Why buy a Wii U? and be saddled with an ugly redundant home console that's kind of useless without the tablet, while the tablet is entirely useless without the console?
Wii U is like an Altair or IMSAI 8080, a fugly box that needed a tacky computer terminal attached. Computers got more successful when they filled both the computer and terminal part like the Commodore PET, Apple II and Switch do.
You had me till the libertarian talking points. Duh, that's one of the libertarian end games : destroy the state and get your ass handed over to faceless giant global corporations. Private police? Rent a cop on facebook. No public health insurance and hospitals? Use Google Health or Apple Health, with detailed profiles like the color of your last shit for the past 35 years, if you want to be able to afford health. No public schools? Get low cost Google Education. Free contracts between individuals? Here's an EULA for our software and "services". Accept or decline.
Say, you run games on a Ryzen CPU, on a 1080p 144Hz or higher monitor so small framerate difference don't go to waste. You can notice it in very specific situations like that.
With early motherboard revisions, early BIOS, first gen CPU you might be almost certain the stupid overclocked speeds won't run though. This story about 4500 MHz RAM is really pointless right now, the only way it's interesting is it signals that in the long run, you might/should have reliable and fairly affordable memory at 3600 or even 4000 speed. Let's say an AMD desktop APU is wanted in year 2018 or 2019. By then getting 3600 RAM in 2018, or 4000+ RAM in 2019 is actually more reasonable since you will get some reasonably Xbox One like 3D performance in games. It's really nerdy to take the time to think of that, sure.
Games differ a lot between them, they can also be where you need a +5% performance the most (even though you can really barely see it)
Yes "useful" photoshop-like tasks vary as well. Benchmark quality and selection varies.
Also generally, for games you might want 4 or 6 cores at 4GHz (or more Hz if possible and cheap), if you're a professional doing media or server/dev things you might like 16 cores at 2.7GHz better (say), have more memory channels and you don't give much of a shit about RAM speed.
I think you have it wrong, video encoding only cares about megaflops pretty much while games react more to memory latency, or more precisely the whole { L1 + L2 + L3 + memory }.
Of course, the über RAM is silly with as you say the extreme frequencies hardly properly working but eventually when RAM will cost say (dummy prices) $103 for DDR 2133, $103 for DDR 2400 and $107 for DDR 3200 you should take the 3200 pretty obviously.
There were huge gains in Core i3 6100 tests, the particularity being it applies to many games there. i3 are not overclockable but have a high clock in the first place. Save for one recent model which is overclockable but costs the same as an i5. The only annoying thing, with Intel, is that you need a motherboard with Z170 or Z270 chipset to clock the RAM higher. So, they make it harder to max out single thread performance on a cheap budget. i3 7100, DDR4 2400 (up from the older 2133 limit) and B250 motherboard will be a relatively cheap option to play ARMA 3 specifically and emulators, but leaves some performance on the table.
It's also 4.2 GHz "data speed" as multiple bits are sent on a single clock cycle, hence I think it's running at 1.05 GHz.
So let's say a 1ns cycle time and the timings suggest latency might be a bit under 20-40 ns although who knows how the numbers have to be added up or not.
It adds up I think, consider the distance traveled, memory controller and CPU memory hierarchy the CPU-to-RAM latency may be something like 50 to 70 ns very roughly.
Others might not like to hear this but this might get fixed for good when Firefox hits version 57 and it doesn't include XUL for the interface anymore. I'm not even blaming XUL, I think it's been great - Firefox just looked like a normal desktop application no matter the OS, I'm just saying the GUI implementation will be different. If that theory works out, you might try Firefox Aurora 57 a few months from now and see if it's fast.
I do think newer versions perform better, i.e. Firefox 51, 52 are good. Slow to open? I've not noticed as for five seconds to open something that will run for hours or weeks, I don't care. But yes it will gobbles resources, unless you seek to use lightweight web sites only! By this point computer hardware without RAM slots feel dumber and dumber. Like, bringing the max RAM limit to 32GB or 16GB (on Atom) or 64GB or possibly more could be a reason to upgrade from old hardware.
