Rather than try to coax a 286 to life, you should look into OpenPOWER based systems like Raptor Computing Systems POWER9 machines (https://www.raptorcs.com/). They are just about to release their microATX form factor boards that are still expensive but not too crazy. These are very open machines, performance competitive with Intel/AMD based systems and can run a number of popular Linux distros. I'm looking forward to ditching Intel once my Blackbird motherboard arrives.
50 people doesn't constitute an "outcry" at a company of 100,000+. 50 people barely constitutes and outcry at a company of 1000. If you don't want to work on a project that's going to be used by the military, don't work on a project that's going to be used by the military*.
* Alternatively, fill your bosses house with a giant tinfoil pan of popcorn.
I don't know if there is a general use case but there are definitely niche use cases. The niche use cases where I've seen it useful are, amusingly, the same as using WINE on Linux. A lot of the data scientist where I work use WSL because the products we engineer don't have a Windows API and will never have a Windows API. So, we tell them to install WSL, then point them at a README that shows them how to apt-get some stuff, git clone some stuff and off they go doing... whatever data scientist do when confronted with a bash prompt... Probably just type "python" or something.
There is also a lot of talk about building a minimum viable product, with the implication that what Waymo is doing isn't it. But I think it's absolutely one (of many possible MVPs): operating in a limited geofenced area with pre-scanned environments that are very simple in the first place
Usually when people (mostly Silicon Valley people) use the term Minimum Viable Product, what they mean is, "We need to toss some half baked shit out the door as quickly as possible so we can get more funding". You can't do that when it comes to self driving cars because your half baked shit is going to literally kill people. The reason self driving cars seem "real close now" and then a few years later are still "real close now" is because the "Fuck it, let's ship this minimum viable product" is meeting the harsh realities of the automotive industry. Waymo can't just slap a "beta" sticker on the side of car, cross their fingers and ship it.
The general gist of this is, "Dang. Stallman was right". I wonder how much more miserable technology would be making our lives without the precedent of things like the GPL. I applaud the man for having the foresight to see the dark days that were coming and trying to hold them back with something that benefits society.
That already happened and probably still happens in data centers. In the late 90's (early 2000s?), Sun Microsystems would sell you an E10k class machine (64 physical CPUs) for cheaper than a fully populated E10k and disable the CPUs you didn't pay for. When you needed more power, you'd call them up, send them a boatload of money and poof... more CPUs started working. I image this kind of thing still happens in datacenters.
The way you've described the mouse synchronization is very interesting and probably true for many workflows but, part of that is the fundamentally broken window manager model of "click to focus". With "click to focus" the thing you are focused on and the mouse are disconnected and so you form habits around that disconnection (like parking the mouse after you've clicked the window). The more traditional UNIX-y model of "focus follows mouse" gives an implicit location of the mouse pointer: It's in the window you are looking at.
In many situations, this makes the mouse more analogous to your keyboard description. For example, if I'm typing in one window and need to type in the window to left of that window, my brain already knows where the mouse is (even if the cursor is hidden) and where it needs to go. It's in the window I'm looking at and it needs to go left so, I just move the mouse to the left until the window I want to type in is highlighted. No synchronization, no searching and you don't even need to see the cursor to do this.
You'll see a lot of old school UNIX guys tiling a bunch of terminals on a screen and just shoving the mouse in the direction of the window they want to have focus relative the window that currently has focus. It's usually faster than alt-tabbing as long as you are able to get your fingers back on the home row keys quickly.
It's strange that he doesn't just carry a pre-paid phone with the battery taken out. I don't carry a cell phone either but, I keep a pre-paid one in my car (or sometimes laptop bag if I'm traveling) with the battery taken out. No tracking, no listening and I basically get a portable pay phone that doesn't need quarters to operate.
A good lidar system doesn't provide depth perception, it provides a centimeter(-ish blah, blah, Guassian) accurate 3D model of the world. It's actually pretty easy to take the point cloud from a lidar and, as long as it's accurately timestamped (PTP works well), synchronize it with GPS and an IMU and get a very accurate 3D model of your surroundings. Some flavors of lidar also provide doppler on every point in the point cloud. You aren't comparing frames against either other to determine if something is moving towards or away from you, every point in the point cloud will tell you that. To the point where you can see a pedestrians legs shift before they decide to cross the street.
