actually getting slightly less performance with 12 SMs than they would have with 8.

Of course this analysis is inherently limited by assuming that 5LPP provides a simple, scalar reduction in power over 8mm. However, I'd still expect roughly similar behaviour. Effectively what we're looking at here is the marginal return on an increased number of TPCs for a given power draw, or equivalently for given clocks. This is related to the power efficiency curve, and we should see that it's 0 at the clock speed which provides peak power efficiency, tending up towards 1 at peak clocks, and it's negative below the peak power efficiency. Hence why the 12 SM GPU at 3.92W performs worse than an 8 SM one does, because it's dealing with clock speeds below the point of peak efficiency.

As a point of reference, if we take the 8nm Orin power curve from above, we can calculate the clock speed which achieves maximum efficiency: 477MHz. This explains why they don't clock below 420MHz on Orin Jetson products and instead disable TPCs at lower power settings, because it actually provides more performance given they're below peak efficiency. If we do the same for the hypothetical 5LPP curve, we get 644MHz as the peak efficiency. This probably doesn't bear much relationship to the actual point of peak efficiency on 5LPP, given the crude nature of applying a scalar shift to the curve, but we should definitely see this peak efficiency point increase as we move onto more efficient manufacturing processes.

I would be very surprised to see Nintendo using clock speeds lower than the peak efficiency point for the process they're using. If they were, then they'd effectively be paying extra for a less powerful GPU. If money weren't an issue, then the hypothetical ideal design for a power-limited chip should be to identify the peak efficiency clock speed and then choose however many TPCs fit in your power budget at that clock speed. Removing TPCs would reduce performance slightly while lowering your costs, while adding TPCs would also reduce performance but raise your costs.

If you're more constrained by cost than power consumption, then the optimal design is simply as many TPCs as you can afford, clocked as high as you can. In a more realistic scenario where you're balancing cost and power draw of various components, the design will sit somewhere between these two extremes, sitting at a sweet spot in the power/clock/cost space where the marginal benefit of adding more TPCs isn't worth the additional cost.

Of course Nintendo actually have two power profiles to be concerned about, portable and docked, but power efficiency is far more important in portable mode, whereas docked mode is going to be more balanced against cost. Running at below peak efficiency clocks in portable mode would effectively mean they've chosen a design which trades away power efficiency in portable mode in favour of improved power efficiency in docked mode, and increased their costs in doing so, which doesn't make a whole lot of sense to me.

I would expect Nintendo to have chosen a GPU such that they're clocking somewhere above peak efficiency clocks in portable mode, and around 2x that in docked mode. This gives them a good balance of performance, power draw and cost, and it's exactly what they did with TX1. At 8nm we can easily see that T239 doesn't have such a GPU, as clocking at peak efficiency clocks of 477MHz would draw 6.47W for the GPU alone in portable mode. Assuming about 3W for the GPU, the optimal number of TPCs on 8nm would be 2.78, or 5.57 SMs.

On our hypothetical 5LPP power curve, with a peak efficiency point at 644MHz, we would get 6.6W for T239's GPU. As I said, this is a very crude, and to be honest is probably a good illustration of why we shouldn't just treat the differences in power consumption between manufacturing processes as a flat percentage. The peak efficiency point is likely to be lower than this, and the 27% power improvement is unlikely to be representative of the lower end of the power curve. Still, I would be surprised if 5LPP were so much more efficient than 8N that 12 SMs would be a sensible design choice. You'd need around a 50% reduction in power draw compared to 8N at the low end of the curve for 12 SMs to make sense. Judging by Ampere/Ada comparisons, 4N does seem to offer around a 50% reduction in power draw over 8N.

Anyway, my point is that using original Switch clocks on any process of 8nm or better would mean that they've chosen a GPU that's too big for their requirements, and are paying more for something that's giving them less performance in portable mode and only marginal improvements docked. As I see it, increases in the minimum viable clock speed (ie peak efficiency clock) with improved manufacturing processes make a clock speed of 500MHz+ in portable mode more likely, and a similar increase to the docked clock. That being the case, it's impossible to justify 12 SMs on 8nm, and honestly hard to justify them on either Samsung 5nm or TSMC 7nm. Only on TSMC's 5nm/4nm processes does 12 SMs seem sensible to me.

A few things that people might miss on here... 5LPP is actually Samsung 4LPX*, which has a further reduction according to the earlier chart you posted of ~10%, putting the power consumption difference between "8nm" and "4nm" at ~40% power reduction in that chart. It's also been noted that Drake has Ada power reduction enhancements via the Nvidia Hack. This should further reduce this power consumption. Comparing nodes multiple generations apart is not a science, too many variables to deal with, but what I've come up with is somewhere around 45% for Drake power reduction vs Orin.

*See dakhil's post below for correction

 

[Dakhil]

A few things that people might miss on here... 5LPP is actually Samsung 4LPE, which has a further reduction according to the earlier chart you posted of ~16%, putting the power consumption difference between "8nm" and "4nm" at ~40% power reduction in that chart. It's also been noted that Drake has Ada power reduction enhancements via the Nvidia Hack. This should further reduce this power consumption. Comparing nodes multiple generations apart is not a science, too many variables to deal with, but what I've come up with is somewhere around 46-47% for Drake power reduction vs Orin.

That's not quite correct. Samsung's 4LPX process node, which the Snapdragon 8 Gen 1 uses, is practically Samsung's 5LPP process node, but rebranded (here and here). Samsung's 4LPE process node, which is used by the Exynos 2200, is quite different from Samsung's 4LPX process node.

 

[Zedark]

Yeah, it puts Samsung's 5nm processes at around the same place as TSMC's 7nm family, which is generally in line with expectations.

Regarding the use of original Switch clocks on the new model, although I don't think that specific rumour about using Samsung 5LPP has any weight to it, it's helpful in highlighting one of the reasons why I think the size of T239's GPU is indicative of a more advanced process, and actually higher clocks than the original Switch.

By poking around at the Jetson Power Tool, we can find that the power curve used for the Orin GPU fits very closely to the following equation:

P = N x 0.4132e^2.01C

Where N is the number of TPCs (or 2x the number of SMs), C is the clock speed measured in GHz, and P is the power consumption measured in Watts. If we take the 27% reduction for 5LPP as a flat value, we can just multiply the equation by 0.73, giving us a hypothetical 5LPP power curve of:

P = N x 0.3016e^2.01C

So, for a 6 TPC design like T239, we would get 3.92W for the GPU at 384MHz and 8.47W for it at 768MHz. Both of these are within the ballpark of what we'd expect for a new Switch, but still don't really explain why they would use such a large GPU. If we invert the equation, we can calculate the clock speed that can be achieved at a given power consumption for a given number of TPCs:

8nm: C = ln( P / (N x 0.4132) ) / 2.01
5nm: C = ln( P / (N x 0.3016) ) / 2.01

Let's assume that Nintendo were choosing between an 8 SM design and a 12 SM design, both on 5LPP with this hypothetical power curve. If their goal was 8.75W for the GPU in docked mode, then they could either have an 8 SM design clocked at 970MHz, providing 1,986 Gflops, or a 12 SM design clocked at 768MHz, providing 2,359 Gflops. Effectively, they're increasing their GPU size by 50%, but only achieving a 19% performance increase out of it. It's not zero return on investment, but it's not great.