It was pretty and safer than XP, the only problem is it required the best grade of everything. Like its successor, it makes 1024x768 feel cramped while XP was kind of comfortable at 800x600. So, you kind of needed a monitor upgrade if you wanted to fit a file manager and music player side by side. Also, sticks of 1GB DDR1 and 10k rpm hard drives weren't that cheap or common.
Well that's one more reason to not come to the US. Believe it or not there are countries where peeing on the side of the road isn't wrong and doesn't make you run the risk of being handcuffed, arrested, and charged with a sex crime.
The brain does seem to utilize what is has better than computers, e.g. in computer chips memory cells store bits and communicate a binary signal. This could be all analog and exploit the voltage range better instead. (Flash memory cells now store a charge which can be at four, eight or sixteen (experimental) levels and translate that to two, three or four bits, that's why you can have a cheap 32GB or 64GB memory card)
A large part of chips is spent ferrying data around and worrying about the clock (although asynchronous CPU exist too). Brain is asynchronous and perhaps does quite some amount of ferrying signals from one place to another too, but along every step there's quite some molecular and chemical level "computing" that we understand about nothing about (I put "computing" between quotes, because comparing it to computing or data processing might be entirely wrong)
The brain is enclosed in an actually small space. It's huge next to a couple square centimeters of chip but next to a 42U cabinet or a room full of these it's fairly small. Computers have to use terribly slow buses like PCIe, Infiniband, Ethernet and even the DRAM is dog slow from the CPU's viewpoint. We might like to see the brain as (even more) terribly slow but it's kind of a 1.5 liter 3D "CPU" continuum with zillions of elements and mega-zillions of connections. And it self-modifies constantly - good luck having your Itanium, ARM or GPU rearrange itself.
It's debatable whether we will ever be able to manufacture anything like that. It also has a whole network to carry fuel and organic materials in, waste and heat out. Other things to allow it not instantly die of cancer or infection. Although, no one knows what happens if you were to completely remove a brain (and spinal cord etc. maybe) from the living body it's attached to : maybe it would go insane in a matter of seconds and die out.
Which is to say, from a hardware stand the brain probably is severely underestimated by most or many people.
Software neural networks are an extremely simplified and idealized representation that only deals with a limited high-level subset of electrical activity we could make sense of some decades ago. There's a lot of smaller scale, or short scale stuff left out as well as all the chemical activity - if neural networks worked a bit more like a brain, we would be able to try the algorithms with caffeine, heroin, cocaine, cannabinoids, opioids and other substances, to study whether and how the algorithms are working differently.
'A simulation of intelligent behavior must appear to be intelligent.'
Does it? I would say a simulation of intelligent behavior must appear to be behavior. That's like teaching dogs to speak maybe, and either AI will never be more than that or we'll never get AI. Even a self-driving car is a system tuned for the goal of "Drive from A to B without killing or maiming anyone" and will never be able to do anything else - save for additional functions like running an inference engine to play Go on the car's GPUs, or putting some Microsoft Cortana kind of software on the dashboard to enter your destination by speaking.
One simple example is a black stone surrounded by white stones. It may happen on the board before removing the stone but I doubt it counts as a position. Have two such occurrences for the position to be definitely illegal.
Turns out, I did once read about it and to sum it up the number of legal positions is about one in 81. It was only computed on Jan 2016:) https://tromp.github.io/go/leg...
The software used for these computations is available at my github repository. Running "make" should compute L(3,3) in about a second. For running an L19 job, a beefy server with 15TB of fast scratch diskspace, 8 to 16 cores, and 192GB of RAM, is recommended. Expect a few months of running time. (...)
Choosing the values might be hard? Not only there's the problems of discarding, clamping, correcting data (if you remember "hide the decline", maybe that's stuff you will have to do) but I wonder if you'll end up doing something silly. E.g., your system ends up increasing the production of men's underwear when the number of pirate attacks is low. Then next year a outlying bout of 1000 pirate attacks happens, your program falls apart but before the underwear company notices they shouldn't blindly trust what comes out of the computer, they will miss a ton of sales and their underwear competitors will mop the floor with them.
And soon, there won't be programmers anymore because computers will program themselves:) You only have to write the program's specification (a square mile of UML diagrams will do).