You are 100% correct but, you are describing frequency modulation. The vast majority of lidar companies (on the order of 99%) are producing amplitude modulation systems that shoot out strong amplitude signals and cross their fingers hoping they can see and distinguish it when it comes back. These systems often use "avalanche detectors" to help their probability of return detection (look it up, it's insane). Driving directly towards a sunset can literally cause these systems to emit their magic smoke. Accidentally hit the wrong angle on a retro-reflector? Smoke. Having another system pointed in their direction means "something made a peak" and there is no way distinguish your light from another sensors light.
These problems are solved in the radar space but, the solutions are taking a long time to trickle down into lidar space. Mostly because it's very easy and very cheap to produce an AM lidar.
Your "static" scenario is exactly what will happen with AM pulsed lidar. It's (one of many) dirty secrets of the lidar industry: AM lidar doesn't scale. There are alternative ways to do lidar (FMCW for one) that can scale much, much better but, these kinds of systems are still a bit expensive and so won't see widespread adoption for a few years.
It's not professional wrestling. You're starting to see a generation of journalists who have had access to internet porn for most of their lives. In the near future, it will be possible to determine a journalists pornographic predilections just by feeding all their headlines into an algorithm.
Unless of course Tesla has developed a telepathic module for their cars. In which case I take back what I said.
I haven't read the patent but telepathy probably isn't necessary. Humans leave a lot of subconscious clues about their intents and, if you have the right sensors, in many cases, you can detect what they are about to do before they have consciously decided to do it. My favorite example of this is FMCW Lidar and pedestrians at a crosswalk. You can literally see subtle doppler shifts in how they are distributing their weight well before they've taken a step and probably before they've even consciously decided that it's time to take a step.
I need to read the patent but, it wouldn't surprise me if seat and steering wheels sensors could provide enough information.
Never attribute to malice that which is adequately explained by a manager with a Gantt Chart. You could probably track down the designer of the board and he would dejectedly tell you, "Yeah, it's a shit design and we had a respin ready but, it didn't fit in the schedule". Or you could track down the embedded software guy and he'd tell you, "We had this elegant upgrade path planned out but no one could figure out how it fit into the Gantt Chart so we dropped it".
The engineers want to do The Right Thing but, when The Right Thing is pitched to management, it's usually just crickets. I genuinely don't think it's malice, I think it's Gantt Charts.
I keep hearing this mantra about "OMG no NoScript!". Apparently people don't realize that the script blocker in uBlock Origin is *far* superior to NoScript. It was updated for the new Firefox months ago so, it's had plenty of time to brew. You can thank me later: https://github.com/gorhill/uBl...
I'd much rather dump heat into my house with 3D printers than bitcoin miners. If your electricity is cheap enough, sure, you are effectively heating your house for free but, you've still paid for the miners and they will be obsolete in 6 months (Actually, they are usually obsolete by the time China ships them to you). After a single winter you are going to have a pile of very hot, very expensive ASICs that no longer mine fast enough to pay for their electricity costs.
A 3D printer is presumably creating something you want so the heat is a beneficial winter-time byproduct. I have two in my home office. It's currently snowing outside and I have my thermostat set at 62F. The rest of the house is a bit cold but, they keep a 12ft x 12ft office nice and cozy and every few hours something interesting pops out of them. I didn't buy them to heat my office but, I suspect that my electric bill will be lower this winter because I'm basically just heating the room I'm usually sitting in.
I've heard that many embedded software vendors respect gray hairs
This. Embedded is where it's at for older programmers. I'll list the awesomeness I've experienced as someone who has switched to embedded:
- You get to write code on a tiny machine that is still 100x as powerful as the 8086 you learned on but nobody else wants to touch because... OMG... C - As soon as someone says Ruby on Rails, you are officially authorized to leave the meeting - Agile? Fuck you. - You get to build systems where understanding how they work is your damn job. You aren't working on layers upon layers of magical APIs that you couldn't debug even if you wanted to. It's your code, libc and the kernel. - You don't have to ask, "What IDE do you guys use?". They use vi and make. I don't mean vim and cmake. I mean vi and make. Which means you get to giggle when someone says, "Why won't this editor backspace?!" - Slow is a bug. If you love doing performance analysis and squeezing every drop of performance out of a system, embedded will bring tears of joy to your eyes.
Frankly, it's glorious. I'd never even consider a non-embedded job at this point.