Portable mode makes less sense, though. An 8 SM GPU within a 3.92W limit could clock to 586MHz, which would give 1,201 Gflops. A 12 SM design clocked at 384MHz consumes the same amount of power, and hits 1,178 Gflops. That is, they're actually getting slightly less performance with 12 SMs than they would have with 8.

Of course this analysis is inherently limited by assuming that 5LPP provides a simple, scalar reduction in power over 8mm. However, I'd still expect roughly similar behaviour. Effectively what we're looking at here is the marginal return on an increased number of TPCs for a given power draw, or equivalently for given clocks. This is related to the power efficiency curve, and we should see that it's 0 at the clock speed which provides peak power efficiency, tending up towards 1 at peak clocks, and it's negative below the peak power efficiency. Hence why the 12 SM GPU at 3.92W performs worse than an 8 SM one does, because it's dealing with clock speeds below the point of peak efficiency.

As a point of reference, if we take the 8nm Orin power curve from above, we can calculate the clock speed which achieves maximum efficiency: 477MHz. This explains why they don't clock below 420MHz on Orin Jetson products and instead disable TPCs at lower power settings, because it actually provides more performance given they're below peak efficiency. If we do the same for the hypothetical 5LPP curve, we get 644MHz as the peak efficiency. This probably doesn't bear much relationship to the actual point of peak efficiency on 5LPP, given the crude nature of applying a scalar shift to the curve, but we should definitely see this peak efficiency point increase as we move onto more efficient manufacturing processes.

I would be very surprised to see Nintendo using clock speeds lower than the peak efficiency point for the process they're using. If they were, then they'd effectively be paying extra for a less powerful GPU. If money weren't an issue, then the hypothetical ideal design for a power-limited chip should be to identify the peak efficiency clock speed and then choose however many TPCs fit in your power budget at that clock speed. Removing TPCs would reduce performance slightly while lowering your costs, while adding TPCs would also reduce performance but raise your costs.

If you're more constrained by cost than power consumption, then the optimal design is simply as many TPCs as you can afford, clocked as high as you can. In a more realistic scenario where you're balancing cost and power draw of various components, the design will sit somewhere between these two extremes, sitting at a sweet spot in the power/clock/cost space where the marginal benefit of adding more TPCs isn't worth the additional cost.

Of course Nintendo actually have two power profiles to be concerned about, portable and docked, but power efficiency is far more important in portable mode, whereas docked mode is going to be more balanced against cost. Running at below peak efficiency clocks in portable mode would effectively mean they've chosen a design which trades away power efficiency in portable mode in favour of improved power efficiency in docked mode, and increased their costs in doing so, which doesn't make a whole lot of sense to me.

I would expect Nintendo to have chosen a GPU such that they're clocking somewhere above peak efficiency clocks in portable mode, and around 2x that in docked mode. This gives them a good balance of performance, power draw and cost, and it's exactly what they did with TX1. At 8nm we can easily see that T239 doesn't have such a GPU, as clocking at peak efficiency clocks of 477MHz would draw 6.47W for the GPU alone in portable mode. Assuming about 3W for the GPU, the optimal number of TPCs on 8nm would be 2.78, or 5.57 SMs.

On our hypothetical 5LPP power curve, with a peak efficiency point at 644MHz, we would get 6.6W for T239's GPU. As I said, this is a very crude, and to be honest is probably a good illustration of why we shouldn't just treat the differences in power consumption between manufacturing processes as a flat percentage. The peak efficiency point is likely to be lower than this, and the 27% power improvement is unlikely to be representative of the lower end of the power curve. Still, I would be surprised if 5LPP were so much more efficient than 8N that 12 SMs would be a sensible design choice. You'd need around a 50% reduction in power draw compared to 8N at the low end of the curve for 12 SMs to make sense. Judging by Ampere/Ada comparisons, 4N does seem to offer around a 50% reduction in power draw over 8N.

Anyway, my point is that using original Switch clocks on any process of 8nm or better would mean that they've chosen a GPU that's too big for their requirements, and are paying more for something that's giving them less performance in portable mode and only marginal improvements docked. As I see it, increases in the minimum viable clock speed (ie peak efficiency clock) with improved manufacturing processes make a clock speed of 500MHz+ in portable mode more likely, and a similar increase to the docked clock. That being the case, it's impossible to justify 12 SMs on 8nm, and honestly hard to justify them on either Samsung 5nm or TSMC 7nm. Only on TSMC's 5nm/4nm processes does 12 SMs seem sensible to me.

This a great analysis! I'm unsure about the choice of fitting an exponential formula to the power curve, however: wouldn't a quadratic relation be more physically plausible? I believe that power scales quadratically with frequency. Is there a way to test a quadratic fit (how did you get the data points btw? If it's an open source software tool, I'd love to play around with it)?

A quadratic fit could change the analysis you presented as well, so it's an important thing to consider I think.

 

[Z0m3le]

This a great analysis! I'm unsure about the choice of fitting an exponential formula to the power curve, however: wouldn't a quadratic relation be more physically plausible? I believe that power scales quadratically with frequency. Is there a way to test a quadratic fit (how did you get the data points btw? If it's an open source software tool, I'd love to play around with it)?

A quadratic fit could change the analysis you presented as well, so it's an important thing to consider I think.

As much as I appreciate the work Thraktor puts in, the analysis should be taken with a grain of salt as Thraktor says. It's using unknown versions of fab processes, results are not 1:1, and Drake has power saving features built into the SoC from Ada, that Orin doesn't have. Unless I missed an exact chip produced at both 8lpp and 5lpp.

There actually is a chip that exists that does this between 8LPP and 7LPP, the Exynos 9820 and Exynos 9825. These are basically the same chip, and we can see that even though the 8LPP phone has a 17% larger battery, a smaller screen and lower clocks, the 7LPP device lasts 8% longer. This roughly actually works out to ~30% less power draw when everything is taken into account. 5LPP according to Samsung is an additional 20% power reduction from 7LPP, This puts it right around ~45%, again Drake also offers some power savings found in Ada, that Orin does not have, meaning ~50% power reduction is what one should expect from Drake when compared to Orin with the same configuration if Drake is 5LPP.

Drake would also be around half the size on 5LPP vs 8LPP, going from silly big at over 250mm^2 to ~130mm^2, much more in line with TX1's 118mm^2 die size.

Again, with the above power reduction taken into account, the DLSS test clock's naming scheme would line up with Samsung 5LPP very easily, (giving us at least an idea of clocks and thus performance) if related to Drake's target power consumption at those clocks (some suggest a big if, but can't give an alternative, meanwhile a straight forward reading / Occam's razor (Occam's razor is a principle often attributed to … 14th century friar William of Ockham that says that if you have two competing ideas to explain the same phenomenon, you should prefer the simpler one.)

 

[Eclissi]

I don't know if it's reliable yet, but keep tweeting about the new Tegra...

