Samsung started shipping 3D NAND in consumer SSDs in mid-2014 with the 850 Pro. It used a 32-layer 40nm process. It was their second-gen 3D NAND, with their first-gen 24-layer product released in 2013 for enterprise SSDs.
IMFT and Toshiba are both sampling 3D NAND, 32-layer for IMFT and 48-layer for Toshiba. Neither is saying what size, but some googling indicates the consensus is somewhere between 35nm and 50nm.
You're still going to need to do wear leveling. If you can write a byte every 2us, but you only have 1000x the endurance of NAND, you've only got 3 million write cycles, which you could burn through in around 6 seconds for a single byte. And since per-byte wear leveling is nuts, you're back to using the same sort of blocks as NAND does.
In terms of quickly eroding the market for NAND, if this stuff is still significantly more expensive than NAND, I don't see how it would have much of an impact. They're saying the cost is "in between there somewhere" in terms of DRAM and NAND. For consumer products, NAND is at ballpark $0.38 per gig, and DRAM is at around $5.00 per gig. So depending on where XPoint is in that range, it could either have a huge impact or a small impact.
Then again, $0.38 is MLC or TLC, and SLC is obviously going to cost a lot more than that. So maybe this new stuff might be more competitive with SLC and supplant that, but the SLC market is probably pretty small: Intel doesn't even use it in their enterprise drives anymore.
I would imagine where this stuff will have the biggest impact is in completely new use cases (places that needed non-volatile memory and so had to use flash, but were significantly performance bottlenecked), or in places where a combination of NAND/DRAM/supercapacitors are currently used. I can see enterprise use being a thing.
The X1 uses a standard ARM Cortex A57 (specifically it's an A57/A53 big.LITTLE 4+4 config), so this says more about ARM's chip than anything nVidia did...
Now if you compared nVidia's Denver CPU, their in-house processor... The Denver is nearly twice as fast as the A57, but only comes in a dual-core config, so it's probably drawing a good deal more power. When you compare a quad-core A57 to a dual-core Denver, the A57 comes out slightly ahead in multicore benchmarks. Of course, single core performance is important too, so I'd be tempted to take a dual-core part over a quad-core if the dual-core had twice the performance per-core...
Why the X1 didn't use a variant of Denver isn't something that nVidia has said, but the assumption most make is that it wasn't ready for the die shrink to 20nm that the X1 entailed.
Why would most people need a terabyte of RAM? There is such a thing as diminishing returns, and you have to be doing pretty crazy stuff to need more than 32GB or so. Ultimately making RAM cheaper might save a few bucks, but that's not justifying the massive hype that non-volatile memory has had over the years.
In terms of 4K recording on a phone, video encoding is done in hardware, not software, and no amount of cache is going to solve that. If you're talking about highspeed, your 64GB cache would store around 11 seconds of raw 240FPS 4K video, and if you're encoding it first, then the speed of the storage isn't really the bottleneck.
I'm proposing that the drones be equipped with this to keep them out of the buffer area, not that the actual airplanes. Airplanes operating over cities are already required to operate 1,000 feet above the highest nearby obstacles, placing them far above any drones. Helicopters would be another story, but they are allowed to operate under your proposed 10 foot cap, so that's kind of already a thing.
Why would you expect it to be cheaper or denser than NAND? IMFT says it'll be more expensive than NAND, and even if the cost drops over time, so too will the cost of NAND.
The whole point of this, though, is integrating something new that works radically differently from existing aircraft. If other technological mechanisms can provide sufficient accuracy, why can't they be used?
It's not like they need to solely rely on AGPS either. Consumer IMUs have been advancing at a rapid pace (there's huge amounts of money being dumped in them due to gaming, VR, and mobile phones), and are capable of high accuracy when combined with an external reference. You can also use laser ranging, which is also very cheap these days (my robotic vacuum cleaner has a LIDAR turret on it, although the range would be less than the few hundred feet required here). If you know where you are, and you know the height above sea level at that location, and you know how far you are from the ground...
There are many tough problems to solve to make what Amazon is proposing practical, but accurately figuring out your altitude a few hundred feet from the ground is certainly not insoluble (or even particularly difficult). There are many things you can do to determine altitude at 300 feet than aren't possible at 30,000 feet.
Good modern GPS implementations (which often include information from multiple constellations and other sources like wifi and cellular towers) can provide altitude with far better than 100' of accuracy. They are not particularly expensive.
> Not quite: Just open a 500MB word document and insert a single character at the top of the file.