Can you do something incredibly plain and unremarkable? e.g., email. Like, everyone hates email as they have bloated web GUIs, all tons of crap mixed in there (administrative, personal, pointless subscriptions) Like, you could give a company, team or group their own squirrelmail, and also their own IRC accessed with the same login for instance. Clean, free of crap and everyone understands what it's about.
You heard about it here lol. Advertise to local businesses. Well, maybe there's only a trickle of money to be made and it's not really a software startup, but I wanted to make the joke. Provide the LEAST amount of innovation possible but with useful benefits to the customer. Like, Google and Wikipedia got absolutely huge by providing people what they wanted - black text on a white background.
Removing underlined Gtk3 letters was the one stupid ass thing. Can't they have a system wide setting so they can hide them on their Gnome or other 3D desktop and look good in their marketing screenshots, but leave them for users of traditionals GTK2/GTK3/Qt etc. desktops?
Other than that, GTK3 still works fine for normal looking, normal working software, usually to the point of not noticing whether your app runs GTK2 or GTK3. I like it, and dislike Qt4/Qt5 instead (for no particular reason).
On occasion I found it useful as there's some networked software where it's pointless not to run the latest version and the stub installer is good for this.
Debian and ubuntu even have it for the whole operating system! The "downloader installer stub" (net install) for debian and ubuntu can use a http proxy, incidentally. Sometimes the problem if you want the full installer is to find the download link. If not available at all, that sucks. Ubuntu is the worst at allowing you to find the downloads on their damn web page, even the vanilla default download! (full iso, main edition, 64bit x86). I think they like to change their web site as well. If you want to find the "installer stub" for Ubuntu that'll be actually a lot harder than the full isos unless you web-search for it.
So much, I guess I can recommend debian or Linux Mint because you can go to the site and find the files to download.
Now, the stupid bit about Ubuntu : I saw the Ubuntu desktop on TV (in some documentary or news report about something else), have downloaded it to rescue a few acquaintances from Windows 7/8/10 and then there's big news coming to say it's deprecated. Well I'll install it probably, since the installations shall be replaced in 2020 or 2021 and by then, who knows what will be deprecated or abandoned anyway.
Intel graphics increasingly are meant to run games : foremost, if some people upgrade to a 4790K or 6600K it might have the best graphics hardware they ever had, because it beats e.g. geforce 7600GT (~2006) or Radeon 5450 (~2009) and others. Then, they got a bit better of their goal of being barely able to run anything. Like, the two cards I mentioned above are deprecated. In fact, look at the low end market and the fact you still can buy geforce 210, Radeon 5450, 6450, R5 230. That's deprecated! (nvidia provides a "legacy" linux driver that you still can use)
R7 240 and 250 are not deprecated officially (but long in the tooth), but don't have a good linux driver yet. So given all the bullshit and museum pieces on the low end graphics market, Intel graphics looks like a good choice. It's good that it sits under the CPU's heatsink.
There is one usable low end graphics board : geforce GT 710. Although, it doesn't have displayport. Roughly similar to Intel graphics in terms of power, but with a better driver. But getting older already.
Even on chip buses are as fast as they need to be, or often too slow (typically, two packs of four CPU cores have a slowish networking between them and aren't to be used arbitrarily as an eight core CPU) You can have a state of the art interconnect between CPU and GPU and memory etc. on a high end chip, sure.
CPU and GPU might be able to share memory addresses, function like AMD's heterogeneous computing promises to do (or does already, just not used very much). Not only AMD, the HSA foundation has the usual mobile CPU/GPU/SoC suspects. Possibly Apple would aim for such abilities in developing a SoC with Apple CPU and Apple GPU, and I wonder if you were thinking of something like that. If so, there will eventually be an announcement of Apple joining the HSA foundation. But whether Apple joins in or not, this might be very interesting technically speaking.. But we'll need software that uses it. If Apple does make such an integrated CPU + GPU combo, it would need e.g. some powerful video editor for iPad Pro and iPad to show off the abilities, or some other example.
I originally wanted to point out the "modular" external GPU is available in just one product, Microsoft Surfacebook with the keyboard dock that includes battery and GPU (so, Intel CPU + GPU combo in the main unit, and low power nvidia in the keyboard dock). So I see no overarching reason why Apple can't do the same with Apple CPU + GPU in an iPad, and Apple GPU in a powered dock. You can even run some code that likes lower latency on the internal GPU, and some code ok with high latency on the external GPU. Just saying it's technically possible, not that they would bother with such a contraption - they'll probably want to sell you a Mac instead, not the equivalent of a "Sega 32x" or sidecar expansion for the iPad/iPad Pro/iPhone plus.
