Out of ideas about phones, Apple decides to become a TV network. Hey, doesn't this conflict with the strategy of selling phones for ever higher prices so there are fewer buyers?
even for intel it would take 2 years of profits to build that fab
So I've been saying this for a while now. Intel has no choice but to go fabless. Not if, but when. I am thinking, Intel will never build a full EUV fab, 10nm deep UV will be their last hurrah. Charlie now starts a rumor that Intel will bow out even before that. Plausible, maybe even sensible, but I'll go with my gut feeling for now: Intel is deep enough into 10nm that they will carry on to the bitter end. (To what incidentally will be called 10th generation, at 10nm, because of the emergency insertion of the stopgap 9th generation on the old 14nm fab lines.)
The combined profits of all the front line semiconductor vendors in the world are more than enough to build the upcoming fab generations comfortably. If they don't pool their resources and do it then they will lose massively because users will stop upgrading. Is this enough to get them cooperating?
After Intel drops that leaves just two. I then expect TSMC to push out Samsung by virtue of a much larger R&D budget. So TSMC wins the game of monopoly and can charge whatever it wants, all other technology companies must pay tribute. Obviously, this isn't a likely outcome because of the massive interests involved, but that is the direction things are headed as of today.
Very cool, this reinforces my decision to switch brand allegiance from Google Nexus to Moto. Last Google branded phone was Nexus 4, it seems to me that Google lost the plot after that, possibly due to Apple Envy.
Now the next thing I want from Motorola: right to upgrade. I should not stop getting ROM updates after just two years or so. Cutting off ROM updates doesn't make me upgrade sooner, it just makes me angry and more likely to switch brands. I'm ok with offline update, it doesn't need to be OTA. Just stand behind your product for more than 2 years.
Of all the apps benchmarked, which ones were compiled with the Intel compiler?
More to the point, modern games don't bottleneck on the CPU any more, all the heavy lifting is done on the GPU. So single core benches just don't rule the world like they used to. Now, Intel's go to strategy for cheating the benchmarks is to avoid benchmarking Vulkan/DX12 games so rendering bottlenecks on a single core, and gloss over the fact that this amounts to optimizing for obsolete game engines. Even that skulduggery isn't going to be enough to hold AMD down in 2019.
BTW, this is the main reason that AMD developed Mantle, to exploit their advantage in cores per dollar. Which advantage AMD still very much has as of Intel's 9th gen release last week, which you can't actually buy, or if you can buy them then you pay too much or they run too hot. All of these things the result of trying to shoehorn marketing's 10nm performance specs into engineering's 14nm node.
Right, the pellicle protects the mask. Everybody with EUV plans (just TSMC, Samsung and Intel now, by my count) is going to start introducing EUV without pellicles, but only for larger features like contacts and vias that are widely separated so that dirt accumulating on the mask is unlikely to create chip defects. For a full EUV process as is absolutely required to go beyond 7nm, not using pellicles will prohibitively shorten the mask life. So they are absolutely required. The difficulty with EUV is, all matter becomes opaque, except for extremely thin membranes. Something like 16nm I think. You can see how that might be a bit tricky to fabricate and work with.
We're all waiting on the big breakthrough that will unlock more nodes at a more reasonable cost. Someone will do it and get mega rich. Until then, the investment will go into better exploiting the current nodes to keep ahead of the game in costs and power use. Someone will do that and get mega rich.
You haven't been following the EUV plot. EUV reduces the light wavelength from the current industry standard 193nm to 13.5nm, improving the resolution by a factor of 14, or a factor of 200 if you think of it in areal terms. This is enough to keep Moore's law chugging on for at least the next decade. EUV is right on the doorstep at the moment. Nobody is wringing their hands hoping that some genius will come up with a fundamental breakthrough, this engineering evolution is really just a whole lot of grinding, hard work. No "pivot" as you imagine.
Note that nobody has actually seen TSMC 7nm volume production hit retail channels as of today. Supposedly, Apple will change that in a few weeks, but until they actually do, everybody is still playing bullshit poker.
amd continues to fail in the single threaded and single core performance metrics comparison with intel
"Fail" is the wrong word. A bit behind would be accurate, and that is Intel's last remaining bragging point. A couple of things. Current Ryzen is still a full node behind Intel, that it manages to clobber Intel in multi-core and put in a respectable showing in single core is truly impressive. Second thing, if TSMC actually delivers on time then AMD will suddenly be a node ahead of Intel for the first time in history. Third thing, buzz has it that Zen 2 improves IPC by 13%, which will bring it roughly even in IPC with Intel, while retaining its massive lead in value.