 

[Z0m3le]

I would expect Nintendo to have chosen a GPU such that they're clocking somewhere above peak efficiency clocks in portable mode, and around 2x that in docked mode. This gives them a good balance of performance, power draw and cost, and it's exactly what they did with TX1. At 8nm we can easily see that T239 doesn't have such a GPU, as clocking at peak efficiency clocks of 477MHz would draw 6.47W for the GPU alone in portable mode. Assuming about 3W for the GPU, the optimal number of TPCs on 8nm would be 2.78, or 5.57 SMs.

On our hypothetical 5LPP power curve, with a peak efficiency point at 644MHz, we would get 6.6W for T239's GPU. As I said, this is a very crude, and to be honest is probably a good illustration of why we shouldn't just treat the differences in power consumption between manufacturing processes as a flat percentage. The peak efficiency point is likely to be lower than this, and the 27% power improvement is unlikely to be representative of the lower end of the power curve. Still, I would be surprised if 5LPP were so much more efficient than 8N that 12 SMs would be a sensible design choice. You'd need around a 50% reduction in power draw compared to 8N at the low end of the curve for 12 SMs to make sense. Judging by Ampere/Ada comparisons, 4N does seem to offer around a 50% reduction in power draw over 8N.

Anyway, my point is that using original Switch clocks on any process of 8nm or better would mean that they've chosen a GPU that's too big for their requirements, and are paying more for something that's giving them less performance in portable mode and only marginal improvements docked. As I see it, increases in the minimum viable clock speed (ie peak efficiency clock) with improved manufacturing processes make a clock speed of 500MHz+ in portable mode more likely, and a similar increase to the docked clock. That being the case, it's impossible to justify 12 SMs on 8nm, and honestly hard to justify them on either Samsung 5nm or TSMC 7nm. Only on TSMC's 5nm/4nm processes does 12 SMs seem sensible to me.

Considering Exynos 9820 and Exynos 9825, real world results of 7LPP is closer to the numbers you are using than 5LPP. Even just taking the raw numbers from the batteries of the Galaxy S10 Plus and Note 10 (4100mah and 3500mah) the Note 10 drew ~22%+ less energy (plus being the negatives against it, higher / more sustained clocks, bigger screen). 7LPP is a less advanced node than 5LPP obviously, Samsung claimed 20% reduction in power over 7LPP. I think combined with the Ada power features Drake borrows, it can get close enough to that 50% power reduction, you and I both believe Drake needs. I'm not trying to prove the "leaker" is accurate btw, I'm just sticking to what I've said all along, that 5nm Samsung makes the logical sense, and when we are looking at such a big chip, 8nm is completely irrelevant, Nintendo didn't go out and spend on a big, expensive, power hungry chip, because Nvidia works with a budget and a power target in mind, and Nvidia certainly knew 8nm wasn't going to work for Drake's configuration inside of a Switch form factor.

I don't know if it's reliable yet, but keep tweeting about the new Tegra...

He is mostly just talking about people like me who DM'd and Tweeted at him about Drake's Linux public leak and the Nvidia hack info, basically he had no idea about anything we've been discussing. From the messages he was sending, he doesn't really seem to know anything about Switch or it's successor, outside of maybe seeing 'Nintendo Tegra 5LPP' somewhere.

 

[Thugstas]

Considering Exynos 9820 and Exynos 9825, real world results of 7LPP is closer to the numbers you are using than 5LPP. Even just taking the raw numbers from the batteries of the Galaxy S10 Plus and Note 10 (4100mah and 3500mah) the Note 10 drew ~22%+ less energy (plus being the negatives against it, higher / more sustained clocks, bigger screen). 7LPP is a less advanced node than 5LPP obviously, Samsung claimed 20% reduction in power over 7LPP. I think combined with the Ada power features Drake borrows, it can get close enough to that 50% power reduction, you and I both believe Drake needs. I'm not trying to prove the "leaker" is accurate btw, I'm just sticking to what I've said all along, that 5nm Samsung makes the logical sense, and when we are looking at such a big chip, 8nm is completely irrelevant, Nintendo didn't go out and spend on a big, expensive, power hungry chip, because Nvidia works with a budget and a power target in mind, and Nvidia certainly knew 8nm wasn't going to work for Drake's configuration inside of a Switch form factor.

He is mostly just talking about people like me who DM'd and Tweeted at him about Drake's Linux public leak and the Nvidia hack info, basically he had no idea about anything we've been discussing. From the messages he was sending, he doesn't really seem to know anything about Switch or it's successor, outside of maybe seeing 'Nintendo Tegra 5LPP' somewhere.

i can confirm this too.

 

[space-coyote]

Even at only having Nintendo published titles dated up until July, only Pikmin 4 and TOTK are actually from EPD, and TOTK is only where it is due to multiple delays. That sounds extremely light for a year or EPD releases.

I don't see many reasons why they'd be quiet if there are indeed more, outside of a new console that they'll be coming out on.

I don't think the amount of EPD releases we know if is suggestive of anything really. If we exclude the launch year of switch, this year is looking roughly on par as. or in many cases better than, the rest of the switch's lifecycle with regard to EPD releases.

• In 2018 we got 5 EPD releases, however 3 of those were different Labo kits, 1 was a partner game (Sushi Striker), and the last one was a port (Captain Toad)
• In 2019 we got 5 EPD releases, 1 was a Labo kit, 1 was a port (NSMBU), 1 was a partner game (Brain Training), with the remaining 2 being fully new games (Ring Fit and Mario Maker 2)
• In 2020 we got 3 EPD releases, 2 being ports (Mario 3D All-Stars, Pikmin 3), and only 1 new game (Animal Crossing)
• In 2021 we got 6 EPD releases, 1 was a port (Bowser's Fury), 3 were developed alongside another studio (Metroid Dread, Miitopia, Warioware), 1 was game builder garage, and the remaining release (Big Brain Academy) doesn't even have a wikipedia page.
• In 2022 we only got 2 EPD releases (Splatoon 3 and Switch Sports)

 

[Concernt]

I don't think the amount of EPD releases we know if is suggestive of anything really. If we exclude the launch year of switch, this year is looking roughly on par as. or in many cases better than, the rest of the switch's lifecycle with regard to EPD releases.

• In 2018 we got 5 EPD releases, however 3 of those were different Labo kits, 1 was a partner game (Sushi Striker), and the last one was a port (Captain Toad)
• In 2019 we got 5 EPD releases, 1 was a Labo kit, 1 was a port (NSMBU), 1 was a partner game (Brain Training), with the remaining 2 being fully new games (Ring Fit and Mario Maker 2)
• In 2020 we got 3 EPD releases, 2 being ports (Mario 3D All-Stars, Pikmin 3), and only 1 new game (Animal Crossing)
• In 2021 we got 6 EPD releases, 1 was a port (Bowser's Fury), 3 were developed alongside another studio (Metroid Dread, Miitopia, Warioware), 1 was game builder garage, and the remaining release (Big Brain Academy) doesn't even have a wikipedia page.
• In 2022 we only got 2 EPD releases (Splatoon 3 and Switch Sports)

The only EPD scheduled for 2023 that we know of was Pikmin 4. Which definitely seems rather exceptionally low. I suppose we'll see. We could get a summer games blowout with 0 hardware, but I doubt it.

