In Windows 8, if I search for 'update', it prominently says "no results". Intuitively I think "oh it must not be there" not "oh I should look over at the right column and see there is a category that has more than '0' to find the results.
IIRC, Win7 will display all the results rather than forcing you to switch categories.
my wife's Prius is nearing the end of its aesthetic life
Curious as to how long you think a car should last. In our household, we have a 1999 vehicle with no plans to replace in the near future (no significant mechanical issues that drive repair cost higher than its worth. While you might live in Japan or have imported on older Prius, I doubt your prius is as old as '99. Some of us amortize the cost of our vehicles over a longer period of time.
If nouveau can do it, and they're not crying trade secrets over it, then nvidia proper can do it too.
If nouveau worked just as well as proprietary drivers, then no one would be bothered. The fact is nVidia is doing *something* that makes them much much much faster than nouveau, and it's difficult to speculate what can and can not be done without destroying performance.
I would assume that could/would drive cost up. Suddenly you need infrastructure on the device to support a piece of code that isn't a problem for 97% of your customers that adds a few dollars to the price, which kills margins on the integrated offerings.
Besides, I presume part of the whole need of the driver is to manage how data flows through the whole system, not just what happens inside the GPU itself.
Challenge being, what Optimus like technology actually works in Linux? I'm unaware of anything off hand. When you have nothing to compare with, what are you supposed to do? Perhaps the life lesson is to look up the product before purchase regardless of vendor to understand what is going to happen.
On the other hand, of the technology that *has* viable alternatives, how does nVidia stack up, capability wise, to the competition? In terms of accelerated 3D, AMD and nVidia proprietary drivers are both very capable, and none of the open source (including Intel GPU) comes close. In the case of Intel, you have to limit the comparison to similarly weak AMD/nVidia hardware, but even then it still lags significantly. One oddity though. With Mythfrontend, any attempt to use GL painter or renderer fails miserably with AMD. This either indicates an AMD bug or a MythTV bug, but the latter is still AMD's problem since clearly the developers do not target their platform.
In terms of video decode acceleration, nVidia VDPAU more or less stands alone. VA-API comes in second, but even with a brand new Ivy Bridge GPU, the experience is far more poor than VDPAU on a low end nVidia GPU that's 4 years old in projects that support both. VDPAU also has more project support. XvBA doesn't even warrant a mention, between the crappy capabilities and only one experimental branch of a major multimedia project bothering.
Video playback tear-free is a massive issue for everyone except nVidia. They are the only ones that do tear-free XV (though usually not needed due to vdpau). They also are the only ones I've tried where OpenGL renderers have worked consistently well.
In terms of kernel module signing, it looked like only RedHat was planning to play that game, and Ubuntu was pursuing a more minimalist approach where the signing requirement is removed as soon as possible. Even when they do 'play the game', I wager it will be like SELinux, strongly encouraged to be on but an optional capability that can be disabled. It probably will disabled first thing by a ton of users because it's too much effort to deal with (much like SELinux). Lack of stable driver ABIs in Linux make signed module requirement in some cases just too limiting to possibly deal with.
Instead of trying to break the internet, refresh your business practices so I can actually legitimately pay for what I want. As it stands, I ignore it as I think infringement is wrong, but from a business perspective my boycott without piracy is no better than my boycott if I were leeching. If you want to actually get money from those infringing *and* those of us who abstain, fix your business processes. Don't force us to endure physical media, drm, and/or streaming.
Right, I do that as well, but the tweaks to advacedsettings.xml are non-obvious and could be simplified. If I had time, that would be an area of interest for me to contribute.
I use XBMC and find it pretty solid. There are things I'd like, mostly not GUI related -ATSC embedded closed caption support (this feature would make me drop mythfrontend in a *second*) -No 'headless' xbmc. For central library, an xbmc instance must be running. It does pretty much everything needed to do api calls to do database maintenance, but has to have a display -A more simplified cookie cutter setup for centralized database. I know a lot of people will say 'PMS', but I've found that less useful (and more burdensome) than XBMC database.
I found the UI pretty decent actually. Now 'x' versus 'Tab' versus 'Escape is a little non-obvious at first in terms of 'how the hell do I get back to the video that didn't stop when I hit escape' or 'why didn't my video stop when I hit escape', but on a remote, things actually can be intuitively mapped.
gPXE nor iPXE in UEFI can start a linux kernel directly, only EFI executables. They can in BIOS mode, but in UEFI mode you'll have to chain to something like... elilo....
Seems like this leaves things open for an MS rootkit. A rootkit that happens to have an entry point resembling a linux kernel seems a likely scenario.
