regardless of the node, this will be a massive jump based on the specs we know. with a better node, the floor is just higherI’ve read your comment like 5 times now. What does the switch times 1.5 mean? Is this just sarcasm?
regardless of the node, this will be a massive jump based on the specs we know. with a better node, the floor is just higherI’ve read your comment like 5 times now. What does the switch times 1.5 mean? Is this just sarcasm?
Depending on how the nanometer stuff pans out, at the extreme ends we might be looking at "Ridiculously more powerful than 2017 Switch" and "Ridiculously more powerful than 2017 Switch times 1.5".
We all know Nintendo can market it however they want, and they can squeeze every ounce of power to make their first party games look amazing, but I’m wondering about the possibility of “miracle” PS5/XSX ports on 8nm Drake.
A die shrink is very unlikely for 8nm.Ah well. I’ll just wait and see. My last thought is if 8nm Drake gets the “Mariko“ treatment and gets a die shrink later on, it won’t be for the benefit of more power but to battery life again. Versus if they did 6nm now, then 3nm in a few years.
I know that’s the conventional wisdom around here, but is a shrink to euv really that much worse than the Mariko shrink to finfet? Is there a source on this?A die shrink is very unlikely for 8nm.
I don't know how much harder it is to shrink to euv and I'm a layman in the field, but even if it's a lot more, NVidia is using AI to reduce die shrink labor by a huge margin, so I wouldn't assume die shrinking from 8nm now would be more expensive to NVidia than Erista's die shrink was at the time.I know that’s the conventional wisdom around here, but is a shrink to euv really that much worse than the Mariko shrink to finfet? Is there a source on this?
Essentially, it's a way of automating the process of migrating a cell, like a fundamental building block of a computer chip, to a newer process node.
...
The tool Nvidia uses for this automated cell migration is called NVCell, and reportedly 92% of the cell library can be migrated using this tool with no errors. Then 12% of those cells were smaller than the human-designed cells.
"This does two things for us. One is it’s a huge labor saving. It’s a group on the order of 10 people will take the better part of a year to port a new technology library. Now we can do it with a couple of GPUs running for a few days. Then the humans can work on those 8 percent of the cells that didn’t get done automatically. And in many cases, we wind up with a better design as well. So it’s labor savings and better than human design."
The bare minimum for low 1080p+60 is a CPU that’s weaker than the ones in the consoles so it isn’t that.It's kind of blowing up right now that Gotham Knights will be limited to 30fps on PS5 and Series S/X. Since their statement says it's more to do with having a big multiplayer open world rather than something they can work around by lowering resolution, sounds like a CPU limitation. Bringing this up as it seems like a rare example of a current game where a Drake version might not be viable.
I know that’s the conventional wisdom around here, but is a shrink to euv really that much worse than the Mariko shrink to finfet? Is there a source on this?
The issue isn’t so much the shrinking process. The issue is that nvdidia has no other product at the Samsung foundry of their newer nodes. The only other foundry in which Nvidia has a product on is at TSMC. If Nvidia is going to shrink the Drake chip to a more recent node, it stands to reason that the most likely scenario is another place in which Nvidia products are, just like the TX1 followed with the Pascal GPUs.I don't know how much harder it is to shrink to euv and I'm a layman in the field, but even if it's a lot more, NVidia is using AI to reduce die shrink labor by a huge margin, so I wouldn't assume die shrinking from 8nm now would be more expensive to NVidia than Erista's die shrink was at the time.
⋮
The semiconductor industry is near the limit. It's near the limit in the sense that we can keep shrinking transistors but we can't shrink atoms — until we discover the same particle that Ant Man discovered. Our transistors are going to find limits and we're at atomic scales. And so [this problem] is a place where material science is really going to come in handy.
A great deal of the semiconductor industry is going to be governed by the advances of material sciences, and the material sciences today is such an enormously complicated problem because things are so small, and without a technology like artificial intelligence we're simply not going to be able to simulate the complicated combination of physics and chemistry that is happening inside these devices. And so artificial intelligence has been proven to be very effective in advancing battery design. It's going to be very effective in discovery and has already contributed to advancing more durable and lightweight materials. And there's no question in my mind it is going to make a contribution in advancing semiconductor physics.