COW filesystems would have no problem with that scenario, especially when they have dynamic block sizes. There might still be some nasty write amplification (such as writing kilobytes of data to insert the one character), but it wouldn't be any slower than appending one byte to the end of the file.
Would that really make much of a difference? Computers already act, for the most part, like they have non-volatile memory. When you shut the lid on a laptop, it writes RAM to disk and goes to sleep. If you wake it up without having cut the power, it wakes up quickly. If you pull the power/battery, it takes a few seconds longer. In either case, it wakes up where it left off.
There is also nothing stopping developers from doing what you describe right now. Storage is fast enough that changes to most files can be saved directly to disk as they're made. When working in the cloud, this sort of "every keystroke saved" thing is already the norm.
I'm not trying to say that really fast and durable non-volatile memory wouldn't make some improvements in some places, but generally the workarounds currently in use have gotten so good that the impact would be relatively minor.
But at that point you're basically just replacing the current approach of a supercapacitor and DRAM with some of this new stuff. You might save a few bucks, but that's a relatively small difference.
We've already got non-volatile memory with extremely high endurance on the mass market (SLC NAND), so what you basically get out of this stuff is "It's like flash, but much faster."
The question becomes, what is enabled by having much faster flash memory? Sure, you might see some minor power efficiency increases in mobile devices if you don't need to keep the RAM powered, but that's not exactly world changing.
I'm not saying this isn't good, just that people are hyping it up, and I'm trying to determine what it might enable that is worthy of hype.
Ion thrusters are rockets. They still use propellant. They simply use electricity to accelerate the propellant instead of a chemical reaction. When an electric rocket runs out of propellant, it can no longer produce thrust, even if the vehicle can still supply electricity. The EM Drive does not use propellant. What has not yet been verified is if it actually produces any thrust. Nobody has yet tested it in such a matter as to conclusively demonstrate this.
Lifting its own weight is irrelevant. Existing electric propulsion thrusters couldn't come remotely close to lifting their own weight, and yet are still in active use in space.
Existing electric propulsion devices (like ion thrusters) still use propellant, they're just really efficient.
The EM drive would appear to use no propellant, meaning the limitations would only be the amount of electricity that could be produced, along with how long the EM drive could operate before it degrades.
While Cell failed as a platform, the concept itself had merit, and the concept of pairing high-performance and low-performance processors can be found in the HPC market today (like Intel's Phi or GPGPU) and in the mobile market (like ARM's big.LITTLE architecture).
A lot of the content (like Homestar Runner and Weebl's Stuff) is also available via their official YouTube channels. You lose all the interactivity, though.
After that, javascript SWF renderers become the only option. Mozilla reported that homestar runner's content mostly sort of worked in 2013, perhaps it's better now.
I meant to write what does the melting point of an element have to do with the melting point of structures formed from atoms of that element, but somehow left out the second "melting point".
What does the melting point of an element have to do with structures formed from atoms of that element? The melting point of black phosphorus is 590 C.
I may have made the mistake of comparing prices today using CAD and historically in USD, and only comparing at the ~1TB level, where price per gig seems to be a bunch higher.
The price-per-gig on the EVO model comes out to around $0.40/GB, which is where SSD prices have more or less been stalled for a few years now. So that's not so great. We really ought to be seeing some price reductions from 3D NAND.
On the other hand, $800 is roughly what I paid for my first SSD, an 80GB Intel G1. Today, for the same price, you can get 2000GB.
Granted CGN hardware isn't cheap, and it takes effort to get it up and running, but upgrading hardware to accommodate growth wasn't going to be cheap or effortless anyhow.
But that's not how it works... The current networks make money because they don't have to cover the R&D or construction. When Iridium went bankrupt, the billions of dollars of debt from building the network evaporated. The modern companies don't have to service that debt. They can charge lower prices because they don't have those costs to recover.
All of the MSS companies went bankrupt. Iridium and ICO in 1999, Orbcomm in 2000, Globalstar and Teledesic (which never got off the ground) in 2002.
Let me put things in perspective: it cost $6 billion to build the Iridium network. The current company bought that network for $35 million. In the grand scheme of things, that's basically free. This has allowed the current Iridium to build up a respectable revenue stream, and now they've got the financial resources to build out their next-gen network. But they could never have done it without getting the "free" network.
I'm skeptical, but I think that they've got a better chance of success than Iridium did. Namely, they have their own launch vehicle (no markup or middleman), they have lower launch costs even for third party launches, and they've got a reputation for building electronics for space on the cheap by re-purposing consumer electronics parts, so they've got a chance at building the satellites themselves for much cheaper.