 

[Dakhil]

https://www.ft.com/content/25ea2570-264d-472d-8ef8-3a02003e87ab

Arm is seeking to raise prices for its chip designs as the SoftBank-owned group aims to boost revenues ahead of a hotly anticipated initial public offering in New York this year.

The UK-based group, which designs blueprints for semiconductors found in more than 95 per cent of all smartphones, has recently informed several of its biggest customers of a radical shift to its business model, according to several industry executives and former employees.

These people said Arm planned to stop charging chipmakers royalties for using its designs based on a chip’s value and instead charge device makers based on the value of the device. This should mean the company earns several times more for each design it sells, as the average smartphone is vastly more expensive than a chip.

The changes represent one of the biggest shake-ups to Arm's business strategy in decades, at a time when SoftBank chief executive Masayoshi Son is seeking to drive up Arm's profits and attract investors to its impending return to the public markets.

"Arm is going to customers and saying 'We would like to get paid more money for broadly the same thing'," said one former senior employee who left the company last year. "What SoftBank is doing at the moment is testing the market value of the monopoly that Arm has."

SoftBank, which acquired Arm for £24.3bn in 2016, plans to retain a majority stake following the IPO. It was aiming to start pushing through the pricing shake-up at the chip designer by as early as next year, but has so far been frustrated by customers' reluctance to accept the new arrangement.

MediaTek, Unisoc and Qualcomm, and multiple Chinese smartphone makers including Xiaomi and Oppo, are among the companies that have been made aware of the proposed change to pricing policy, according to several people familiar with the talks.

The Cambridge-headquartered company licenses its designs to various chipmakers for them to use to make the semiconductors found in smartphones, computers and cars. It charges a licence fee for obtaining its blueprints, and then a recurring royalty for each chip shipped.

Arm has also become more aggressive in pushing price increases within its existing sales model for royalties and licences over the past year, particularly for customers making chips for smartphones, where it has a dominant market position, according to people with knowledge of the recent moves.

Arm has been locked in a bitter legal battle with its largest customer, Qualcomm, since the second half of last year, accusing the mobile chipmaker of using its intellectual property without permission. In its counterclaim, the chipmaker alleged that Arm had told "one or more of Qualcomm's customers" that it would cease licensing central processing units (CPU) to all chip companies and would only provide licences to device makers themselves.

According to the new business model being presented by Arm, royalties would be set according to the average selling price (ASP) of mobile devices rather than that of the chips. The changes will mainly involve Arm's most prominent "Cortex-A" designs, essential for the development of smartphone processors.

Charging based on device price is a widespread practice across the telecoms equipment market, with Qualcomm, Nokia and Ericsson all using a similar model for their patents. The problem for Arm is that it is attempting to change its pricing strategy long after it established a different sales model.

The average price for a smartphone computing chip is about $40 for Qualcomm, $17 for MediaTek and $6 for Unisoc. Arm charges royalties of about 1 to 2 per cent of the value of each chip sold based on its designs, according to Sravan Kundojjala, an analyst at TechInsights. By contrast, the average smartphone sold for $335 in 2022. While it is unlikely Arm would seek as much as 1 to 2 per cent of the value of each device, those familiar with the matter said the company would set its new pricing in a way that significantly increases overall earnings.

"The [royalty] amount will be at least several times higher than what Arm gets now," said an executive from a leading Chinese smartphone maker which has so far refused to back the proposed plan. "We are told that they hope such changes could start from 2024."

Some of Arm's customers, including Apple, are both chipmakers and device makers, and have special licensing and royalty agreements with Arm. The iPhone maker is not involved in discussions about the change to Arm's business model, said executives with knowledge of the company’s recent discussions.

Arm, SoftBank, Qualcomm, MediaTek, Unisoc, Xiaomi and Oppo declined to comment.

Son is relying on a blockbuster Arm IPO to help mount a turnround at SoftBank, which has suffered heavy losses over the past year as the value of its tech investments was hammered in a broader industry downturn.

Son announced last year that he would step back from day-to-day operations at SoftBank to devote himself to turbocharging Arm's growth. People close to the billionaire said he felt Arm had taken a smaller slice of the industry profits over the past decade than chipmakers such as Nvidia, Broadcom and Qualcomm, especially considering how crucial and ubiquitous its designs were in mobiles.

Investors in the US, UK and Japan have told the Financial Times they have applied valuations to Arm of between $30bn and $70bn. They say the broad range stems from the difficulty in identifying any comparable companies, and a lack of clarity over the company’s precise growth strategy in recent years.

For several years under SoftBank’s ownership, Arm posted stagnating revenue and falling profits. Arm's costs increased from $716mn in 2015 to $1.6bn in 2019, according to SoftBank data. Revenues gained 20 per cent to $1.9bn over the same period, while profits fell almost 70 per cent to $276mn by 2019.

It has since reversed its fortunes, posting a 35 per cent rise in revenues to $2.7bn in 2021, the latest date for which there are annual figures, and a 68 per cent rise in adjusted earnings to $1bn.

Additional reporting by Qianer Liu in Hong Kong and Leo Lewis in Tokyo

 

[Z0m3le]

This pricing structure makes no sense. It would benefit cheap phones with high performance SoCs but lacking other features. SBCs would be wildly priced, with cheap SoCs not making sense anymore. This structure doesn't make sense to me.

 

[LuckyLuke]

Do you think there will be a hardware innovation typical of Nintendo? The "magic" with a new game at launch as 1-2-Switch.

 

[Skittzo]

Do you think there will be a hardware innovation typical of Nintendo? The "magic" with a new game at launch as 1-2-Switch.

Probably something neat using tensor cores. Maybe Nintendogs with learning dog AI.

 

[Goodtwin]

First Party Nintendo Games First Half 2023:

January: Fire Emblem Engage
February: Metroid Prime Remastered, Kirby's Return to Dreamland
March: Bayonetta Origins
April: Advance Wars 1+2
May: Zelda TotK
June: ?
July: Pikmin 4 (yes I know its technically second half, but just barely.)

When you look at the lineup along with some notable release from third parties for the first half of the year, its hard to justify a strong first half of the year only to have a soft second half. If Switch Redacted isn't coming this year and Nintendo doesn't have any big hitters planned for the second half, why wouldn't they structure the first half of releases differently to space them out more to better cover the entire year? The second half of the year is intentionally wide open right now and the idea that Nintendo is just planning on a soft second half doesn't make sense when it could be avoided by moving around a couple of releases from the first half. This doesn't guarantee new hardware by no stretch of the imagination, but if Nintendo does announce some big hitters for later this year with no new hardware in sight, such as a new 3D Mario, that makes new hardware coming any time soon look a lot less likely. I do not see Nintendo rolling out their successor with no Zelda or 3D Mario at launch. Nintendo could have a very strong year without new hardware if they release a few big hitters in the second half, but at some point Nintendo has to start staging software for the next hardware, unless of course that hardware isn't coming anytime soon. If that is the case, this will end up being another blunder for Nintendo, because Switch sales are going to start to fall pretty hard very soon.