Also surprised with efilinux. It can load from block devices only, which omits network boot. I understand that grub2 GPL3 concerns make sense, but you would think they might go with elilo. It may be less 'active', but it is capable of doing more than efilinux, notably network deployment.
The problem is I acknowledge the market is too small to get to acceptable volumes by itself. I fear a future where the few of us that like it can no longer share a common hardware platform with a popular software stack.
Problem being, today the relative minority enjoy the hardware out there primarily designed to run MS software but also can run Linux due to the way things historically panned out.
If tomorrow the 95% of hardwer designed to run MS is tilted to run MS *exclusively*, the minority no longer has a significant, affordable market to buy in.
You do realize that MS by and large already dictates hardware standards? ACPI, EFI, Direct3D, and all sorts of things have been either dictated or highly influenced by MS. Hardware vendors must accomodate MS, not vice-versa. This is already the case, but MS just entertains a much wider scope of hardware.
Can you point to a case where MS did work on their part to support hardware the vendor has made no effort to assist? Of course not, it has never happened. There are examples where the vendor was not interested in MS support, and MS just withholds certification for those pieces of hardware.
I understand how to fix many things with my car. I can go down to an auto parts store and pick up all sorts of components for cars. Admittedly, there is less standardization than you see in computing today, but I can reliably find oil filters, air filters, alternators, brake pads, and so on and so fourth for my car. People can do far more drastic things with their cars than I do (electric conversions, adding forced induction, signififcant suspension tuning, and so on).
This is *not* where the consumer computer market seems to be going. Instead, we are talking about devices where any hardware or software problem causes a need to dispose of the whole thing. If software on a piece of hardware is 'old', the hardware is ditched too.
Most of the datacenter equipment has less in common with the current desktop components than most would like. For example, generally more expensive registered DIMMs with ECC are used. Xeon EN or higher, meaning pretty expensive processors. Where GPUs are employed, they are frequently models without any video ports at all. Motherboards are frequently proprietary form factor custom designed for the chassis they go into. About the only solace is that pretty typical SATA drives have a pretty strong presence alongside higher end 15k SAS drives. However, the driver of use of SATA for storage is largely the commodity pricing from commonality with desktops, without desktop volume, SATA prices will rise and/or be ignored for SAS interfaces.
Basically, if the market ditches socketed consumer components, the server market is not going to be a force to keep prices where they are now.
It gets interesting down the list too... SuperMUC is a Sandy Bridge install without GPUs, gets more performance than Tianhe also at lower power. One of the big big things about GPGPU was drastically better performance per watt, but SuperMUC beating Tianhe is really interesting. I really expected a stronger GPGPU showing, but it just isn't happening...
The very rough idea is obvious. Going through the specifics of instrumenting a facility, of determining what the acceptable temperature and flow rate are to keep cpu die temperature at an acceptable level (note, if your cpu is still kicking, that may not be enough, voltage leakage increases with temperature, meaning power draw goes up, and you are being inefficient by letting the die get *too* hot. Also, this is the fastest x86 based system in the world. In part because the cooling is adequate to let the cpu frequency bump up more consistently. Go too far in the other direction, and you are spending too much money on cooling for diminishing returns in die temperatures and the benefits lower die temperatures bring.
We are also talking about a plumbing setup designed to keep ~10,000 servers serviceable without getting the conductive water in the wrong places. This means some significant consideration in how the plumbing connects and how to make it quickly disconnect without leaking to replace parts. The risk for a single water cooled system is small, you paid maybe too much for it and a limited volume of water is in the closed loop. This scale is a different thing.
This isn't just some overblown overclocking setup hooked into an aquarium. This is a non-trivial amount of engineering that goes into it.
This was a BlueGene, meaning in all likelihood the processors and circuit boards were manufactured in New York. There were probably memory,storage, and other components sourced from southeast asia, but the most expensive bits probably were actually fabricated domestic.
At least in the *specific* performance characteristic of 64bit precision linear algebra, it's perfectly likely that the biggest player is reported.
In the cases where secrecy is probably preventing you from knowing about it, it probably is optimized for 32-bit precision floating point and/or large storage throughput to fuel data mining.
Of course, then there are collections of systems that could probably easily place in the list that are at least moderately well-known but not submitted, if it wouldn't be a financial catastrophe to take it down for a few days to dedicate to an xhpl run. An EC2 datacenter comes to mind.
Slashdot Mirror
In Windows 8, if I search for 'update', it prominently says "no results". Intuitively I think "oh it must not be there" not "oh I should look over at the right column and see there is a category that has more than '0' to find the results.
IIRC, Win7 will display all the results rather than forcing you to switch categories.