I genuinely admire what you're doing and really wish that a vocational "Software Engineering Drudgery" degree would be a thing but, I just don't see how it's possible. The drudgery requires just as much logic skills as the product. I would almost say that the best software teams are the ones who make their smartest guys build the infrastructure (including Makefiles, networking, etc). Everything else floats on that raft. I sure as hell don't want my raft built by a 16 year old.
I say this as a guy who dropped out of college as a junior at the age of 18. 20 years later, my lack of degree has had *zero* effect on my ability to get a job but, I'm acutely aware of how bad I was at doing... well... anything... at the age of 18.
I'd love to have a vocational software assistant but, software is complex enough that I barely trust experienced co-workers to write it, let alone a 16 year old kid.
I agree with you in spirit but not in practice. I work for a small company and we had a top-notch, experienced EE doing design *and* soldering work for a while. Once we hired a technician to do the soldering work, the EE's productivity increased dramatically. I don't think the same can be said for many/most software jobs. I can't hire cheap labor to do my dirty work because there is no part of the process that can be pushed onto people with underwhelming qualifications. There is no equivalent of "the guy who solders my boards".
We hire interns whenever we can but, I've long thought that maybe I spend more time helping the intern than I would if I'd just written it myself. And, when the intern leaves, it's actually pretty common to just rewrite what they did. So, it's almost certain that they are, at best, a cheap prototype vehicle.
The tedious work in computer science is actually what a technician is *least* qualified to do. You want a 16 year old kid to create your Makefiles? Fuck that. You want a 16 year old kid to grok your network? Fuck that. Those are hard things to do and there is a reason that people make a lot of money doing them: If you are good at doing that level of tedious stuff, you are worth a lot of money. It's actually very hard to do.
So, no, we aren't going to see a huge surge of technicians in CS. We've already seen it. It's called offshoring. And the quality of software (and support) has dramatically decreased because of it. Cheap labor and quality software are not compatible ideas. A product that involves creative thought does not lend itself to technicians. And that's what offshoring gives you: Technicians.
According to Google (who's a big LIDAR proponent), it's still $7,5k per unit. It still messes up your aerodynamics and looks dorky. It still can't see in adverse weather conditions, meaning you have to have developed an optical / radar based world-modeling system anyway. And you have to have image processing regardless to read signs, road lines, identify objects, see brake lights, and so forth.
There's real hope for further improvements in LIDAR and its variants in the future, however. We'll see where it goes.
Good lidar systems see much, much better than camera based systems in adverse weather. I work on FMCW lidar systems and I recall driving to work one day where the fog was so bad that I couldn't even find the road to my office. Once I got work, I turned on the lidar system I was working on and it imaged a building 100 meters away without issue. Road lines are trivial to identify in a lidar system since they have much different reflectivity than the road surface. Objects are also easier to identify because you aren't trying to pull three dimensional information out of two dimensional images. On an FMCW lidar system, you also get doppler information for free. You don't have to try and decide if an object is moving towards or away from you by comparing subsequent images. Every single point in the point cloud includes a meters/second doppler value.
I have to assume your familiarity is with those awful spinning Velodyne systems. They are utter garbage. No self driving car company that I've interacted with is even vaguely entertaining the idea of using them in a production car. They don't even really like using them in their mule cars but, until very recently, they were the only real option available.
The real problem with lidar is that people aren't good at consuming lidar data yet. Once they start to get some experience with it, I have zero doubt that lidar will be the primary sensor on the car. It's the only way you can really build a model of your surroundings with high accuracy, high refresh rate and high tolerance to ambient conditions. So, I actually agree with the GM guy here: Tesla is full of shit. They aren't going to make a level 5 autonomous car with cameras and radar.
Crappy rural broadband that services the 90% of the continental U.S. where the least-crowded 50% of the population lives just doesn't cut it for streaming services.
This is exactly why I used it for years. You can't stream anything of quality over a rural 3Mbit ADSL connection and you can't afford to stream anything over a higher bandwidth satellite connection. Thankfully, the movie studios have done me a huge favor by only releasing like 1-2 good movies a year so, there is no need for any of that nonsense anymore.