 

[majimazio]

First Party Nintendo Games First Half 2023:

January: Fire Emblem Engage
February: Metroid Prime Remastered, Kirby's Return to Dreamland
March: Bayonetta Origins
April: Advance Wars 1+2
May: Zelda TotK
June: ?
July: Pikmin 4 (yes I know its technically second half, but just barely.)

When you look at the lineup along with some notable release from third parties for the first half of the year, its hard to justify a strong first half of the year only to have a soft second half. If Switch Redacted isn't coming this year and Nintendo doesn't have any big hitters planned for the second half, why wouldn't they structure the first half of releases differently to space them out more to better cover the entire year? The second half of the year is intentionally wide open right now and the idea that Nintendo is just planning on a soft second half doesn't make sense when it could be avoided by moving around a couple of releases from the first half. This doesn't guarantee new hardware by no stretch of the imagination, but if Nintendo does announce some big hitters for later this year with no new hardware in sight, such as a new 3D Mario, that makes new hardware coming any time soon look a lot less likely. I do not see Nintendo rolling out their successor with no Zelda or 3D Mario at launch. Nintendo could have a very strong year without new hardware if they release a few big hitters in the second half, but at some point Nintendo has to start staging software for the next hardware, unless of course that hardware isn't coming anytime soon. If that is the case, this will end up being another blunder for Nintendo, because Switch sales are going to start to fall pretty hard very soon.

There have been three 3D Mario releases for the Switch, including the port.
I don't believe they will release any more 3D Mario for Switch.
The new 3D Mario should be a [REDACTED] launch title and MP4 should be cross-gen.

Apart from that, I want a new 2D Mario for the Switch for the Holidays.

 

[LuckyLuke]

First Party Nintendo Games First Half 2023:

January: Fire Emblem Engage
February: Metroid Prime Remastered, Kirby's Return to Dreamland
March: Bayonetta Origins
April: Advance Wars 1+2
May: Zelda TotK
June: ?
July: Pikmin 4 (yes I know its technically second half, but just barely.)

When you look at the lineup along with some notable release from third parties for the first half of the year, its hard to justify a strong first half of the year only to have a soft second half. If Switch Redacted isn't coming this year and Nintendo doesn't have any big hitters planned for the second half, why wouldn't they structure the first half of releases differently to space them out more to better cover the entire year? The second half of the year is intentionally wide open right now and the idea that Nintendo is just planning on a soft second half doesn't make sense when it could be avoided by moving around a couple of releases from the first half. This doesn't guarantee new hardware by no stretch of the imagination, but if Nintendo does announce some big hitters for later this year with no new hardware in sight, such as a new 3D Mario, that makes new hardware coming any time soon look a lot less likely. I do not see Nintendo rolling out their successor with no Zelda or 3D Mario at launch. Nintendo could have a very strong year without new hardware if they release a few big hitters in the second half, but at some point Nintendo has to start staging software for the next hardware, unless of course that hardware isn't coming anytime soon. If that is the case, this will end up being another blunder for Nintendo, because Switch sales are going to start to fall pretty hard very soon.

There are Pokémon DLCs in the second half. Sure they're just DLC, but it's Pokémon.

However I think they will do like Switch. Announcement October 2023 with some new games. Metroid Prime 4, 3D Mario etc... and console release in March 2024.

 

[Link_enfant]

First Party Nintendo Games First Half 2023:

January: Fire Emblem Engage
February: Metroid Prime Remastered, Kirby's Return to Dreamland
March: Bayonetta Origins
April: Advance Wars 1+2
May: Zelda TotK
June: ?
July: Pikmin 4 (yes I know its technically second half, but just barely.)

When you look at the lineup along with some notable release from third parties for the first half of the year, its hard to justify a strong first half of the year only to have a soft second half. If Switch Redacted isn't coming this year and Nintendo doesn't have any big hitters planned for the second half, why wouldn't they structure the first half of releases differently to space them out more to better cover the entire year? The second half of the year is intentionally wide open right now and the idea that Nintendo is just planning on a soft second half doesn't make sense when it could be avoided by moving around a couple of releases from the first half. This doesn't guarantee new hardware by no stretch of the imagination, but if Nintendo does announce some big hitters for later this year with no new hardware in sight, such as a new 3D Mario, that makes new hardware coming any time soon look a lot less likely. I do not see Nintendo rolling out their successor with no Zelda or 3D Mario at launch. Nintendo could have a very strong year without new hardware if they release a few big hitters in the second half, but at some point Nintendo has to start staging software for the next hardware, unless of course that hardware isn't coming anytime soon. If that is the case, this will end up being another blunder for Nintendo, because Switch sales are going to start to fall pretty hard very soon.

That makes sense. It's especially true with Nintendo, as we very well know about their tendency to "sit" on finished games to release them when they feel like it. They could have just done that a bit more to have big releases for H2 2023, not sure which games though. Maybe Bayonetta Origins and Kirby?

 

[Dakhil]

https://www.ft.com/content/25ea2570-264d-472d-8ef8-3a02003e87ab

Arm is seeking to raise prices for its chip designs as the SoftBank-owned group aims to boost revenues ahead of a hotly anticipated initial public offering in New York this year.

The UK-based group, which designs blueprints for semiconductors found in more than 95 per cent of all smartphones, has recently informed several of its biggest customers of a radical shift to its business model, according to several industry executives and former employees.

These people said Arm planned to stop charging chipmakers royalties for using its designs based on a chip’s value and instead charge device makers based on the value of the device. This should mean the company earns several times more for each design it sells, as the average smartphone is vastly more expensive than a chip.

The changes represent one of the biggest shake-ups to Arm's business strategy in decades, at a time when SoftBank chief executive Masayoshi Son is seeking to drive up Arm's profits and attract investors to its impending return to the public markets.

"Arm is going to customers and saying 'We would like to get paid more money for broadly the same thing'," said one former senior employee who left the company last year. "What SoftBank is doing at the moment is testing the market value of the monopoly that Arm has."

SoftBank, which acquired Arm for £24.3bn in 2016, plans to retain a majority stake following the IPO. It was aiming to start pushing through the pricing shake-up at the chip designer by as early as next year, but has so far been frustrated by customers' reluctance to accept the new arrangement.

MediaTek, Unisoc and Qualcomm, and multiple Chinese smartphone makers including Xiaomi and Oppo, are among the companies that have been made aware of the proposed change to pricing policy, according to several people familiar with the talks.

The Cambridge-headquartered company licenses its designs to various chipmakers for them to use to make the semiconductors found in smartphones, computers and cars. It charges a licence fee for obtaining its blueprints, and then a recurring royalty for each chip shipped.

Arm has also become more aggressive in pushing price increases within its existing sales model for royalties and licences over the past year, particularly for customers making chips for smartphones, where it has a dominant market position, according to people with knowledge of the recent moves.