The Northbridge and PCIe for GPU are integrated in all desktop/laptops CPUs now (save for AM3+ FX 6000 and 8000 series that you still can buy) Of course this doesn't change much at all. PCIe bandwith and latency and joule per bit on PCIe still are the same.
What would work better is an MCM, CPU and GPU are packaged close together and use custom or semi-custom interface. It does exist, that's AMD's upcoming Opteron that includes on the same package : two Ryzen dies, one Vega 2048 SP / 32 CU die, HBM2 memory for the GPU die. I'm sure it will be awesome but it will work great at 180 watts or 250 watts, not 1.8 watt or 2.5 watts. There's also Haswell or later at lowest wattage : on a single package, a CPU (with GPU and shit integrated) and a chipset are connected, closely and low powery. The "MacBook" (thing with only one USB port) and Surface Pro can use that. But it's a bit similar to a phone using a "do everything" SoC and a radio/modem chip on the side.
All of that to support your conclusion but theoretically, a large Ipad / Ipad Pro can use a discrete GPU, though still very unlikely. Can optic communication between two chips reduce the power? That would be one way to make it more possible, although likely again some expensive thing for $1000 and up tablets or other small mobile computers.
Ultra fast external communications are power hungry in general, what's telling is 4K 60Hz and up may require real time compression even between a device's chip and the device's display panel (there's a VESA standard for this) and 10 gigabit RJ45 ethernet failed just because of this.
If it's there, and you can see it . . . If it's not there, but you can see it . . . If it's not there, and you can't see it . ..
We should ask Donald Rumsfeld about this then. Also, for the case of "If it's there, but you can't see it". Alternatively you might have something to teach Rumsfeld about!
Slashdot Mirror
Why buy a Wii U? and be saddled with an ugly redundant home console that's kind of useless without the tablet, while the tablet is entirely useless without the console?
Wii U is like an Altair or IMSAI 8080, a fugly box that needed a tacky computer terminal attached. Computers got more successful when they filled both the computer and terminal part like the Commodore PET, Apple II and Switch do.
You had me till the libertarian talking points. Duh, that's one of the libertarian end games : destroy the state and get your ass handed over to faceless giant global corporations.
Private police? Rent a cop on facebook.
No public health insurance and hospitals? Use Google Health or Apple Health, with detailed profiles like the color of your last shit for the past 35 years, if you want to be able to afford health.
No public schools? Get low cost Google Education.
Free contracts between individuals? Here's an EULA for our software and "services". Accept or decline.
Say, you run games on a Ryzen CPU, on a 1080p 144Hz or higher monitor so small framerate difference don't go to waste.
You can notice it in very specific situations like that.
With early motherboard revisions, early BIOS, first gen CPU you might be almost certain the stupid overclocked speeds won't run though.
This story about 4500 MHz RAM is really pointless right now, the only way it's interesting is it signals that in the long run, you might/should have reliable and fairly affordable memory at 3600 or even 4000 speed.
Let's say an AMD desktop APU is wanted in year 2018 or 2019. By then getting 3600 RAM in 2018, or 4000+ RAM in 2019 is actually more reasonable since you will get some reasonably Xbox One like 3D performance in games.
It's really nerdy to take the time to think of that, sure.
Games differ a lot between them, they can also be where you need a +5% performance the most (even though you can really barely see it)
Yes "useful" photoshop-like tasks vary as well. Benchmark quality and selection varies.
Also generally, for games you might want 4 or 6 cores at 4GHz (or more Hz if possible and cheap), if you're a professional doing media or server/dev things you might like 16 cores at 2.7GHz better (say), have more memory channels and you don't give much of a shit about RAM speed.
I think you have it wrong, video encoding only cares about megaflops pretty much while games react more to memory latency, or more precisely the whole { L1 + L2 + L3 + memory }.
Of course, the über RAM is silly with as you say the extreme frequencies hardly properly working but eventually when RAM will cost say (dummy prices) $103 for DDR 2133, $103 for DDR 2400 and $107 for DDR 3200 you should take the 3200 pretty obviously.