Slashdot Mirror
Samsung started shipping 3D NAND in consumer SSDs in mid-2014 with the 850 Pro. It used a 32-layer 40nm process. It was their second-gen 3D NAND, with their first-gen 24-layer product released in 2013 for enterprise SSDs.
IMFT and Toshiba are both sampling 3D NAND, 32-layer for IMFT and 48-layer for Toshiba. Neither is saying what size, but some googling indicates the consensus is somewhere between 35nm and 50nm.
You're still going to need to do wear leveling. If you can write a byte every 2us, but you only have 1000x the endurance of NAND, you've only got 3 million write cycles, which you could burn through in around 6 seconds for a single byte. And since per-byte wear leveling is nuts, you're back to using the same sort of blocks as NAND does.
In terms of quickly eroding the market for NAND, if this stuff is still significantly more expensive than NAND, I don't see how it would have much of an impact. They're saying the cost is "in between there somewhere" in terms of DRAM and NAND. For consumer products, NAND is at ballpark $0.38 per gig, and DRAM is at around $5.00 per gig. So depending on where XPoint is in that range, it could either have a huge impact or a small impact.
Then again, $0.38 is MLC or TLC, and SLC is obviously going to cost a lot more than that. So maybe this new stuff might be more competitive with SLC and supplant that, but the SLC market is probably pretty small: Intel doesn't even use it in their enterprise drives anymore.
I would imagine where this stuff will have the biggest impact is in completely new use cases (places that needed non-volatile memory and so had to use flash, but were significantly performance bottlenecked), or in places where a combination of NAND/DRAM/supercapacitors are currently used. I can see enterprise use being a thing.
Sorry, I meant not an *nVidia* CPU.
The X1 uses a standard ARM Cortex A57 (specifically it's an A57/A53 big.LITTLE 4+4 config), so this says more about ARM's chip than anything nVidia did...
Now if you compared nVidia's Denver CPU, their in-house processor... The Denver is nearly twice as fast as the A57, but only comes in a dual-core config, so it's probably drawing a good deal more power. When you compare a quad-core A57 to a dual-core Denver, the A57 comes out slightly ahead in multicore benchmarks. Of course, single core performance is important too, so I'd be tempted to take a dual-core part over a quad-core if the dual-core had twice the performance per-core...
Why the X1 didn't use a variant of Denver isn't something that nVidia has said, but the assumption most make is that it wasn't ready for the die shrink to 20nm that the X1 entailed.
Why would most people need a terabyte of RAM? There is such a thing as diminishing returns, and you have to be doing pretty crazy stuff to need more than 32GB or so. Ultimately making RAM cheaper might save a few bucks, but that's not justifying the massive hype that non-volatile memory has had over the years.
In terms of 4K recording on a phone, video encoding is done in hardware, not software, and no amount of cache is going to solve that. If you're talking about highspeed, your 64GB cache would store around 11 seconds of raw 240FPS 4K video, and if you're encoding it first, then the speed of the storage isn't really the bottleneck.
I'm proposing that the drones be equipped with this to keep them out of the buffer area, not that the actual airplanes. Airplanes operating over cities are already required to operate 1,000 feet above the highest nearby obstacles, placing them far above any drones. Helicopters would be another story, but they are allowed to operate under your proposed 10 foot cap, so that's kind of already a thing.
Why would you expect it to be cheaper or denser than NAND? IMFT says it'll be more expensive than NAND, and even if the cost drops over time, so too will the cost of NAND.
The whole point of this, though, is integrating something new that works radically differently from existing aircraft. If other technological mechanisms can provide sufficient accuracy, why can't they be used?
It's not like they need to solely rely on AGPS either. Consumer IMUs have been advancing at a rapid pace (there's huge amounts of money being dumped in them due to gaming, VR, and mobile phones), and are capable of high accuracy when combined with an external reference. You can also use laser ranging, which is also very cheap these days (my robotic vacuum cleaner has a LIDAR turret on it, although the range would be less than the few hundred feet required here). If you know where you are, and you know the height above sea level at that location, and you know how far you are from the ground...
There are many tough problems to solve to make what Amazon is proposing practical, but accurately figuring out your altitude a few hundred feet from the ground is certainly not insoluble (or even particularly difficult). There are many things you can do to determine altitude at 300 feet than aren't possible at 30,000 feet.
Good modern GPS implementations (which often include information from multiple constellations and other sources like wifi and cellular towers) can provide altitude with far better than 100' of accuracy. They are not particularly expensive.
> Not quite: Just open a 500MB word document and insert a single character at the top of the file.
