They could be but it would use less energy, need less overall materials, ram and SSD prices would go down. Even if the APU price stayed the same more or less, they'd be able to cut a lot from the price
-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
TSMC: Better yield on 5nm than 7nm, first products are expected later this year. 3nm is planned to enter volume production in 2H22.
- Thread starter kostacurtas
- Start date
- News
Source: TSMC will start mass production of 3nm chips in the second half of 2022 - (gizchina.com)
Taiwanese chipmaker, TSMC, will commence the mass production of 3nm chips in the second half of 2022. This is the latest report with respect to the development of TSMC's 3nm manufacturing process. In addition, the monthly production capacity of this process will be 55,000 pieces. According to TSMC's Chairman, Liu Deyin, the number of employees in the Tainan Science Park will reach about 20,000 when the 3nm chip starts mass production. The current number is 15,000.
SAMSUNG TO MASS-PRODUCE 3NM CHIPS IN 2022: STRIVING TO OVERTAKE TSMCSamsung and TSMC are the only two companies that have the capacity to make chips with the 5nm process. However, Samsung seems to be in the shadow of TSMC but it is putting up a good fight. According to recent reports, Samsung Electronics is struggling to catch up with TSMC. The South Korean manufacturing giant plans to mass-produce 3nm chips in 2022.
Samsung Electronics executive Park Jae-hong said at an event recently that the
company has set a goal to mass-produce 3-nanometer chips in 2022. The executive revealed that the company is already developing preliminary design tools with major partners.
Presently, it has invested $116 billion in its next-generation chip business. This includes manufacturing chips for external customers.
Nvidia is going to make their SoC.
There's no indication otherwise.
Wishful thinking mostly. I would expect Nintendo to opt for a cheaper older process for their next gen console too. But tbf by the time they will launch it this process will probably be current TSMC's cutting edge N7 or N6.
Nintendo should be on 5nm for their next gen.
Think that was just Jen-Hsun hyping for investors iirc. Though now they own ARM I don't see Nintendo going elsewhere for at least next gen.
For years node names have been more marketing then anything. many years only part of the node process would actually be at the announced size. Intel was the last company who would only call a node say 15nm if everything was 15nm. Dunno if they still do that though. So new nodes are always smaller, but maybe not so much as you'd think by the announcement.
Think that was just Jen-Hsun hyping for investors iirc. Though now they own ARM I don't see Nintendo going elsewhere for at least next gen.
For years node names have been more marketing then anything. many years only part of the node process would actually be at the announced size. Intel was the last company who would only call a node say 15nm if everything was 15nm. Dunno if they still do that though. So new nodes are always smaller, but maybe not so much as you'd think by the announcement.
im glad intel finally gets kicked in the butt. they have basically stagnated for a decade. now everybody is owning them.
Yea looking it up just says it could last two decades. No guarantee.
Actual gate sizes hasn't been as small as the process name suggests since around 16nm.
"N3" isn't really 3nm in gate size either, it's bigger.
This is why TSMC's "7nm" process is more or less similar to Intel's 10nm.
It's also why it has been seemingly easy for TSMC to go past their 7nm - their 5nm and 3nm aren't really this small.
With the consoles and AMD's line of products + Apple silicon, is there enough capacity for Nvidia ? I think they would've used TSMC 7nm already in Ampere if they could.
This is all your headcanon. Switch was cutting edge in 2017 and Nvidia gaming revenues have increased substantially to record levels thanks to Switch SoC sales. There are huge financial incentives to keep the partnership going.
No manufacturing process from any semiconductor manufacturer is true to the Gate Pitch and haven't been since a long time(Since 180nm). It just stayed for marketing reasons. Also, Intel 10 nm is comparable to TSMC 7nm because Intel 14 nm was already a much more dense process than any other manufacturer. That's why(Coupled with other factors) Intel had so much trouble. They wanted a 2.7x process scaling, but hit hard limits. TSMC was much more conservative and went with 1.7/1.8x node jump. Their 5nm, 3nm and the future 2nm are much more advanced than anything else, state of art manufacturing. Intel 7nm is supposed to launch in 2021/2 and is only comparable to TSMC 5nm.
I love my ps5 and like it's wacky look but it's big AF. I'm hopeful 3nm, if on track for late 2022, means we get a slim ps5 in time for Sept. 2023. I know the price wouldn't go down much for the APU, but I'd expect it to go down at least a tiny bit but most of all they would save money on parts like the case, heat sink, power supply, etc.
How much do we think 3nm will save energy wise at the same performance vs 7nm? I would hope 40%+
How much do we think 3nm will save energy wise at the same performance vs 7nm? I would hope 40%+
I don't know if we're going to see a revised PS5 or Pro or Xbox Series Ultra. The economics of manufacturing changed a lot. But, answering your question, TSMC 3nm is supposed to have a 25-30 % energy consumption reduction compared to 5nm. I don't remember now which manufacturing node consoles are(N7/N7+/N7P), but your prediction of 40~50% is more or less right on money.
So what you're saying is, if Intel manages to get their 10nm processor out the door, they will have caught up (and possibly surpassed) the industry?
There will almost certainly be a slim version of the PS5.