When something dies? It might be reincarnated, but it dies. The question is, what's the definition of Moore’s law? And just to be serious, I think that the definition of Moore's law is about the fact that computers and advanced computers could allow us to do 10 times more computing every five years — it's two times every one and a half years — but it's easier to go 10 times every five years, with a lower cost so that you could do 10 times more processing at the same cost.
Nobody actually denies it at the physics level. Dennard scaling ended close to 10 years ago. And you could see the curves flattened. Everybody's seen the curves flatten, I'm not the only person. So the ability for us to continue to scale 10 times every five years is behind us. Now, of course, for the first five years after, it's the difference between two times and 10 times — you could argue about it a little bit and we're running about two times every five years. You could argue a little bit about it, you can nip and tuck it, you could give people a discount, you could work a little harder, so on and so forth. But over 10 years now, the disparity between Moore's law is 100 times versus four times, and in 15 years, it's 1,000 times versus eight.
We could keep our head in the sand, but we have to acknowledge the fact that we have to do something different. That's what it's really about. If we don't do something different and we don't apply a different way of computing, then what's going to happen is the world's data centers are going to continue to consume more and more of the world's total power. It's already noticeable, isn't that right? It means the moment it gets into a few percent, then every year after that, it will [continue]. Every five years it will increase by a factor of 10.
So this is an imperative. It's an imperative that we change the way we compute, there's no question about it. And it's not denied by any computer scientists. We just have to not ignore it. We can't deny it. And we just have to deal with it. The world's method of computation cannot be the way it used to be. And it is widely recognized that the right approach is to go domain by domain of application and get accelerated with new computer science.
⋮
Leaks aren’t guaranteed from production, though we’ve had several possible production leaks already. It’s just that until the product releases we won’t know which are valid.The lack of news is disheartening. Assuming a May launch, when should production (and therefore leaks) start?
I think it's possible, if not likely, manufacturing has already begun in part, with final assembly and firmware flashing either ongoing or about to go online.Before their November 2020 launches the PS5 entered manufacturing in June 2020 and the Series X and Series S entered manufacturing in late July 2020. They would need to start manufacturing sometime around December - February, possibly sooner if they are going for a March or April launch
I think the reality is that chip makers have to start entering purpose-built computing instead of general-purpose computing to make up the lack of efficiency gains from die shrinks alone. And Nvidia seems to understand this, I have no doubt that its specialty cores in its GPU designs are there precisely to address this problem for at least a few industries, and we’re likely to see more of that moving forward.Nvidia CEO Jensen Huang: ‘The semiconductor industry is near the limit’
Jensen Huang and his wife, Lori Huang, are donating $50 million to Oregon State University to help fund the development of a new innovation complex that will include a Nvidia supercomputer.www.protocol.com
Many of them have. The entire mobile industry functions this way. The recent updates from More Moore that @Dakhil posted are pretty insightful in how little additional specialization and consolidation has left to give usI think the reality is that chip makers have to start entering purpose-built computing instead of general-purpose computing to make up the lack of efficiency gains from die shrinks alone.
You’re absolutely right, especially on exotic materials. We’re likely a hundred years away from a generational leap on power transmission, unfortunately. That’s a situation that will get worse for a long time before it gets better.But materials (read: advancements like graphene) need to come into play sooner rather than later. And in terms of ramp-up of power consumption, resolving the electricity loss in generation and transmission (which will require investment in infrastructure) could cushion the blow he describes, as well.
Yeah I'm pretty sure I learned about graphene and its amazing potential in high school lolGraphene is kinda becoming the fusion of the materials world, it's always 10-15 years away from being viable. I was doing research work on graphene like 13 years ago and was told how it would change the world in like 5 years and we're no closer today towards achieving viable mass production.
So Graphine is the new silicon?Graphene is kinda becoming the fusion of the materials world, it's always 10-15 years away from being viable. I was doing research work on graphene like 13 years ago and was told how it would change the world in like 5 years and we're no closer today towards achieving viable mass production.