Arm has been locked in a bitter legal battle with its largest customer, Qualcomm, since the second half of last year, accusing the mobile chipmaker of using its intellectual property without permission. In its counterclaim, the chipmaker alleged that Arm had told "one or more of Qualcomm's customers" that it would cease licensing central processing units (CPU) to all chip companies and would only provide licences to device makers themselves.

According to the new business model being presented by Arm, royalties would be set according to the average selling price (ASP) of mobile devices rather than that of the chips. The changes will mainly involve Arm's most prominent "Cortex-A" designs, essential for the development of smartphone processors.

Charging based on device price is a widespread practice across the telecoms equipment market, with Qualcomm, Nokia and Ericsson all using a similar model for their patents. The problem for Arm is that it is attempting to change its pricing strategy long after it established a different sales model.

The average price for a smartphone computing chip is about $40 for Qualcomm, $17 for MediaTek and $6 for Unisoc. Arm charges royalties of about 1 to 2 per cent of the value of each chip sold based on its designs, according to Sravan Kundojjala, an analyst at TechInsights. By contrast, the average smartphone sold for $335 in 2022. While it is unlikely Arm would seek as much as 1 to 2 per cent of the value of each device, those familiar with the matter said the company would set its new pricing in a way that significantly increases overall earnings.

"The [royalty] amount will be at least several times higher than what Arm gets now," said an executive from a leading Chinese smartphone maker which has so far refused to back the proposed plan. "We are told that they hope such changes could start from 2024."

Some of Arm's customers, including Apple, are both chipmakers and device makers, and have special licensing and royalty agreements with Arm. The iPhone maker is not involved in discussions about the change to Arm's business model, said executives with knowledge of the company’s recent discussions.

Arm, SoftBank, Qualcomm, MediaTek, Unisoc, Xiaomi and Oppo declined to comment.

Son is relying on a blockbuster Arm IPO to help mount a turnround at SoftBank, which has suffered heavy losses over the past year as the value of its tech investments was hammered in a broader industry downturn.

Son announced last year that he would step back from day-to-day operations at SoftBank to devote himself to turbocharging Arm's growth. People close to the billionaire said he felt Arm had taken a smaller slice of the industry profits over the past decade than chipmakers such as Nvidia, Broadcom and Qualcomm, especially considering how crucial and ubiquitous its designs were in mobiles.

Investors in the US, UK and Japan have told the Financial Times they have applied valuations to Arm of between $30bn and $70bn. They say the broad range stems from the difficulty in identifying any comparable companies, and a lack of clarity over the company’s precise growth strategy in recent years.

For several years under SoftBank’s ownership, Arm posted stagnating revenue and falling profits. Arm's costs increased from $716mn in 2015 to $1.6bn in 2019, according to SoftBank data. Revenues gained 20 per cent to $1.9bn over the same period, while profits fell almost 70 per cent to $276mn by 2019.

It has since reversed its fortunes, posting a 35 per cent rise in revenues to $2.7bn in 2021, the latest date for which there are annual figures, and a 68 per cent rise in adjusted earnings to $1bn.

Additional reporting by Qianer Liu in Hong Kong and Leo Lewis in Tokyo

I can't see Nintendo being happy about being forced to pay twice: pay Nvidia to design an Arm based SoC for Nintendo, and pay Arm a device royalty fee.

And I have to wonder: are companies paying Arm device royalty fees going to be prohibited by Arm to buy SoC designs with a mix of Arm IP and non-Arm IP, and are forced to only buy SoC designs with only Arm IP? (I'm going by Qualcomm's claims.) Nintendo could definitely be in trouble after 2024 if that's the case. (I suppose one solution is for Nintendo to ask Nvidia to design a custom Arm based CPU for future SoCs.)

 

[Foltzie]

This pricing structure makes no sense. It would benefit cheap phones with high performance SoCs but lacking other features. SBCs would be wildly priced, with cheap SoCs not making sense anymore. This structure doesn't make sense to me.

As the article points out, ASP of the final device is common for patents and other licenses. Still, changing terms where they are now based on chip ASP, to device ASP, while offering no additional value is ballsy. I'm curious if it invites antitrust scrutiny as a result.

I dont think we'll see manufacturers trying to end run around the license change by selling high powered, but cheaper devices since it would impact their own profit margin.

Softbank bought ARM with an intention to make a profit by selling it, so when the Nvidia deal was blocked I expected something like this in a lead up to an IPO.

I wonder if Google, Microsoft, Qualcomm, & others should have considered banding together to buy ARM and leave it as a neutral separate company.

 

[Dakhil]

I wonder if Google, Microsoft, Qualcomm, & others should have considered banding together to buy ARM and leave it as a neutral separate company.

Good luck trying to convince regulators, especially regulators in China, to approve such an acquisition, especially with the geopolitical situation between the US and China.

 

[ILikeFeet]

Probably something neat using tensor cores. Maybe Nintendogs with learning dog AI.

don't see why you need tensor cores for that. a sufficiently advanced state machine in a finite options world would give you the same results

 

[kvetcha]

don't see why you need tensor cores for that. a sufficiently advanced state machine in a finite options world would give you the same results

I want a dog that can convincingly simulate rabies

 

[Foltzie]

Good luck trying to convince regulators, especially regulators in China, to approve such am acquisition, especially with the geopolitical situation between the US and China.

I think US regulators might have been on board with such a deal, and ARM China is already largely owned by China. I dont know if China would have have been in a position to delay the deal.

Still, it's moot, it wouldn't have happened as it would have required a bunch of companies to agree to be neutral on an acquisition and not try to position themselves over one another in the deal.

 

[iag]

Maybe it’s time for Nintendo to invest in RISC-V?

 

[Thraktor]

This a great analysis! I'm unsure about the choice of fitting an exponential formula to the power curve, however: wouldn't a quadratic relation be more physically plausible? I believe that power scales quadratically with frequency. Is there a way to test a quadratic fit (how did you get the data points btw? If it's an open source software tool, I'd love to play around with it)?

A quadratic fit could change the analysis you presented as well, so it's an important thing to consider I think.

That's a fair point. The power estimator tool is here, by the way (I'm not sure if you need a Nvidia developer account to access). To get the numbers I used, I switched to Jetson AGX Orin 64GB (as it allows the largest range of clock speeds), set the GPU load level to high, and took power measurements at each available clock speed for both the 4 TPC and 8 TPC configuration, then disabled the GPU to get a baseline reading with which to isolate the GPU power draw. I took the per-TPC power draw for each of the 4 TPC and 8 TPC configurations, and then took the average of the two as my figure for each clock speed (although the difference between the two was very small in any case).

As for my reason for fitting an exponential formula, I wanted to choose the simplest representation (ie fewest variables) which provided a good fit. I only have nine data points, so over-fitting the data is a bigger concern than attempting to most accurately represent the underlying phenomena. An exponential fit provided a very high level of fit (R^2=0.9993) with two independent variables. We can also get a quadratic with a very good fit (R^2=0.9992), but moving from two to three independent variables significantly increases the risk of over-fitting when we've only got nine data points. FYI, here's the quadratic form:

P = T*4.5832C^2 - 2.7592C + 1.3446

If we look at a graph of this, we can see that it slopes up below 300MHz, implying that as clocks reduce below 300MHz, overall power consumption goes up, which is obviously not what we should expect. This is a clear instance of over-fitting, and occurs only slightly outside our data set (the lowest data point I have is 420MHz). An exponential form has the benefit of guaranteeing that power consumption increases monotonically w.r.t. clocks at all points of the curve, which means we get sensible behaviour outside our data set.