I know this is hard for you, but the post you responded to is about the most accurate thing you could possibly say to OP. If you can't see why then I am sorry for you.
Great job "forecasting" there grandpa, lol? Why don't you give unsolicited micro-lessons on a subject you know something about or actually work in instead? Who knows, someone might even listen slightly or care in the least. K thx bai
Somebody modded it up, and you down, sigh. OTOH, it wasn't necessary to be quite that abrasive. OTOOH, you were responding to possibly the worst post in the entire universe. The only sorrier thing I can think of is, somebody modded it up.
5nm (using TSMC terminology) will be even more nuts than 10nm/7nm because it will be fully EUV. EUV is just scary. Consider there: it was only in the last couple of months that ASML managed to achieve the necessary 250 watt EUV output power, that takes a megawatt of input power. Optical materials are opaque to EUV so its all done with mirrors. Each mirror in the chain, and there are many, aborbs 20-30% of the power. That all turns into heat. So each stepper has a medium sized creek flowing through it to keep it from melting. Those are just a couple of the horrors.
Another one, pellicles (what's a pellicle?) They aren't ready for prime time. There is exactly one company, ASML, making all the EUV equipment and they are currently burning the midnight oil trying to develop usable pellicles. As I understand it, they currently aren't quite transparent enough (88% minimum required vs 83% current maximum achieved) or durable enough, by about a factor of three. Without pellicles, nobody is making any chips with EUV, and nobody is getting past 7nm.
We have reach an inflection point where the foundry industry must become a commons to make further progress. I really can't think of any historical analogy, this is something new. The costs and risks are just too high for any one player to bear them alone, and yet, competing vendors cannot tolerate one single company getting a choke hold on the entire industry.
Samsung is showing impressive resilience, they will be the last player to fold, leaving the entire pot to TSMC. Intel will give up their fabs within two years, just run them until they can't sell any more 14nm and 10nm parts, then bulldoze them.
If industry players are unable to negotiate a formula for sharing fabs at future nodes then governments will. That will be messy.
The article you linked compares Intel to Global Foundries, which doesn't even have a 7nm process as of today. You're still right, but try mylinks
Note that Samsung also went for 36nm minimum metal pitch and for what it's worth, also seem to be behind TSMC by about the same lag as Intel. It's starting to look like TSMC went for exactly the right amount of conservative.
I think that Intel is behind by exactly as much as I think:) Translation: Intel is now behind by about a year, or half a node, whatever that is.
Correction: Intel's 10nm minimum metal pitch is only 10% smaller than TSMC's 7nm, not 22% as I wrote above. That is still enough to explain why Intel is having much worse problems than TSMC at the extreme limit of what deep UV can do.
Two days before AMD reports, three days before Intel reports, Semiaccurate floats this rumor with pretty much nothing to back it up. Here's the meat of their argument Note: The following is analysis for professional level subscribers only. So this is about signing up subscribers? Or an attempt at illegal stock manipulation? Both? It is certainly not about quality journalism.
I am definitely an AMD fanboy, full disclosure there. But that doesn't make me an Intel hater, at least not when they lay off the dirty tricks, which appears to pretty much the situation at the moment. So... balanced assessment: no reason to doubt Intel's revised 10nm production schedule. This is all about yields as Semiaccurate is fond of pointing out.
You can see from this that Intel's 10nm fin pitch is a bit more aggressive than TSMC's 7nm, 6% smaller. Intel's minimum metal pitch is a lot more aggressive, 22%. This is all right at the limit of what deep UV alone can do, so that might be Intel's bridge too far right there. I have a whole lot of difficultly believing that Intel did not learn enough from their aborted ramp up last spring to know exactly what they need to do to hit their yields, most probably including respinning their masks to a density nearly identical to TSMC.
Buried in there somewhere I did find one credible little nugget... Semiaccurate pointed out that last spring's 8121U Cannon Lake part, produced in limited quantities and only ever seen in the hands of a few reviewers, is specced without a GPU. Not because it doesn't have one, but because does have one but it doesn't work. I find that credible. Debugging both a processor and a GPU is much more work that just a processor or GPU alone. In contrast, AMD doesn't try to fab APUs until both the processor and GPU have been successfully fabbed separately. Excellent strategy, a big risk reduction.