I'm not the poster, but no. Intel is outdated. TSMC 7nm(Which is comparable in logic density to Intel 10nm) is already old news. TSMC is manufacturing 5nm chips right now for the iPhones 12 and Huawei Mate(I think it's called that?) which is much more advanced than Intel 10 nm and is supposedly comparable to Intel 7nm which will be(supposedly again) available in 2021/2022. By this time, TSMC and Samsung are going to start manufacturing of 3nm/N3 process and TSMC development of 2nm/N2 process is going smoothly.
I think we will, tho certainly the timeline can change and it may not follow the 3 year routine Sony has.
with the world economy hopefully recovering from COVID by 2023, I think they will definitely have a cost cutting model in 23-24. The APU tho may not cut too much costs of course.
they might pay the same they do today (let's say they pay $125 per APU) even if it's 3nm, but I imagine they can cut costs on the case, size, PSU size, Heatsink and fan size, while switching to less memory chips on the board, and of course SSD prices should hopefully come down a decent amount as they become more mainstream.
I could see the Cost of 3nm pushing budget SKUs midgen to $349 instead of $299 tho.
I guess so, if going by past generations trends. But i'm not so certain this time. Microsoft lecturing regarding manufacturing at Hot Chips and the Eurogamer interview regarding Xbox Series S were an eye opener.
Their 10nm processors are out of the door for some time now, just not for the desktops.
And yes, they are roughly on par with TSMC's 7nm Zen2/3 in transistor density at least.
A lot of factors. Design margin, defect density, excursions at fab, etc.
Did you find anything decent on this topic? I'd love to see something like that.
The whole industry should just agree to leave nanometers behind and measure improvements in transistor density terms.
The tegra pricesor was released on 2015....
We are not taking how successful is the switch but about TSMC roadmap.
Semi conductors was my most hated and fascinating class in electrical engineering. Fucking pain because this thing is nearly witchcraft and the teacher was basically sadistic, but super fascinating concepts and scale.
EDIT :
And outside of just tools to make it happen, which is fascinating on it's own, what i always found interesting in engineering/maths is the foundation, the mathematical model that made all of this possible.
Schrödinger's wave equation, in 1925.
Schrödinger's wavefunction, simplified to a time-dependent equation in one dimension (eigenvalue equation). With quantum mechanics, unlike three basis vectors in real splace, these eigen functions span an infinite number of dimensions in an abstract space, Hilbert space. Almost impossible to solve.. unless you have a potential energy well, with infinite walls (as would be seen from an electron, ohhhh). This becomes a "particle in a box". Controlling this tunneling, is effectively the physics and probabilities of controlling the hyper complex electron waveform of the electron, consistently. A purely mathematical model of the transistor.
They solved the equations for Hydrogen first, the "easiest" part, to introduce 4 quantum numbers, dictated by quantum mechanical rules, and extrapolated these ideas/solutions to Silicon. Way beyond my own comprehension to be honest, i did not got there. But anyway, 95 years later, we have the world's library, communicate through air, take megapixel photos on a wallet size device.
Schrödinger was not human..
EDIT :
And outside of just tools to make it happen, which is fascinating on it's own, what i always found interesting in engineering/maths is the foundation, the mathematical model that made all of this possible.
Schrödinger's wave equation, in 1925.
Schrödinger's wavefunction, simplified to a time-dependent equation in one dimension (eigenvalue equation). With quantum mechanics, unlike three basis vectors in real splace, these eigen functions span an infinite number of dimensions in an abstract space, Hilbert space. Almost impossible to solve.. unless you have a potential energy well, with infinite walls (as would be seen from an electron, ohhhh). This becomes a "particle in a box". Controlling this tunneling, is effectively the physics and probabilities of controlling the hyper complex electron waveform of the electron, consistently. A purely mathematical model of the transistor.
They solved the equations for Hydrogen first, the "easiest" part, to introduce 4 quantum numbers, dictated by quantum mechanical rules, and extrapolated these ideas/solutions to Silicon. Way beyond my own comprehension to be honest, i did not got there. But anyway, 95 years later, we have the world's library, communicate through air, take megapixel photos on a wallet size device.
Schrödinger was not human..
Last edited:
Tegra was not mature tech. In 2015 it was very advanced. I mean if you're goign by year PS4 used 2011 CPUs and 2012 GPUs in their late 2013 consoles. But even that isn't fair to the consoles. But I always only hear the 2015 figure mentioned in reference to the Switch, I wonder why.
If you're talking process nodes
Nintendo apparently did get a killer deal on the outdated 20nm process for the launch because Nvidia had excess unsued capacity but once that was eaten up (the 20 million chips) they shifted to the smaller node which coincided with the new Nintendo Switch and Switch Lite launching a year later. nvidia/tsmc and probably Nintendo themselves couldn't wait to get off that 20nm node.
Oh come on. Stop acting like Nintendo is some kind of victim here. People complained all the time about the POS Jaguar APUs in the PS4 and Xbox One. And then you act like we should be celebrating process node shrinks that happen with pretty much every console in the last 20 years when all they yielded was greater energy efficiency.
The Switch used dated technology at launch. In a configuration that bottlenecked the system. We know that. Let's move on, and hope for a more balanced system in the succ. It won't be hard for Nintendo.
That would be true if it was a 2018 console, when it was in development the chip was the best one they have access to. We have endless discussions about this in the Switch Pro/2 discussion.
It was only as dated as any other console that needs 1-2 years to come together using primarily existing designs already out in the market, which its competitors pretty much followed. So singling them out is certainly something I found worth pointing out.