I'd also argue that, if we really want to properly represent behaviour at low clock speeds, assuming a quadratic relationship between power consumption and clock speed would actually be an oversimplification itself. Power consumption is more accurately quadratic with respect to voltage, and linear with respect to clock speed. If voltage increases linearly w.r.t. clock speeds, then we'd be fine making such a simplifying assumption, but we're actually looking at exactly the point where that assumption breaks down. Specifically, there is a minimum voltage required for the chip to function, and we should expect the peak efficiency clock speed to occur at or around the place in the power curve where the voltage bottoms out. Or, to put it another way, we should expect the peak efficiency clock to be the maximum clock which can be achieved with the minimum voltage.

To illustrate this, here's the basic formula we should expect power consumption to follow (excluding the Poole-Frenkel effect):

P = a * V^2 + b * C * V^2

Where P is overall power, C is clock speed, V is voltage, and a and b are fixed variables reflecting the properties of the chip. The first part of the formula is the static power consumption, and the second part is the dynamic power consumption.

Let's assume that voltages are completely flat up to some clock speed X, and above X increase linearly in proportion to clock speed. In the region below X, we have a fixed static power consumption, and then a dynamic power consumption that's proportional to clock speed. So long as static power consumption is greater than zero (a pretty safe assumption), power efficiency increases in this region as clock speeds increase. In the region above X, we have both static and dynamic power consumption increasing proportional to the square of the voltage, which is increasing with clock speed, so it's pretty clear that power efficiency in this region decreases as clock speed increases. So, if we've got a lower region where efficiency is increasing, and a higher region where it's decreasing, the peak efficiency is obviously at the boundary of those two regions, ie at clock speed X.

Of course this doesn't mean that an exponential curve accurately represents this behaviour properly either, but I believe it's the least-worst balance of representing the data set while avoiding over-fitting. The specific peak efficiency clock it provided shouldn't be taken as gospel, but is probably in the ballpark for Orin on 8nm (I had previously speculated that it was 420MHz, as that's the lowest clock speed that Nvidia supports). The peak efficiency clock for 5LPP definitely shouldn't be taken as gospel, and if anything should just be taken as evidence that you shouldn't treat node changes as flat reductions in power consumption.

 

[Dakhil]

I think US regulators might have been on board with such a deal, and ARM China is already largely owned by China. I dont know if China would have have been in a position to delay the deal.

Still, it's moot, it wouldn't have happened as it would have required a bunch of companies to agree to be neutral on an acquisition and not try to position themselves over one another in the deal.

I think there's a possibility that the US government hypothetically only approves the acquisition if the companies that plan to acquire Arm don't sell high end Arm IP to companies based in China, especially with the current geopolitical situation between the US and China.

And China was definitely one of the reasons Nvidia's attempted acquisition of Arm failed.

 

[KeijiKG]

https://www.ft.com/content/25ea2570-264d-472d-8ef8-3a02003e87ab

I understand ARM is trying to protect its bottom line, but just looks like micromanaging to me.

 

[Foltzie]

And China was definitely one of the reasons Nvidia's attempted acquisition of Arm failed.

I thought the UK killed the deal.

 

[Dakhil]

I thought the UK killed the deal.

The UK government's definitely one of the reasons Nvidia's attempted acquisition of Arm failed, but not the only reason. (Hence, why I said China's one of the reasons Nvidia's attempted acquisition of Arm failed.)

 

[Concernt]

Probably something neat using tensor cores. Maybe Nintendogs with learning dog AI.

How about a "learning world"? A world that morphs and changes its lighting and textures as you play, reacting to you. As you jump, it grows more vertical, the harder you try, the harder it gets. Nothing extravagant, maybe an escape-the-room type deal where you have to "manipulate" the AI into letting you out.

 

[mjayer]

Am I crazy or are the graphics in this trailer just... really really good? I know this is done all the time but... but... are they too good?

 

[Thraktor]

[snip]

Actually, this is probably best illustrated by ignoring all the formulae and just looking at the data itself. Here's the raw data:

Clock      W per TPC
0.42075    0.96
0.52275    1.14
0.62475    1.45
0.72675    1.82
0.82875    2.21
0.93075    2.73
1.03275    3.32
1.23675    4.89
1.30050    5.58

From these numbers, we can easily calculate the power efficiency for each point, measured in Gflops per W:

Clock    W per TPC    Gflops/W
0.42075    0.96       223.8
0.52275    1.14       233.9
0.62475    1.45       220.4
0.72675    1.82       205.0
0.82875    2.21       192.1
0.93075    2.73       174.3
1.03275    3.32       159.2
1.23675    4.89       129.4
1.30050    5.58       119.4

The efficiency is higher at 522MHz than it is at 420MHz, and then drops off again at 624MHz and above. Therefore, the peak efficiency clock should be somewhere between 420MHz and 522MHz.

 

[ShadowFox08]

That's one I'm not sure about. On the memory controller side Nvidia just announced at GTC that their Grace CPU is sampling, which means they've already got a product manufactured on TSMC 4N with an LDDR5X controller in the hands of partners. I believe Grace was designed by the Tegra group, like T239, so conceivably if T239 were manufactured on 4N there shouldn't have been any reason not to use it across both products.

What I'm more concerned about is LPDDR5X part cost and availability. We're only just seeing it in phones, and the smallest capacity I've seen is 8GB (with a 64-bit interface), which means for a 128-bit bus you've got a minimum of 16GB, which likely pushes up the price quite a bit compared to what seems to be a 12GB baseline using standard LPDDR5 parts. Then again, I don't know if this is just because LPDDR5X hasn't made it to mid-range and low-end phones yet. In 2019, when LPDDR4X was around a year and a half old, Nintendo switched from LPDDR4 to 4X, without even using the higher clock speeds. Since then LPDDR4X seems to have completely displaced LPDDR4, so it seems likely that Nintendo were aware that 4X would be a better choice for availability long-term, and that's what informed the change. If the memory industry expects 5X to completely displace LPDDR5 over the next few years, then we could see the same thing here.

Crazy thing also is that even if we do get lpddr5x on t239, there's a 50% chance they won't use the extra bandwidth, but rather towards saving power draw. God I hope they future roof this thing thing though.

Edit: I'm also a little skeptical about those theoretical specs (3.45 tflops GPU, ~1.7 CPU 8 core) being achievable at 11 watts docked on 5nm Samsung as well, especially when 5nm Samsung performs identically to as 7nm TSMC, and comparing it to steam deck (yeah I know, CPU clocks won't be nearly ther high). That would also give the more reason to use lpddr5x for saving power draw, over the 33% bandwidth speed. Not holding my breath on lpddr5x on 5nm Samsung though.