I think it's moreso the benefits of switching to graphene don't outweigh the cost of switching atm in the eyes of companiesYeah I'm pretty sure I learned about graphene and its amazing potential in high school lol
I think a newer and better node (ahem, TSMC) does allow for a significantly higher upper bound in handheld, which would also affect docked mode. Nintendo is likely targeting a specific amount of watts for both modes. If you can get a 1.5 Tflops handheldNo.
The clocks on 8nm won't be all that different from clocks on 5nm, there is a lower bound they can't really pass and an upper bound that comes up pretty quickly regardless of node. We already more or less have the full specs, it gives us a pretty narrow range of performance, this thing will not be able to be all that different just based on clocks. If they want to call it a "pro" or a "2" or a ganache is totally up to them, the process node is like, the least likely determining factor in that decision.
As I said yesterday the worry over node is getting a bit silly.
Witcher 3 is 30fps on last gen consoles and switch.It's kind of blowing up right now that Gotham Knights will be limited to 30fps on PS5 and Series S/X. Since their statement says it's more to do with having a big multiplayer open world rather than something they can work around by lowering resolution, sounds like a CPU limitation. Bringing this up as it seems like a rare example of a current game where a Drake version might not be viable.
It's already in commercialized products (notably ultracapacitors and concrete additives to increase strength and lower volume to lower CO2 in construction), so viability has been proven and makes the fusion comparison rather insulting, since from discovery of monolayer graphene to commercialization has taken... at longest 12 years compared to fusion's never?Graphene is kinda becoming the fusion of the materials world, it's always 10-15 years away from being viable. I was doing research work on graphene like 13 years ago and was told how it would change the world in like 5 years and we're no closer today towards achieving viable mass production.
One way to get a more stable, less grainy image is to aggressively filter the pixel samples, but that doesn’t mean the resulting image is a more accurate representation of the ground truth. This article doesn’t pass the smell test for me because it makes no attempt to consider or quantify this.I wonder if Nvidia's going to make improvements to DLSS in the denoising department?
Tellusim Technology Inc. found that DLSS 2.4 performs worse than FSR 2.1 in Tellusim Engine in the denoising department.
Monolayer graphene was discovered in the 60s, if not earlier. They managed to create a reproducible method for fabrication only recently (2004 I think) which involved taking graphite and scotch tape and repeatedly peeling the tape on and off of the graphite.It's already in commercialized products (notably ultracapacitors and concrete additives to increase strength and lower volume to lower CO2 in construction), so viability has been proven and makes the fusion comparison rather insulting, since from discovery of monolayer graphene to commercialization has taken... at longest 12 years compared to fusion's never?
So production of graphene for semiconductor applications is still not as close as necessary and more work needs to be done in mass production, but considering the turnaround on monolayer graphene's commercialization in other applications, yeah, we're going to figure out the means to get there a hell of a lot sooner than it'll take to crack the fusion nut.
and fusion has some, um, limited use cases tooMonolayer graphene was discovered in the 60s, if not earlier. They managed to create a reproducible method for fabrication only recently (2004 I think) which involved taking graphite and scotch tape and repeatedly peeling the tape on and off of the graphite.
Yes, it's usable and producible in very minute quantities and can be used now in very expensive and part limited applications but we are no closer to figuring out an economic method of mass producing graphene. I think right now the most viable method of production is epitaxial CVD on a silicon carbide wafer, and when you're talking about applications that need square feet of graphene rather than square cms it is an extremely inefficient production method.
there's still a bunch of Ampere gpus that will be in production for a bit in addition to Orin products. while having no future, it'll linger on life support for a bit. hard to say if Samsung will keep enough capacity for the addition of Drake
Who has been saying it'll release 1 day before TotK? Most people thinking it'll release near the game like myself expect it to release the same day, not one day before. They launch all their hardware at the same time as a new game.Sorry if I am late to the prty but what makes people think it will release 1 day before Tears of the Kingdom. If it was going to release in may I think people should expect the first week of may but no later than that.