There were huge gains in Core i3 6100 tests, the particularity being it applies to many games there. i3 are not overclockable but have a high clock in the first place. Save for one recent model which is overclockable but costs the same as an i5.
The only annoying thing, with Intel, is that you need a motherboard with Z170 or Z270 chipset to clock the RAM higher. So, they make it harder to max out single thread performance on a cheap budget. i3 7100, DDR4 2400 (up from the older 2133 limit) and B250 motherboard will be a relatively cheap option to play ARMA 3 specifically and emulators, but leaves some performance on the table.
It's also 4.2 GHz "data speed" as multiple bits are sent on a single clock cycle, hence I think it's running at 1.05 GHz.
So let's say a 1ns cycle time and the timings suggest latency might be a bit under 20-40 ns although who knows how the numbers have to be added up or not.
It adds up I think, consider the distance traveled, memory controller and CPU memory hierarchy the CPU-to-RAM latency may be something like 50 to 70 ns very roughly.
L1 cache still is well over 10x faster.
Others might not like to hear this but this might get fixed for good when Firefox hits version 57 and it doesn't include XUL for the interface anymore. I'm not even blaming XUL, I think it's been great - Firefox just looked like a normal desktop application no matter the OS, I'm just saying the GUI implementation will be different.
If that theory works out, you might try Firefox Aurora 57 a few months from now and see if it's fast.
I do think newer versions perform better, i.e. Firefox 51, 52 are good.
Slow to open? I've not noticed as for five seconds to open something that will run for hours or weeks, I don't care. But yes it will gobbles resources, unless you seek to use lightweight web sites only!
By this point computer hardware without RAM slots feel dumber and dumber. Like, bringing the max RAM limit to 32GB or 16GB (on Atom) or 64GB or possibly more could be a reason to upgrade from old hardware.
It was pretty and safer than XP, the only problem is it required the best grade of everything. Like its successor, it makes 1024x768 feel cramped while XP was kind of comfortable at 800x600. So, you kind of needed a monitor upgrade if you wanted to fit a file manager and music player side by side. Also, sticks of 1GB DDR1 and 10k rpm hard drives weren't that cheap or common.
Well that's one more reason to not come to the US. Believe it or not there are countries where peeing on the side of the road isn't wrong and doesn't make you run the risk of being handcuffed, arrested, and charged with a sex crime.
The brain does seem to utilize what is has better than computers, e.g. in computer chips memory cells store bits and communicate a binary signal. This could be all analog and exploit the voltage range better instead. (Flash memory cells now store a charge which can be at four, eight or sixteen (experimental) levels and translate that to two, three or four bits, that's why you can have a cheap 32GB or 64GB memory card)
A large part of chips is spent ferrying data around and worrying about the clock (although asynchronous CPU exist too). Brain is asynchronous and perhaps does quite some amount of ferrying signals from one place to another too, but along every step there's quite some molecular and chemical level "computing" that we understand about nothing about (I put "computing" between quotes, because comparing it to computing or data processing might be entirely wrong)
The brain is enclosed in an actually small space. It's huge next to a couple square centimeters of chip but next to a 42U cabinet or a room full of these it's fairly small. Computers have to use terribly slow buses like PCIe, Infiniband, Ethernet and even the DRAM is dog slow from the CPU's viewpoint. We might like to see the brain as (even more) terribly slow but it's kind of a 1.5 liter 3D "CPU" continuum with zillions of elements and mega-zillions of connections.
And it self-modifies constantly - good luck having your Itanium, ARM or GPU rearrange itself.
It's debatable whether we will ever be able to manufacture anything like that. It also has a whole network to carry fuel and organic materials in, waste and heat out. Other things to allow it not instantly die of cancer or infection. Although, no one knows what happens if you were to completely remove a brain (and spinal cord etc. maybe) from the living body it's attached to : maybe it would go insane in a matter of seconds and die out.
Which is to say, from a hardware stand the brain probably is severely underestimated by most or many people.