Slashdot Mirror
Rather than try to coax a 286 to life, you should look into OpenPOWER based systems like Raptor Computing Systems POWER9 machines (https://www.raptorcs.com/). They are just about to release their microATX form factor boards that are still expensive but not too crazy. These are very open machines, performance competitive with Intel/AMD based systems and can run a number of popular Linux distros. I'm looking forward to ditching Intel once my Blackbird motherboard arrives.
50 people doesn't constitute an "outcry" at a company of 100,000+. 50 people barely constitutes and outcry at a company of 1000. If you don't want to work on a project that's going to be used by the military, don't work on a project that's going to be used by the military*.
* Alternatively, fill your bosses house with a giant tinfoil pan of popcorn.
I don't know if there is a general use case but there are definitely niche use cases. The niche use cases where I've seen it useful are, amusingly, the same as using WINE on Linux. A lot of the data scientist where I work use WSL because the products we engineer don't have a Windows API and will never have a Windows API. So, we tell them to install WSL, then point them at a README that shows them how to apt-get some stuff, git clone some stuff and off they go doing... whatever data scientist do when confronted with a bash prompt... Probably just type "python" or something.
There is also a lot of talk about building a minimum viable product, with the implication that what Waymo is doing isn't it. But I think it's absolutely one (of many possible MVPs): operating in a limited geofenced area with pre-scanned environments that are very simple in the first place
Usually when people (mostly Silicon Valley people) use the term Minimum Viable Product, what they mean is, "We need to toss some half baked shit out the door as quickly as possible so we can get more funding". You can't do that when it comes to self driving cars because your half baked shit is going to literally kill people. The reason self driving cars seem "real close now" and then a few years later are still "real close now" is because the "Fuck it, let's ship this minimum viable product" is meeting the harsh realities of the automotive industry. Waymo can't just slap a "beta" sticker on the side of car, cross their fingers and ship it.
The general gist of this is, "Dang. Stallman was right". I wonder how much more miserable technology would be making our lives without the precedent of things like the GPL. I applaud the man for having the foresight to see the dark days that were coming and trying to hold them back with something that benefits society.
That already happened and probably still happens in data centers. In the late 90's (early 2000s?), Sun Microsystems would sell you an E10k class machine (64 physical CPUs) for cheaper than a fully populated E10k and disable the CPUs you didn't pay for. When you needed more power, you'd call them up, send them a boatload of money and poof... more CPUs started working. I image this kind of thing still happens in datacenters.
The way you've described the mouse synchronization is very interesting and probably true for many workflows but, part of that is the fundamentally broken window manager model of "click to focus". With "click to focus" the thing you are focused on and the mouse are disconnected and so you form habits around that disconnection (like parking the mouse after you've clicked the window). The more traditional UNIX-y model of "focus follows mouse" gives an implicit location of the mouse pointer: It's in the window you are looking at.
In many situations, this makes the mouse more analogous to your keyboard description. For example, if I'm typing in one window and need to type in the window to left of that window, my brain already knows where the mouse is (even if the cursor is hidden) and where it needs to go. It's in the window I'm looking at and it needs to go left so, I just move the mouse to the left until the window I want to type in is highlighted. No synchronization, no searching and you don't even need to see the cursor to do this.
You'll see a lot of old school UNIX guys tiling a bunch of terminals on a screen and just shoving the mouse in the direction of the window they want to have focus relative the window that currently has focus. It's usually faster than alt-tabbing as long as you are able to get your fingers back on the home row keys quickly.
It's strange that he doesn't just carry a pre-paid phone with the battery taken out. I don't carry a cell phone either but, I keep a pre-paid one in my car (or sometimes laptop bag if I'm traveling) with the battery taken out. No tracking, no listening and I basically get a portable pay phone that doesn't need quarters to operate.
A good lidar system doesn't provide depth perception, it provides a centimeter(-ish blah, blah, Guassian) accurate 3D model of the world. It's actually pretty easy to take the point cloud from a lidar and, as long as it's accurately timestamped (PTP works well), synchronize it with GPS and an IMU and get a very accurate 3D model of your surroundings. Some flavors of lidar also provide doppler on every point in the point cloud. You aren't comparing frames against either other to determine if something is moving towards or away from you, every point in the point cloud will tell you that. To the point where you can see a pedestrians legs shift before they decide to cross the street.