Another huge thing AMD did to cut the 7nm risk was, jumping into bed with the phone industry. Intel convinced themselves it was a good idea to go it alone as usual, and were proved colossally wrong. Though I am not going to claim any special inside information, I think that Intel is going to bring up its Cannon Lake production successfully, 3 or 4 years behind schedule as they say, and that this is the end of the line for Intel as an independent fab. It's simple: the days of always being a node ahead are over, today they are half a node behind. From here on, there are no advantages to running an independent fab, only disadvantages. When Intel finally does ramp up Cannon Lake they will be in an excellent position to negotiate a new, cooperative deal with the rest of the industry, but if they persist in marching to their own drumbeat they will pay an enormous cost in market share and operating income over the next few years.
I am going to take a wild guess here: Intel plays around with EUV a bit, gets some first hand data on what horribly nasty stuff that is, then makes a deal with TSMC. Intel is going to do just fine as a pure Engineering/IP player like AMD but they risk everything by running their own vanity fab.
Codes of Conduct are only "required" if you have a sudden influx of intolerable people...such as the ones that show up screeching for a "Code of Conduct."
Not sure what picture you are trying to paint, but in every case that I know of, code of conduct was introduced by long term project members following discussions amongst long term project members. If you have an example where "intolerable people" demanded and got a code of conduct forced upon a project, could you please provide the link.
With a few exceptions, every single project I've worked in over the course of two decades has been open and welcoming.
Not sure what "with a few exceptions, every single project" means. Anyway there are notable exceptions. If your project starts making headlines in the mainstream news for being dysfunctional then clearly something needs to be done. It is not a question of whether, but what.
Yes, it is a bad code of conduct. Mixed together with many high minded desiderata there are multiple explicit religious exhortations such as "deny oneself in order to follow Christ." There is also a great deal of fluff having little to do with software development or community behavior such as "chastise the body" and "clothe the naked." I dunno, what is the one piece of advice I most want to give to D. Richard Hipp? Maybe: "review before posting".
In which alternate universe is that post flamebait?
Slashdot Mirror
Out of ideas about phones, Apple decides to become a TV network. Hey, doesn't this conflict with the strategy of selling phones for ever higher prices so there are fewer buyers?
Apple vs Netflix, I will bet on Netflix.
even for intel it would take 2 years of profits to build that fab
So I've been saying this for a while now. Intel has no choice but to go fabless. Not if, but when. I am thinking, Intel will never build a full EUV fab, 10nm deep UV will be their last hurrah. Charlie now starts a rumor that Intel will bow out even before that. Plausible, maybe even sensible, but I'll go with my gut feeling for now: Intel is deep enough into 10nm that they will carry on to the bitter end. (To what incidentally will be called 10th generation, at 10nm, because of the emergency insertion of the stopgap 9th generation on the old 14nm fab lines.)
The combined profits of all the front line semiconductor vendors in the world are more than enough to build the upcoming fab generations comfortably. If they don't pool their resources and do it then they will lose massively because users will stop upgrading. Is this enough to get them cooperating?
After Intel drops that leaves just two. I then expect TSMC to push out Samsung by virtue of a much larger R&D budget. So TSMC wins the game of monopoly and can charge whatever it wants, all other technology companies must pay tribute. Obviously, this isn't a likely outcome because of the massive interests involved, but that is the direction things are headed as of today.
add: how long the vendor provides ROM updates.
Very cool, this reinforces my decision to switch brand allegiance from Google Nexus to Moto. Last Google branded phone was Nexus 4, it seems to me that Google lost the plot after that, possibly due to Apple Envy.
Now the next thing I want from Motorola: right to upgrade. I should not stop getting ROM updates after just two years or so. Cutting off ROM updates doesn't make me upgrade sooner, it just makes me angry and more likely to switch brands. I'm ok with offline update, it doesn't need to be OTA. Just stand behind your product for more than 2 years.
Convenient for Charlie that we need to wait a whole year to find out for sure.
Of all the apps benchmarked, which ones were compiled with the Intel compiler?
More to the point, modern games don't bottleneck on the CPU any more, all the heavy lifting is done on the GPU. So single core benches just don't rule the world like they used to. Now, Intel's go to strategy for cheating the benchmarks is to avoid benchmarking Vulkan/DX12 games so rendering bottlenecks on a single core, and gloss over the fact that this amounts to optimizing for obsolete game engines. Even that skulduggery isn't going to be enough to hold AMD down in 2019.
BTW, this is the main reason that AMD developed Mantle, to exploit their advantage in cores per dollar. Which advantage AMD still very much has as of Intel's 9th gen release last week, which you can't actually buy, or if you can buy them then you pay too much or they run too hot. All of these things the result of trying to shoehorn marketing's 10nm performance specs into engineering's 14nm node.