Could we really even have an acceptable battery life at those speeds at 11 watts? At the very least Nintendo would aim to match v1 switch battery life, and then followed up with a revision (3nm) .

I really hope its it's 4nm TSMC though. But I'm trying not to think about it too much..

 

[Concernt]

Am I crazy or are the graphics in this trailer just... really really good? I know this is done all the time but... but... are they too good?

No, but also, yes. Maybe. We'll see.

 

[KeijiKG]

Am I crazy or are the graphics in this trailer just... really really good? I know this is done all the time but... but... are they too good?

I don't see what you're getting at.
Just looks like any other LEGO game.

 

[Thraktor]

Am I crazy or are the graphics in this trailer just... really really good? I know this is done all the time but... but... are they too good?

It's probably PC/PS5/Series X footage, and they're just using the same footage for all announcement videos.

Crazy thing also is that even if we do get lpddr5x on t239, there's a 50% chance they won't use the extra bandwidth, but rather towards saving power draw. God I hope they future roof this thing thing though.

I definitely don't expect full memory clocks in portable mode (even if we're just talking LPDDR5), but if they are using 5X I'd be surprised if they didn't support full clocks in docked mode. In terms of overall performance, they're probably better off pushing a little bit extra power to RAM to improve bandwidth than they are using the same power to increase GPU clocks.

 

[Bosintang]

Am I crazy or are the graphics in this trailer just... really really good? I know this is done all the time but... but... are they too good?

Considering how secretive and careful Nintendo is concerning their new hardware (and about everything honestly), I really don't believe that they would let a random third party game showing the first Drake-powered footage.
I'm voting PS5 footage or PC.

 

[LukasManak22]

Even at only having Nintendo published titles dated up until July, only Pikmin 4 and TOTK are actually from EPD, and TOTK is only where it is due to multiple delays. That sounds extremely light for a year or EPD releases.

I don't see many reasons why they'd be quiet if there are indeed more, outside of a new console that they'll be coming out on.

The only EPD releases in 2018 were LABO kits. The "second half titles are not announced because of the new console" are the least proof of a new console imo. 2D Mario announcement in June for Holiday is still a decent lineup (with TotK selling millions till Holidays).

 

[JoshuaJSlone]

so theoretically do we think this thing could manage 4k resolution in docked mode, but everything is essentially on the lowest settings possible depending on the game?

"Depending on the game" is pretty big. For some games it should do 4K at high settings. For some 4K at low settings. For some 4K would be infeasible. For some any resolution would be infeasible.

How about a "learning world"? A world that morphs and changes its lighting and textures as you play, reacting to you. As you jump, it grows more vertical, the harder you try, the harder it gets. Nothing extravagant, maybe an escape-the-room type deal where you have to "manipulate" the AI into letting you out.

So far this seems like a set of if-thens that could be done on most gaming hardware from the last several decades in some form.

 

[Kenka]

To all of you who advocated higher clocks for Drake than the ones we reasonably assumed until now.

@Thraktor just saved your butts.

However, this peak efficiency is too low for what we can expect from docked mode, right?

 

[Skittzo]

To all of you who advocated higher clocks for Drake than the ones we reasonably assumed until now.

@Thraktor just saved your butts.

However, this peak efficiency is too low for what we can expect from docked mode, right?

Efficiency is much less important for docked mode for obvious reasons. They'll still want to avoid thermal issues but not needing to rely on a battery will help.

 

[Concernt]

The only EPD releases in 2018 were LABO kits. The "second half titles are not announced because of the new console" are the least proof of a new console imo. 2D Mario announcement in June for Holiday is still a decent lineup (with TotK selling millions till Holidays).

I have my doubts that 2D Mario will be announced before the next system, if at all.

 

[LukasManak22]

I have my doubts that 2D Mario will be announced before the next system, if at all.

The fact that number of Splatoon 2 designers were missing from 3 just leads me to believe that its coming. 3D as a launch title, 2D for this Switch, easy title to sell especially after Mario Movie.

 

[Bosintang]

I have my doubts that 2D Mario will be announced before the next system, if at all.

Oh be prepared to be surprised

 

[Deleted member 887]

To all of you who advocated higher clocks for Drake than the ones we reasonably assumed until now.

@Thraktor just saved your butts.

However, this peak efficiency is too low for what we can expect from docked mode, right?

I will note that halfway between 430 and 522 is 476 - and that the higher docked mode profile that Breath of the Wild uses is 460MHz. 2.25x that is 1071MHz. I would call that the high end of my usual pessimistic estimates and the low end of the optimistic ones, which sounds about right.

And yeah, you don't have to live in peak efficiency in docked mode in the same way. Heat still matters, but you can afford to be less efficient. The point that @Thraktor (and others) have been hitting since even before the leak, is why would you go with so many SMs? And the answer is obviously, more performance, but they could also do that with higher clocks. In the case of a mobile device, you add more silicon because it is cheaper, power wise, than pushing clocks.

By knowing that the "maximally efficient clock" is well past that 300MHz range helps us dial in both where the clock will land but also, the node. 12SMs always felt like a lot, but if peak efficiency was closer to 300MHz, then maybe it really makes sense to go with that many on 8nm. But since it's closer to 500MHz, it really feels like Nintendo is on a different node. Not just "better", but different, a place where 12SMs isn't just possible, but where 12SMs is the maximally power efficient path to their perf goals.

 

[Concernt]

Oh be prepared to be surprised

Source?

 

[Concernt]

The fact that number of Splatoon 2 designers were missing from 3 just leads me to believe that its coming. 3D as a launch title, 2D for this Switch, easy title to sell especially after Mario Movie.

That sounds plausible, though I would expect both to come out for, well, both.

 

[LukasManak22]

That sounds plausible, though I would expect both to come out for, well, both.

You just have to trust the process. EPD10 delivering 2 games in a short period of time isn't new. New Super Mario Bros. U - November 18th 2012, Pikmin 3 - July 13th 2012. In fact, it would be quite funny if they mirror it this year. Pikmin 4 in July and new 2D Mario in November.

 

[Concernt]

You just have to trust the process. EPD10 delivering 2 games in a short period of time isn't new. New Super Mario Bros. U - November 18th 2012, Pikmin 3 - July 13th 2012. In fact, it would be quite funny if they mirror it this year. Pikmin 4 in July and new 2D Mario in November.

As I said! Plausible! Especially around the launch of a new console.

 

[ILikeFeet]

Nvidia shown off a use case for the tensor cores today, though it's not for gameplay but for rendering. it's also more path tracing based, but it's being implemented in RTXGI 2.0, so a game using it isn't too crazy

NVIDIA #GTC2024 Conference Session Catalog

Register now. Streamed online. March 18-21, 2024.

www.nvidia.com

they also shown off their latest RTX Showcase (no release date) that heavily leverages RTXDI to have 4700 shadow casting lights with minimal performance loss. I definitely expect restir-based techniques to get a lot of use this generation since it's a technique that's been shown to be quite scalable. I can't wait for the demo's release so we can see just how scalable it is

NVIDIA #GTC2024 Conference Session Catalog

Register now. Streamed online. March 18-21, 2024.

www.nvidia.com