Software neural networks are an extremely simplified and idealized representation that only deals with a limited high-level subset of electrical activity we could make sense of some decades ago. There's a lot of smaller scale, or short scale stuff left out as well as all the chemical activity - if neural networks worked a bit more like a brain, we would be able to try the algorithms with caffeine, heroin, cocaine, cannabinoids, opioids and other substances, to study whether and how the algorithms are working differently.
'A simulation of intelligent behavior must appear to be intelligent.'
Does it? I would say a simulation of intelligent behavior must appear to be behavior. That's like teaching dogs to speak maybe, and either AI will never be more than that or we'll never get AI. Even a self-driving car is a system tuned for the goal of "Drive from A to B without killing or maiming anyone" and will never be able to do anything else - save for additional functions like running an inference engine to play Go on the car's GPUs, or putting some Microsoft Cortana kind of software on the dashboard to enter your destination by speaking.
One simple example is a black stone surrounded by white stones. It may happen on the board before removing the stone but I doubt it counts as a position. Have two such occurrences for the position to be definitely illegal.
Turns out, I did once read about it and to sum it up the number of legal positions is about one in 81. It was only computed on Jan 2016 :)
https://tromp.github.io/go/leg...
The software used for these computations is available at my github repository. Running "make" should compute L(3,3) in about a second. For running an L19 job, a beefy server with 15TB of fast scratch diskspace, 8 to 16 cores, and 192GB of RAM, is recommended. Expect a few months of running time. (...)
Then you might want to count legal or possible positions, which might have been done or merely estimated but likely not an easy problem in itself.
Choosing the values might be hard?
Not only there's the problems of discarding, clamping, correcting data (if you remember "hide the decline", maybe that's stuff you will have to do) but I wonder if you'll end up doing something silly. E.g., your system ends up increasing the production of men's underwear when the number of pirate attacks is low. Then next year a outlying bout of 1000 pirate attacks happens, your program falls apart but before the underwear company notices they shouldn't blindly trust what comes out of the computer, they will miss a ton of sales and their underwear competitors will mop the floor with them.
And soon, there won't be programmers anymore because computers will program themselves :)
You only have to write the program's specification (a square mile of UML diagrams will do).
Can you do something incredibly plain and unremarkable? e.g., email. Like, everyone hates email as they have bloated web GUIs, all tons of crap mixed in there (administrative, personal, pointless subscriptions)
Like, you could give a company, team or group their own squirrelmail, and also their own IRC accessed with the same login for instance. Clean, free of crap and everyone understands what it's about.
You heard about it here lol.
Advertise to local businesses. Well, maybe there's only a trickle of money to be made and it's not really a software startup, but I wanted to make the joke. Provide the LEAST amount of innovation possible but with useful benefits to the customer. Like, Google and Wikipedia got absolutely huge by providing people what they wanted - black text on a white background.
Removing underlined Gtk3 letters was the one stupid ass thing. Can't they have a system wide setting so they can hide them on their Gnome or other 3D desktop and look good in their marketing screenshots, but leave them for users of traditionals GTK2/GTK3/Qt etc. desktops?
Other than that, GTK3 still works fine for normal looking, normal working software, usually to the point of not noticing whether your app runs GTK2 or GTK3. I like it, and dislike Qt4/Qt5 instead (for no particular reason).
On occasion I found it useful as there's some networked software where it's pointless not to run the latest version and the stub installer is good for this.
Debian and ubuntu even have it for the whole operating system! The "downloader installer stub" (net install) for debian and ubuntu can use a http proxy, incidentally.
Sometimes the problem if you want the full installer is to find the download link. If not available at all, that sucks.
Ubuntu is the worst at allowing you to find the downloads on their damn web page, even the vanilla default download! (full iso, main edition, 64bit x86). I think they like to change their web site as well. If you want to find the "installer stub" for Ubuntu that'll be actually a lot harder than the full isos unless you web-search for it.
So much, I guess I can recommend debian or Linux Mint because you can go to the site and find the files to download.
Now, the stupid bit about Ubuntu : I saw the Ubuntu desktop on TV (in some documentary or news report about something else), have downloaded it to rescue a few acquaintances from Windows 7/8/10 and then there's big news coming to say it's deprecated. Well I'll install it probably, since the installations shall be replaced in 2020 or 2021 and by then, who knows what will be deprecated or abandoned anyway.