Right, the pellicle protects the mask. Everybody with EUV plans (just TSMC, Samsung and Intel now, by my count) is going to start introducing EUV without pellicles, but only for larger features like contacts and vias that are widely separated so that dirt accumulating on the mask is unlikely to create chip defects. For a full EUV process as is absolutely required to go beyond 7nm, not using pellicles will prohibitively shorten the mask life. So they are absolutely required. The difficulty with EUV is, all matter becomes opaque, except for extremely thin membranes. Something like 16nm I think. You can see how that might be a bit tricky to fabricate and work with.
We're all waiting on the big breakthrough that will unlock more nodes at a more reasonable cost. Someone will do it and get mega rich. Until then, the investment will go into better exploiting the current nodes to keep ahead of the game in costs and power use. Someone will do that and get mega rich.
You haven't been following the EUV plot. EUV reduces the light wavelength from the current industry standard 193nm to 13.5nm, improving the resolution by a factor of 14, or a factor of 200 if you think of it in areal terms. This is enough to keep Moore's law chugging on for at least the next decade. EUV is right on the doorstep at the moment. Nobody is wringing their hands hoping that some genius will come up with a fundamental breakthrough, this engineering evolution is really just a whole lot of grinding, hard work. No "pivot" as you imagine.
Note that nobody has actually seen TSMC 7nm volume production hit retail channels as of today. Supposedly, Apple will change that in a few weeks, but until they actually do, everybody is still playing bullshit poker.
amd continues to fail in the single threaded and single core performance metrics comparison with intel
"Fail" is the wrong word. A bit behind would be accurate, and that is Intel's last remaining bragging point. A couple of things. Current Ryzen is still a full node behind Intel, that it manages to clobber Intel in multi-core and put in a respectable showing in single core is truly impressive. Second thing, if TSMC actually delivers on time then AMD will suddenly be a node ahead of Intel for the first time in history. Third thing, buzz has it that Zen 2 improves IPC by 13%, which will bring it roughly even in IPC with Intel, while retaining its massive lead in value.
So "fail" is the wrong word, indeed.
Intel needs to give up its fabs like AMD. Sooner or later they will. Sooner will be less painful.
I know this is hard for you, but the post you responded to is about the most accurate thing you could possibly say to OP. If you can't see why then I am sorry for you.
Great job "forecasting" there grandpa, lol? Why don't you give unsolicited micro-lessons on a subject you know something about or actually work in instead? Who knows, someone might even listen slightly or care in the least. K thx bai
Somebody modded it up, and you down, sigh. OTOH, it wasn't necessary to be quite that abrasive. OTOOH, you were responding to possibly the worst post in the entire universe. The only sorrier thing I can think of is, somebody modded it up.
5nm (using TSMC terminology) will be even more nuts than 10nm/7nm because it will be fully EUV. EUV is just scary. Consider there: it was only in the last couple of months that ASML managed to achieve the necessary 250 watt EUV output power, that takes a megawatt of input power. Optical materials are opaque to EUV so its all done with mirrors. Each mirror in the chain, and there are many, aborbs 20-30% of the power. That all turns into heat. So each stepper has a medium sized creek flowing through it to keep it from melting. Those are just a couple of the horrors.
Another one, pellicles (what's a pellicle?) They aren't ready for prime time. There is exactly one company, ASML, making all the EUV equipment and they are currently burning the midnight oil trying to develop usable pellicles. As I understand it, they currently aren't quite transparent enough (88% minimum required vs 83% current maximum achieved) or durable enough, by about a factor of three. Without pellicles, nobody is making any chips with EUV, and nobody is getting past 7nm.
We have reach an inflection point where the foundry industry must become a commons to make further progress. I really can't think of any historical analogy, this is something new. The costs and risks are just too high for any one player to bear them alone, and yet, competing vendors cannot tolerate one single company getting a choke hold on the entire industry.
Samsung is showing impressive resilience, they will be the last player to fold, leaving the entire pot to TSMC. Intel will give up their fabs within two years, just run them until they can't sell any more 14nm and 10nm parts, then bulldoze them.
If industry players are unable to negotiate a formula for sharing fabs at future nodes then governments will. That will be messy.
The article you linked compares Intel to Global Foundries, which doesn't even have a 7nm process as of today. You're still right, but try my links
Note that Samsung also went for 36nm minimum metal pitch and for what it's worth, also seem to be behind TSMC by about the same lag as Intel. It's starting to look like TSMC went for exactly the right amount of conservative.