Is October usually this slow for game news?
I have seen no one here saying that, specifically. Some think some time before Zelda, others think after Zelda, and many think on the same daySorry if I am late to the prty but what makes people think it will release 1 day before Tears of the Kingdom. If it was going to release in may I think people should expect the first week of may but no later than that.
In everyone who has kept up with this' opinion, how far are we from an initial official announcement/acknowledgement?
It's kind of blowing up right now that Gotham Knights will be limited to 30fps on PS5 and Series S/X. Since their statement says it's more to do with having a big multiplayer open world rather than something they can work around by lowering resolution, sounds like a CPU limitation. Bringing this up as it seems like a rare example of a current game where a Drake version might not be viable.
Announcement Oct 31st, release Nov 18th.
This would only help Nov/Dec sales, not hurt.
Since this doesn’t have the burden of being a “new console successor”, trying to maximize sales out of the gate isn’t an issue. It will sell out along side the other models that continue to sell as if the lifespan was still at its midpoint, not end.
It's like you describing the OLED model which was not a successor at all and still had about three months between announcement and release.Announcement Oct 31st, release Nov 18th.
This would only help Nov/Dec sales, not hurt.
Since this doesn’t have the burden of being a “new console successor”, trying to maximize sales out of the gate isn’t an issue. It will sell out along side the other models that continue to sell as if the lifespan was still at its midpoint, not end.
Personally I don't see the need for all that much marketing myself, none of the arguments for it make all that much sense.It's like you describing the OLED model which was not a successor at all and still had about three months between announcement and release.
Drake is a major upgrade, whatever Nintendo decides to do in their marketing so I don't really see less than a 2 months timeframe happening.
Announcement Oct 31st, release Nov 18th.
This would only help Nov/Dec sales, not hurt.
Since this doesn’t have the burden of being a “new console successor”, trying to maximize sales out of the gate isn’t an issue. It will sell out along side the other models that continue to sell as if the lifespan was still at its midpoint, not end.
So, adding check with name "nvidia,tegra239" as it is present in tegra239-sim-vdk.dts.
The NVIDIA vGPU software Management SDK enables third party applications to monitor and control NVIDIA physical GPUs and virtual GPUs that are running on virtualization hosts. The NVIDIA vGPU software Management SDK supports control and monitoring of GPUs from both the hypervisor host system and from within guest VMs.
NVIDIA vGPU software enables multiple virtual machines (VMs) to have simultaneous, direct access to a single physical GPU, using the same NVIDIA graphics drivers that are deployed on non-virtualized operating systems. For an introduction to NVIDIA vGPU software, see Virtual GPU Software User Guide.
I think the references to VDK for both Orin and Drake likely concern the simulator environment in general, which is used within Nvidia for hardware and driver development, rather than being intended for customers. If Nintendo ever used the simulator, it would have been for OS development, since it's not useful for game developers. Game development can take place before the silicon exists, but it doesn't need a VDK, since NVN and the NintendoSDK can run on Windows, providing developers close enough behavior to the final hardware that they can start porting engines or games.I’m only going to post a theory, it’s entirely speculative but can put a plausible idea on this that can perhaps explain a few things and give a potential cohesive reasoning for it:
A few weeks ago, @oldpuck posted a few GitHub links, one of these links was of interest:
As you may or may not know, VDK stands for “Virtual Developer Kit”
Nvidia actually offers virtual SDKs for devs to use their cloud servers in software development:
NVIDIA Virtual GPU software management SDK
developer.nvidia.com
Management SDK User Guide :: NVIDIA Virtual GPU Software Documentation
Documentation for C application programmers that explains how to use the NVIDIA virtual GPU software management SDK to integrate NVIDIA virtual GPU management with third-party applications.docs.nvidia.com
To link a few, here’s an interesting remark:
It isn’t far fetched to believe that developers can use nVidia servers to remotely develop games for a device…. and stay with me here…. Doesn’t physically exist.
If anyone remembers, Nintendo made a comment at the time, and so did Zynga, denying that A) the former sent out developer kits for a new piece of hardware and B) the latter actually received a developer kit for a new piece of hardware.