You are 100% correct but, you are describing frequency modulation. The vast majority of lidar companies (on the order of 99%) are producing amplitude modulation systems that shoot out strong amplitude signals and cross their fingers hoping they can see and distinguish it when it comes back. These systems often use "avalanche detectors" to help their probability of return detection (look it up, it's insane). Driving directly towards a sunset can literally cause these systems to emit their magic smoke. Accidentally hit the wrong angle on a retro-reflector? Smoke. Having another system pointed in their direction means "something made a peak" and there is no way distinguish your light from another sensors light.
These problems are solved in the radar space but, the solutions are taking a long time to trickle down into lidar space. Mostly because it's very easy and very cheap to produce an AM lidar.
Your "static" scenario is exactly what will happen with AM pulsed lidar. It's (one of many) dirty secrets of the lidar industry: AM lidar doesn't scale. There are alternative ways to do lidar (FMCW for one) that can scale much, much better but, these kinds of systems are still a bit expensive and so won't see widespread adoption for a few years.
More likely these guys dialed it up to 11 for CES demos and got caught.
It's not professional wrestling. You're starting to see a generation of journalists who have had access to internet porn for most of their lives. In the near future, it will be possible to determine a journalists pornographic predilections just by feeding all their headlines into an algorithm.
These are not the numbers you are looking for...
also for Google internally where it is safe to identify you gender as Dragon
I always brush off the Google recruiters when they call but, I didn't realize this was an option... Tell me more...
Unless of course Tesla has developed a telepathic module for their cars. In which case I take back what I said.
I haven't read the patent but telepathy probably isn't necessary. Humans leave a lot of subconscious clues about their intents and, if you have the right sensors, in many cases, you can detect what they are about to do before they have consciously decided to do it. My favorite example of this is FMCW Lidar and pedestrians at a crosswalk. You can literally see subtle doppler shifts in how they are distributing their weight well before they've taken a step and probably before they've even consciously decided that it's time to take a step.
I need to read the patent but, it wouldn't surprise me if seat and steering wheels sensors could provide enough information.
Never attribute to malice that which is adequately explained by a manager with a Gantt Chart. You could probably track down the designer of the board and he would dejectedly tell you, "Yeah, it's a shit design and we had a respin ready but, it didn't fit in the schedule". Or you could track down the embedded software guy and he'd tell you, "We had this elegant upgrade path planned out but no one could figure out how it fit into the Gantt Chart so we dropped it".
The engineers want to do The Right Thing but, when The Right Thing is pitched to management, it's usually just crickets. I genuinely don't think it's malice, I think it's Gantt Charts.
I keep hearing this mantra about "OMG no NoScript!". Apparently people don't realize that the script blocker in uBlock Origin is *far* superior to NoScript. It was updated for the new Firefox months ago so, it's had plenty of time to brew. You can thank me later: https://github.com/gorhill/uBl...
I'd much rather dump heat into my house with 3D printers than bitcoin miners. If your electricity is cheap enough, sure, you are effectively heating your house for free but, you've still paid for the miners and they will be obsolete in 6 months (Actually, they are usually obsolete by the time China ships them to you). After a single winter you are going to have a pile of very hot, very expensive ASICs that no longer mine fast enough to pay for their electricity costs.
A 3D printer is presumably creating something you want so the heat is a beneficial winter-time byproduct. I have two in my home office. It's currently snowing outside and I have my thermostat set at 62F. The rest of the house is a bit cold but, they keep a 12ft x 12ft office nice and cozy and every few hours something interesting pops out of them. I didn't buy them to heat my office but, I suspect that my electric bill will be lower this winter because I'm basically just heating the room I'm usually sitting in.
I've heard that many embedded software vendors respect gray hairs
This. Embedded is where it's at for older programmers. I'll list the awesomeness I've experienced as someone who has switched to embedded:
- You get to write code on a tiny machine that is still 100x as powerful as the 8086 you learned on but nobody else wants to touch because... OMG... C
- As soon as someone says Ruby on Rails, you are officially authorized to leave the meeting
- Agile? Fuck you.
- You get to build systems where understanding how they work is your damn job. You aren't working on layers upon layers of magical APIs that you couldn't debug even if you wanted to. It's your code, libc and the kernel.
- You don't have to ask, "What IDE do you guys use?". They use vi and make. I don't mean vim and cmake. I mean vi and make. Which means you get to giggle when someone says, "Why won't this editor backspace?!"
- Slow is a bug. If you love doing performance analysis and squeezing every drop of performance out of a system, embedded will bring tears of joy to your eyes.