IBM x3950 x5 can do 3TB of ram with Intel processors.
my wife's Prius is nearing the end of its aesthetic life
Curious as to how long you think a car should last. In our household, we have a 1999 vehicle with no plans to replace in the near future (no significant mechanical issues that drive repair cost higher than its worth. While you might live in Japan or have imported on older Prius, I doubt your prius is as old as '99. Some of us amortize the cost of our vehicles over a longer period of time.
If nouveau can do it, and they're not crying trade secrets over it, then nvidia proper can do it too.
If nouveau worked just as well as proprietary drivers, then no one would be bothered. The fact is nVidia is doing *something* that makes them much much much faster than nouveau, and it's difficult to speculate what can and can not be done without destroying performance.
I would assume that could/would drive cost up. Suddenly you need infrastructure on the device to support a piece of code that isn't a problem for 97% of your customers that adds a few dollars to the price, which kills margins on the integrated offerings.
Besides, I presume part of the whole need of the driver is to manage how data flows through the whole system, not just what happens inside the GPU itself.
I haven't figured out a way with just the OSS tooling to correct for overscanning on my TV.
Challenge being, what Optimus like technology actually works in Linux? I'm unaware of anything off hand. When you have nothing to compare with, what are you supposed to do? Perhaps the life lesson is to look up the product before purchase regardless of vendor to understand what is going to happen.
On the other hand, of the technology that *has* viable alternatives, how does nVidia stack up, capability wise, to the competition? In terms of accelerated 3D, AMD and nVidia proprietary drivers are both very capable, and none of the open source (including Intel GPU) comes close. In the case of Intel, you have to limit the comparison to similarly weak AMD/nVidia hardware, but even then it still lags significantly. One oddity though. With Mythfrontend, any attempt to use GL painter or renderer fails miserably with AMD. This either indicates an AMD bug or a MythTV bug, but the latter is still AMD's problem since clearly the developers do not target their platform.
In terms of video decode acceleration, nVidia VDPAU more or less stands alone. VA-API comes in second, but even with a brand new Ivy Bridge GPU, the experience is far more poor than VDPAU on a low end nVidia GPU that's 4 years old in projects that support both. VDPAU also has more project support. XvBA doesn't even warrant a mention, between the crappy capabilities and only one experimental branch of a major multimedia project bothering.
Video playback tear-free is a massive issue for everyone except nVidia. They are the only ones that do tear-free XV (though usually not needed due to vdpau). They also are the only ones I've tried where OpenGL renderers have worked consistently well.
In terms of kernel module signing, it looked like only RedHat was planning to play that game, and Ubuntu was pursuing a more minimalist approach where the signing requirement is removed as soon as possible. Even when they do 'play the game', I wager it will be like SELinux, strongly encouraged to be on but an optional capability that can be disabled. It probably will disabled first thing by a ton of users because it's too much effort to deal with (much like SELinux). Lack of stable driver ABIs in Linux make signed module requirement in some cases just too limiting to possibly deal with.
Instead of trying to break the internet, refresh your business practices so I can actually legitimately pay for what I want. As it stands, I ignore it as I think infringement is wrong, but from a business perspective my boycott without piracy is no better than my boycott if I were leeching. If you want to actually get money from those infringing *and* those of us who abstain, fix your business processes. Don't force us to endure physical media, drm, and/or streaming.
Right, I do that as well, but the tweaks to advacedsettings.xml are non-obvious and could be simplified. If I had time, that would be an area of interest for me to contribute.
1920x1080 MPEG2 is actually still challenging (e.g. ATSC content).
'Scene' rules haven't stopped me from getting higher profiles on various pieces of content.
I use XBMC and find it pretty solid. There are things I'd like, mostly not GUI related
-ATSC embedded closed caption support (this feature would make me drop mythfrontend in a *second*)
-No 'headless' xbmc. For central library, an xbmc instance must be running. It does pretty much everything needed to do api calls to do database maintenance, but has to have a display
-A more simplified cookie cutter setup for centralized database. I know a lot of people will say 'PMS', but I've found that less useful (and more burdensome) than XBMC database.
I found the UI pretty decent actually. Now 'x' versus 'Tab' versus 'Escape is a little non-obvious at first in terms of 'how the hell do I get back to the video that didn't stop when I hit escape' or 'why didn't my video stop when I hit escape', but on a remote, things actually can be intuitively mapped.
gPXE nor iPXE in UEFI can start a linux kernel directly, only EFI executables. They can in BIOS mode, but in UEFI mode you'll have to chain to something like... elilo....
Seems like this leaves things open for an MS rootkit. A rootkit that happens to have an entry point resembling a linux kernel seems a likely scenario.