Now I know that at the time, this was just for that moment, it’s fine for them to lie because they don’t gain anything from telling the truth only harming the company, but let’s put the hypothetical scenario that they were actually just telling the truth.
The reason for why they will be telling the truth is that Nintendo wouldn’t be the one distributing actual developer kits, that would be nVidia doing it, and it will be a software that developers can access to by a portal. Likewise Zynga, even if there was an investigation done on them, wouldn’t have had a developer kit for an unannounced hardware, because it doesn’t actually physically exist at the studio.
Now, what is a supporting evidence I have of this, how would the developers even realize that they are developing precisely for what they are targeting? Nvidia has a super computer that can actually simulate a certain performance target that they are aiming to achieve, and it uses AI. They also have a super computer that can speed up the process of developing a chip, which greatly reduce the amount of time required to develop a whole new architecture/series of chips.
Nvidia R&D Chief on How AI is Improving Chip Design
Getting a glimpse into Nvidia’s R&D has become a regular feature of the spring GTC conference with Bill Dally, chief scientist and senior vice president of research, providing an overview of Nvidia’s R&D organization and a few details on current priorities. This year, Dally focused mostly on AI...www.hpcwire.com
Nvidia wants to use GPUs and AI to accelerate and improve future chip design
During a talk at this year's GPU Technology Conference, Nvidia's chief scientist and senior vice president of research, Bill Dally, talked a great deal about using GPUs...www.techspot.com
It is not impossible that developers are developing on virtual hardware, a.k.a. not physical hardware, since 2020. Henceforth, this will explain why there has been zero leaks about what the hardware looks like. Granted it was going to look like a switch most likely, however Nintendo would have simply discussed the performance target that they desire over the switch and Nvidia is simply tasked with delivering set performance target because Nintendo was willing to pay for that.
As a result, the project started in late 2019/2020, the first report of developer is being task to have games “4K ready“ was in the second half of 2020. Despite all this time, there has been nothing about a physical form of the device and I know I’ve seen people hear comments on it. in essence, it would be more difficult for information to leak this time because it is not a traditional piece of hardware cycle.
It is probably not like the PlayStation5 or the Xbox series which had a physical piece of hardware and developers worked with those physical pieces of hardware that got changed numerous times over the course of its development. In this case, there is no need to update it per se when the simulated hardware is supposed to act the same for each developer that works with it. It’s not real hardware.
This is simply a theory, I’m missing a few bits of information but this should get the gear rolling on a speculative perspective of this.
Yeah, I was gonna say isn't that what SDKs have always been for? You'll still need a physical devkit for the final, final development optimization which would be true for virtual devkits too I would imagine, at least for testing, but probably 90-95% of it can be done on development PCs.I think the references to VDK for both Orin and Drake likely concern the simulator environment in general, which is used within Nvidia for hardware and driver development, rather than being intended for customers. If Nintendo ever used the simulator, it would have been for OS development, since it's not useful for game developers. Game development can take place before the silicon exists, but it doesn't need a VDK, since NVN and the NintendoSDK can run on Windows, providing developers close enough behavior to the final hardware that they can start porting engines or games.
Nobody knows anything about the GPU, going by how vague Qualcomm's specs for the Snapdragon G3x Gen 1 are.Not really related but interesting from a cost perspective is that Razer recently announced the edge, a $400 gaming tablet with detachable controllers, rocking an octocore CPU and 1024 core GPU on TSMC 4nm. Comes with 128gb storage and 8gb of ram.
Razer Edge handheld Android console announced
It's equipped with Qualcomm's Snapdragon G3x Gen 1 chipset and comes in a 5G version for streaming PC and console games on the go.m.gsmarena.com
There's some stuff Nintendo wouldn't use that is on this device and well they need to include a Dock but cost wise it's a good comparison, did I mention the SoC is on TSMC 4nm for $400?
I daresay the volumes Nintendo would be looking at would confer them a discount compared to razer too. Costs aren't as high as some may have guessed it seems.