COW filesystems would have no problem with that scenario, especially when they have dynamic block sizes. There might still be some nasty write amplification (such as writing kilobytes of data to insert the one character), but it wouldn't be any slower than appending one byte to the end of the file.
Would that really make much of a difference? Computers already act, for the most part, like they have non-volatile memory. When you shut the lid on a laptop, it writes RAM to disk and goes to sleep. If you wake it up without having cut the power, it wakes up quickly. If you pull the power/battery, it takes a few seconds longer. In either case, it wakes up where it left off.
There is also nothing stopping developers from doing what you describe right now. Storage is fast enough that changes to most files can be saved directly to disk as they're made. When working in the cloud, this sort of "every keystroke saved" thing is already the norm.
I'm not trying to say that really fast and durable non-volatile memory wouldn't make some improvements in some places, but generally the workarounds currently in use have gotten so good that the impact would be relatively minor.
But at that point you're basically just replacing the current approach of a supercapacitor and DRAM with some of this new stuff. You might save a few bucks, but that's a relatively small difference.
We've already got non-volatile memory with extremely high endurance on the mass market (SLC NAND), so what you basically get out of this stuff is "It's like flash, but much faster."
The question becomes, what is enabled by having much faster flash memory? Sure, you might see some minor power efficiency increases in mobile devices if you don't need to keep the RAM powered, but that's not exactly world changing.
I'm not saying this isn't good, just that people are hyping it up, and I'm trying to determine what it might enable that is worthy of hype.
Ion thrusters are rockets. They still use propellant. They simply use electricity to accelerate the propellant instead of a chemical reaction. When an electric rocket runs out of propellant, it can no longer produce thrust, even if the vehicle can still supply electricity. The EM Drive does not use propellant. What has not yet been verified is if it actually produces any thrust. Nobody has yet tested it in such a matter as to conclusively demonstrate this.
Lifting its own weight is irrelevant. Existing electric propulsion thrusters couldn't come remotely close to lifting their own weight, and yet are still in active use in space.
Existing electric propulsion devices (like ion thrusters) still use propellant, they're just really efficient.
The EM drive would appear to use no propellant, meaning the limitations would only be the amount of electricity that could be produced, along with how long the EM drive could operate before it degrades.
While Cell failed as a platform, the concept itself had merit, and the concept of pairing high-performance and low-performance processors can be found in the HPC market today (like Intel's Phi or GPGPU) and in the mobile market (like ARM's big.LITTLE architecture).
A lot of the content (like Homestar Runner and Weebl's Stuff) is also available via their official YouTube channels. You lose all the interactivity, though.
After that, javascript SWF renderers become the only option. Mozilla reported that homestar runner's content mostly sort of worked in 2013, perhaps it's better now.
I meant to write what does the melting point of an element have to do with the melting point of structures formed from atoms of that element, but somehow left out the second "melting point".
What does the melting point of an element have to do with structures formed from atoms of that element? The melting point of black phosphorus is 590 C.
I may have made the mistake of comparing prices today using CAD and historically in USD, and only comparing at the ~1TB level, where price per gig seems to be a bunch higher.
The price-per-gig on the EVO model comes out to around $0.40/GB, which is where SSD prices have more or less been stalled for a few years now. So that's not so great. We really ought to be seeing some price reductions from 3D NAND.
On the other hand, $800 is roughly what I paid for my first SSD, an 80GB Intel G1. Today, for the same price, you can get 2000GB.
Granted CGN hardware isn't cheap, and it takes effort to get it up and running, but upgrading hardware to accommodate growth wasn't going to be cheap or effortless anyhow.
The fact that there wasn't any real impact highlights how all the hype about IPv4 doomsday were overblown.
But that's not how it works... The current networks make money because they don't have to cover the R&D or construction. When Iridium went bankrupt, the billions of dollars of debt from building the network evaporated. The modern companies don't have to service that debt. They can charge lower prices because they don't have those costs to recover.
All of the MSS companies went bankrupt. Iridium and ICO in 1999, Orbcomm in 2000, Globalstar and Teledesic (which never got off the ground) in 2002.
Let me put things in perspective: it cost $6 billion to build the Iridium network. The current company bought that network for $35 million. In the grand scheme of things, that's basically free. This has allowed the current Iridium to build up a respectable revenue stream, and now they've got the financial resources to build out their next-gen network. But they could never have done it without getting the "free" network.
I'm skeptical, but I think that they've got a better chance of success than Iridium did. Namely, they have their own launch vehicle (no markup or middleman), they have lower launch costs even for third party launches, and they've got a reputation for building electronics for space on the cheap by re-purposing consumer electronics parts, so they've got a chance at building the satellites themselves for much cheaper.