Frankly, it's glorious. I'd never even consider a non-embedded job at this point.
You can't fight in here! This is the war room!
-- From Dr. Strangelove
I genuinely admire what you're doing and really wish that a vocational "Software Engineering Drudgery" degree would be a thing but, I just don't see how it's possible. The drudgery requires just as much logic skills as the product. I would almost say that the best software teams are the ones who make their smartest guys build the infrastructure (including Makefiles, networking, etc). Everything else floats on that raft. I sure as hell don't want my raft built by a 16 year old.
I say this as a guy who dropped out of college as a junior at the age of 18. 20 years later, my lack of degree has had *zero* effect on my ability to get a job but, I'm acutely aware of how bad I was at doing... well... anything... at the age of 18.
I'd love to have a vocational software assistant but, software is complex enough that I barely trust experienced co-workers to write it, let alone a 16 year old kid.
I agree with you in spirit but not in practice. I work for a small company and we had a top-notch, experienced EE doing design *and* soldering work for a while. Once we hired a technician to do the soldering work, the EE's productivity increased dramatically. I don't think the same can be said for many/most software jobs. I can't hire cheap labor to do my dirty work because there is no part of the process that can be pushed onto people with underwhelming qualifications. There is no equivalent of "the guy who solders my boards".
We hire interns whenever we can but, I've long thought that maybe I spend more time helping the intern than I would if I'd just written it myself. And, when the intern leaves, it's actually pretty common to just rewrite what they did. So, it's almost certain that they are, at best, a cheap prototype vehicle.
The tedious work in computer science is actually what a technician is *least* qualified to do. You want a 16 year old kid to create your Makefiles? Fuck that. You want a 16 year old kid to grok your network? Fuck that. Those are hard things to do and there is a reason that people make a lot of money doing them: If you are good at doing that level of tedious stuff, you are worth a lot of money. It's actually very hard to do.
So, no, we aren't going to see a huge surge of technicians in CS. We've already seen it. It's called offshoring. And the quality of software (and support) has dramatically decreased because of it. Cheap labor and quality software are not compatible ideas. A product that involves creative thought does not lend itself to technicians. And that's what offshoring gives you: Technicians.
According to Google (who's a big LIDAR proponent), it's still $7,5k per unit. It still messes up your aerodynamics and looks dorky. It still can't see in adverse weather conditions, meaning you have to have developed an optical / radar based world-modeling system anyway. And you have to have image processing regardless to read signs, road lines, identify objects, see brake lights, and so forth.
There's real hope for further improvements in LIDAR and its variants in the future, however. We'll see where it goes.
Good lidar systems see much, much better than camera based systems in adverse weather. I work on FMCW lidar systems and I recall driving to work one day where the fog was so bad that I couldn't even find the road to my office. Once I got work, I turned on the lidar system I was working on and it imaged a building 100 meters away without issue. Road lines are trivial to identify in a lidar system since they have much different reflectivity than the road surface. Objects are also easier to identify because you aren't trying to pull three dimensional information out of two dimensional images. On an FMCW lidar system, you also get doppler information for free. You don't have to try and decide if an object is moving towards or away from you by comparing subsequent images. Every single point in the point cloud includes a meters/second doppler value.
I have to assume your familiarity is with those awful spinning Velodyne systems. They are utter garbage. No self driving car company that I've interacted with is even vaguely entertaining the idea of using them in a production car. They don't even really like using them in their mule cars but, until very recently, they were the only real option available.
The real problem with lidar is that people aren't good at consuming lidar data yet. Once they start to get some experience with it, I have zero doubt that lidar will be the primary sensor on the car. It's the only way you can really build a model of your surroundings with high accuracy, high refresh rate and high tolerance to ambient conditions. So, I actually agree with the GM guy here: Tesla is full of shit. They aren't going to make a level 5 autonomous car with cameras and radar.
Crappy rural broadband that services the 90% of the continental U.S. where the least-crowded 50% of the population lives just doesn't cut it for streaming services.
This is exactly why I used it for years. You can't stream anything of quality over a rural 3Mbit ADSL connection and you can't afford to stream anything over a higher bandwidth satellite connection. Thankfully, the movie studios have done me a huge favor by only releasing like 1-2 good movies a year so, there is no need for any of that nonsense anymore.