I think that Intel is behind by exactly as much as I think :) Translation: Intel is now behind by about a year, or half a node, whatever that is.
So Apple Makes Intel Great Again. You know, AMIGA.
OK, that's funny. Pointless but funny. Where's your funny mod?
Correction: Intel's 10nm minimum metal pitch is only 10% smaller than TSMC's 7nm, not 22% as I wrote above. That is still enough to explain why Intel is having much worse problems than TSMC at the extreme limit of what deep UV can do.
Two days before AMD reports, three days before Intel reports, Semiaccurate floats this rumor with pretty much nothing to back it up. Here's the meat of their argument Note: The following is analysis for professional level subscribers only. So this is about signing up subscribers? Or an attempt at illegal stock manipulation? Both? It is certainly not about quality journalism.
I am definitely an AMD fanboy, full disclosure there. But that doesn't make me an Intel hater, at least not when they lay off the dirty tricks, which appears to pretty much the situation at the moment. So... balanced assessment: no reason to doubt Intel's revised 10nm production schedule. This is all about yields as Semiaccurate is fond of pointing out.
You can see from this that Intel's 10nm fin pitch is a bit more aggressive than TSMC's 7nm, 6% smaller. Intel's minimum metal pitch is a lot more aggressive, 22%. This is all right at the limit of what deep UV alone can do, so that might be Intel's bridge too far right there. I have a whole lot of difficultly believing that Intel did not learn enough from their aborted ramp up last spring to know exactly what they need to do to hit their yields, most probably including respinning their masks to a density nearly identical to TSMC.
Buried in there somewhere I did find one credible little nugget... Semiaccurate pointed out that last spring's 8121U Cannon Lake part, produced in limited quantities and only ever seen in the hands of a few reviewers, is specced without a GPU. Not because it doesn't have one, but because does have one but it doesn't work. I find that credible. Debugging both a processor and a GPU is much more work that just a processor or GPU alone. In contrast, AMD doesn't try to fab APUs until both the processor and GPU have been successfully fabbed separately. Excellent strategy, a big risk reduction.
Another huge thing AMD did to cut the 7nm risk was, jumping into bed with the phone industry. Intel convinced themselves it was a good idea to go it alone as usual, and were proved colossally wrong. Though I am not going to claim any special inside information, I think that Intel is going to bring up its Cannon Lake production successfully, 3 or 4 years behind schedule as they say, and that this is the end of the line for Intel as an independent fab. It's simple: the days of always being a node ahead are over, today they are half a node behind. From here on, there are no advantages to running an independent fab, only disadvantages. When Intel finally does ramp up Cannon Lake they will be in an excellent position to negotiate a new, cooperative deal with the rest of the industry, but if they persist in marching to their own drumbeat they will pay an enormous cost in market share and operating income over the next few years.
I am going to take a wild guess here: Intel plays around with EUV a bit, gets some first hand data on what horribly nasty stuff that is, then makes a deal with TSMC. Intel is going to do just fine as a pure Engineering/IP player like AMD but they risk everything by running their own vanity fab.
So you don't have any actual argument or evidence, only invective.
respect is earned, it is not a right
Similarly, disrespect is not a right, though trolls who rail against conduct guidelines seem to believe it is.
Codes of Conduct are only "required" if you have a sudden influx of intolerable people...such as the ones that show up screeching for a "Code of Conduct."
Not sure what picture you are trying to paint, but in every case that I know of, code of conduct was introduced by long term project members following discussions amongst long term project members. If you have an example where "intolerable people" demanded and got a code of conduct forced upon a project, could you please provide the link.
With a few exceptions, every single project I've worked in over the course of two decades has been open and welcoming.
Not sure what "with a few exceptions, every single project" means. Anyway there are notable exceptions. If your project starts making headlines in the mainstream news for being dysfunctional then clearly something needs to be done. It is not a question of whether, but what.
I don't see any of your mythical SJWs posting, but I do see a some trolls such as you posting, apparently trying to stir up controversy.
is it a bad code of conduct?
Yes, it is a bad code of conduct. Mixed together with many high minded desiderata there are multiple explicit religious exhortations such as "deny oneself in order to follow Christ." There is also a great deal of fluff having little to do with software development or community behavior such as "chastise the body" and "clothe the naked." I dunno, what is the one piece of advice I most want to give to D. Richard Hipp? Maybe: "review before posting".
In which alternate universe is that post flamebait?