Also surprised with efilinux. It can load from block devices only, which omits network boot. I understand that grub2 GPL3 concerns make sense, but you would think they might go with elilo. It may be less 'active', but it is capable of doing more than efilinux, notably network deployment.
The problem is I acknowledge the market is too small to get to acceptable volumes by itself. I fear a future where the few of us that like it can no longer share a common hardware platform with a popular software stack.
64 cores should be enough for anyone!
My concern being does this bleed over to the desktop and laptop market in time. Like it or not, MS dominates that particular space.
Problem being, today the relative minority enjoy the hardware out there primarily designed to run MS software but also can run Linux due to the way things historically panned out.
If tomorrow the 95% of hardwer designed to run MS is tilted to run MS *exclusively*, the minority no longer has a significant, affordable market to buy in.
You do realize that MS by and large already dictates hardware standards? ACPI, EFI, Direct3D, and all sorts of things have been either dictated or highly influenced by MS. Hardware vendors must accomodate MS, not vice-versa. This is already the case, but MS just entertains a much wider scope of hardware.
Can you point to a case where MS did work on their part to support hardware the vendor has made no effort to assist? Of course not, it has never happened. There are examples where the vendor was not interested in MS support, and MS just withholds certification for those pieces of hardware.
I understand how to fix many things with my car. I can go down to an auto parts store and pick up all sorts of components for cars. Admittedly, there is less standardization than you see in computing today, but I can reliably find oil filters, air filters, alternators, brake pads, and so on and so fourth for my car. People can do far more drastic things with their cars than I do (electric conversions, adding forced induction, signififcant suspension tuning, and so on).
This is *not* where the consumer computer market seems to be going. Instead, we are talking about devices where any hardware or software problem causes a need to dispose of the whole thing. If software on a piece of hardware is 'old', the hardware is ditched too.
Most of the datacenter equipment has less in common with the current desktop components than most would like. For example, generally more expensive registered DIMMs with ECC are used. Xeon EN or higher, meaning pretty expensive processors. Where GPUs are employed, they are frequently models without any video ports at all. Motherboards are frequently proprietary form factor custom designed for the chassis they go into. About the only solace is that pretty typical SATA drives have a pretty strong presence alongside higher end 15k SAS drives. However, the driver of use of SATA for storage is largely the commodity pricing from commonality with desktops, without desktop volume, SATA prices will rise and/or be ignored for SAS interfaces.
Basically, if the market ditches socketed consumer components, the server market is not going to be a force to keep prices where they are now.
It gets interesting down the list too... SuperMUC is a Sandy Bridge install without GPUs, gets more performance than Tianhe also at lower power. One of the big big things about GPGPU was drastically better performance per watt, but SuperMUC beating Tianhe is really interesting. I really expected a stronger GPGPU showing, but it just isn't happening...
The very rough idea is obvious. Going through the specifics of instrumenting a facility, of determining what the acceptable temperature and flow rate are to keep cpu die temperature at an acceptable level (note, if your cpu is still kicking, that may not be enough, voltage leakage increases with temperature, meaning power draw goes up, and you are being inefficient by letting the die get *too* hot. Also, this is the fastest x86 based system in the world. In part because the cooling is adequate to let the cpu frequency bump up more consistently. Go too far in the other direction, and you are spending too much money on cooling for diminishing returns in die temperatures and the benefits lower die temperatures bring.
We are also talking about a plumbing setup designed to keep ~10,000 servers serviceable without getting the conductive water in the wrong places. This means some significant consideration in how the plumbing connects and how to make it quickly disconnect without leaking to replace parts. The risk for a single water cooled system is small, you paid maybe too much for it and a limited volume of water is in the closed loop. This scale is a different thing.
This isn't just some overblown overclocking setup hooked into an aquarium. This is a non-trivial amount of engineering that goes into it.
Also, this is a Blue Gene, meaning IBM, not Intel fabs.
This was a BlueGene, meaning in all likelihood the processors and circuit boards were manufactured in New York. There were probably memory,storage, and other components sourced from southeast asia, but the most expensive bits probably were actually fabricated domestic.
At least in the *specific* performance characteristic of 64bit precision linear algebra, it's perfectly likely that the biggest player is reported.
In the cases where secrecy is probably preventing you from knowing about it, it probably is optimized for 32-bit precision floating point and/or large storage throughput to fuel data mining.
Of course, then there are collections of systems that could probably easily place in the list that are at least moderately well-known but not submitted, if it wouldn't be a financial catastrophe to take it down for a few days to dedicate to an xhpl run. An EC2 datacenter comes to mind.