Intel graphics increasingly are meant to run games : foremost, if some people upgrade to a 4790K or 6600K it might have the best graphics hardware they ever had, because it beats e.g. geforce 7600GT (~2006) or Radeon 5450 (~2009) and others.
Then, they got a bit better of their goal of being barely able to run anything. Like, the two cards I mentioned above are deprecated. In fact, look at the low end market and the fact you still can buy geforce 210, Radeon 5450, 6450, R5 230. That's deprecated! (nvidia provides a "legacy" linux driver that you still can use)
R7 240 and 250 are not deprecated officially (but long in the tooth), but don't have a good linux driver yet.
So given all the bullshit and museum pieces on the low end graphics market, Intel graphics looks like a good choice. It's good that it sits under the CPU's heatsink.
There is one usable low end graphics board : geforce GT 710. Although, it doesn't have displayport. Roughly similar to Intel graphics in terms of power, but with a better driver. But getting older already.
(thanks, that's interesting)
Even on chip buses are as fast as they need to be, or often too slow (typically, two packs of four CPU cores have a slowish networking between them and aren't to be used arbitrarily as an eight core CPU)
You can have a state of the art interconnect between CPU and GPU and memory etc. on a high end chip, sure.
CPU and GPU might be able to share memory addresses, function like AMD's heterogeneous computing promises to do (or does already, just not used very much). Not only AMD, the HSA foundation has the usual mobile CPU/GPU/SoC suspects. Possibly Apple would aim for such abilities in developing a SoC with Apple CPU and Apple GPU, and I wonder if you were thinking of something like that. If so, there will eventually be an announcement of Apple joining the HSA foundation.
But whether Apple joins in or not, this might be very interesting technically speaking.. But we'll need software that uses it.
If Apple does make such an integrated CPU + GPU combo, it would need e.g. some powerful video editor for iPad Pro and iPad to show off the abilities, or some other example.
I originally wanted to point out the "modular" external GPU is available in just one product, Microsoft Surfacebook with the keyboard dock that includes battery and GPU (so, Intel CPU + GPU combo in the main unit, and low power nvidia in the keyboard dock).
So I see no overarching reason why Apple can't do the same with Apple CPU + GPU in an iPad, and Apple GPU in a powered dock. You can even run some code that likes lower latency on the internal GPU, and some code ok with high latency on the external GPU. Just saying it's technically possible, not that they would bother with such a contraption - they'll probably want to sell you a Mac instead, not the equivalent of a "Sega 32x" or sidecar expansion for the iPad/iPad Pro/iPhone plus.
The Northbridge and PCIe for GPU are integrated in all desktop/laptops CPUs now (save for AM3+ FX 6000 and 8000 series that you still can buy)
Of course this doesn't change much at all. PCIe bandwith and latency and joule per bit on PCIe still are the same.
What would work better is an MCM, CPU and GPU are packaged close together and use custom or semi-custom interface. It does exist, that's AMD's upcoming Opteron that includes on the same package : two Ryzen dies, one Vega 2048 SP / 32 CU die, HBM2 memory for the GPU die. I'm sure it will be awesome but it will work great at 180 watts or 250 watts, not 1.8 watt or 2.5 watts.
There's also Haswell or later at lowest wattage : on a single package, a CPU (with GPU and shit integrated) and a chipset are connected, closely and low powery. The "MacBook" (thing with only one USB port) and Surface Pro can use that. But it's a bit similar to a phone using a "do everything" SoC and a radio/modem chip on the side.
All of that to support your conclusion but theoretically, a large Ipad / Ipad Pro can use a discrete GPU, though still very unlikely.
Can optic communication between two chips reduce the power? That would be one way to make it more possible, although likely again some expensive thing for $1000 and up tablets or other small mobile computers.
Ultra fast external communications are power hungry in general, what's telling is 4K 60Hz and up may require real time compression even between a device's chip and the device's display panel (there's a VESA standard for this) and 10 gigabit RJ45 ethernet failed just because of this.
If it's there, and you can see it . . . .
If it's not there, but you can see it . . .
If it's not there, and you can't see it . .
We should ask Donald Rumsfeld about this then.
Also, for the case of "If it's there, but you can't see it".
Alternatively you might have something to teach Rumsfeld about!