IF it's really as low as $250K that doesn't actually sound like too bad a deal. it's "only" $125 per gigabyte.
Comparatively I built a 48GB dual quad box a while ago for about £3000 (including VAT but also including our discounts from the supplier) which at current exchange rates works out to just under $100 per gigabyte. I could have got that price lower with slower processors.
Not as large a difference as I expected really but I still think in a ram dominated virtualisation situation the dual socket boxes are likely to win out.
and the Nehalem-EX with 8 sockets go as high as 2TB of RAM per server. IIRC in theory you get four memory buffers per processor each with two channels. With two modules per channel, two channels per expander, four expanders and 8 processors you would get 128 modules. With 16GB modules you would indeed have 2TB of ram.
But does anyone actually sell a board that will take that much ram (the only nahelm ex board i've seen for sale was from supermicro and only had four CPUs and one module per channel so only supported 16 modules)? and how much would the bloody thing cost? (last I checked there was a huge premium on cost per gigabyte for 8GB modules, I dread to think what 16GB is like).
And still people want more RAM per box. Some people do but for many workloads i'd think that beyond a certain point the TCO of adding more boxes would be lower than the TCO of trying to stuff more ram into one box (depending on how much your cost to admin and possiblly license an extra box is of course).
There are at least two seperate issues that need to come together to cause this denial of service flaw.
1: systems that process untrusted data (and yes that means web browsers) need resource limits. If some aspect of loading a page sucks up unacceptablly large ammounts of memory then the browser should stop and give an error. Firefox clearly doesn't. 2: The animiation should not take this much memory to handle. Yes for a short loop of a few frames storing everything uncompressed in memory makes sense. However for long loops and one-shot animations it makes far more sense to work in a streaming manner.
working around an arbitrary limitation in a not-too-far-in-the-future legacy 32-bit platform. Actually on a 64-bit system this security vulerability is potentially MUCH worse. On a modern 32-bit system it likely just means the browser crashes. On a 64-bit system the exploit file it will push the entire system deep into swap potentially bringing the entire system to it's knees.
Really there is a deeper issue, crashing or eating up insane ammounts of swap is not an appropriate response to an item causing excessive memory use in an animation decoder/displayer whether that is the result of a legitimate codepath (in this case converting an animated gif to a sequance of uncompressed images) or a leak or an exploit. Systems like browsers that are exposed to malicious remote entities need resource limits.
3th party chip sets Less high power high speed chips means less power spent on interfacing between them and less complex system cooling. For most systems that is probablly worth sacrificing the ability to choose northbridges.
also apple. They can't stay core2 for ever on the mini / some of there laptops and intel video does not fit in there gpu api. If true that sucks for them but ultimately I don't think in.
It appears (see http://slashdot.org/comments.pl?sid=1786182&cid=33570924 for caveats) that with sandy bridge intel is now beating the best nvidia chipsets for the C2D so it won't be a downgrade on general graphics performance for laptop builders to move from C2D+nvidia chipset to sandy bridge.
With consoles each console has a distinct pool of games. Most games are either released for a single console or for a group of consoles with similar capbilities. Sometimes with a PC release as well but the PC release is often either crap, late or both.
If you want to play recent Mario games you have to buy a Wii (or maybe screw arround with emulation on a PC but most people won't bother to go to that much trouble). If you want to play GTA4 you have to buy a PS3, an xbox 360 or a high end gaming PC. If you want to play recent halo you have to buy an xbox 360. If you want to play recent ratchet and clank you have to buy a PS3.
OTOH PCs all draw from the same pool of games. With a low end PC some games are unplayable and others require you to turn down the settings and/or tolerate slowdowns. With a high end PC you should be able to run all games well (other than driver bugs which can hit both low and high end systems:( ).
My question would be are those wanting a laptop that is too small for a discrete GPU and who care about graphics performance going to be better off staying with a core 2 duo with a nvidia chipset? or will they be better off with sandy bridge?
Unfortunately anandtech ( http://www.anandtech.com/show/3871/the-sandy-bridge-preview-three-wins-in-a-row/7 ) didn't include nvidia integrated graphics in their comparison. Also they were using a desktop not a laptop chip afaict (though they don't seem to know for sure exactly what they had from reading the comments). Integrated graphics performance is far less relevant for desktops than for laptops since it's so easy to add dedicated graphics to a desktop.
Currently with intel stuff (it's a while since i've looked at the AMD side) the laptop and low end desktop platforms have 2 channels and at least with the boards i've seen the max configuration supported is 4x4GB for a total of 16GB.
The current intel high end desktop platform has 3 channels and at least with the boards i've seen the max configuration supported is 4x4GB for a total of 16GB
Workstation/server platforms go much higher, with the right board and a big enough budget you can get 18x8GB (maybe more now if larger modules have appeared) on a dual socket platform and more on the 4/8 socket platform (though all the boards i've seen for that platform have not come even close to maxing out the chipsets memory capability)
All the details i've seen about sandy bridge seem to reffer to a laptop/low end desktop platform. I've seen a mention of a high end variant but no real details of it.
Ultimately for laptops and low end desktops moving all the high speed logic (graphics, CPU, memory controller) into one chip makes a lot of sense from both a cost and a power point of view.
Yes it's annoying that the option of a nvidia chipset with integrated graphics that were better than intel's while being cheaper and lower power than a dedicated graphics chip with it's own memory has been frozen out by this change.
Yes it's annoying that you can no longer use a low end CPU with a high end platform or vice versa like you could in the core 2 days.
But while the former of those is annoying for laptop gamers and the latter is annoying for those whose CPU and IO requirements are mismatched that does NOT mean that the decision to go for greater integration was the wrong one for intel.
in fact the 4x4GB DDR2 is mostly replaced by 6x2GB DDR3 as the "top of the line" at mortal prices. Hmm, afaict only the last generation of core 2 boards supported 4x4GB whereas nearly every i series board i've seen supported that configuration and ddr2 and ddr3 seem similar in price.
Your right though there doesn't seem to be much pressure for more ram from desktop apps. I have 8GB in my desktop at uni and it's more than enough for all the normal desktop stuff (I also use an app that can drive a 48GB box into swapping but I defaintely wouldn't consider such an app typical.
It doesn't guarantee that the driver will run the device properly, but it does guarantee that the driver won't crash the OS. I don't see how thats possible. My understanding of a bus master DMA based system (e.g. pretty much any PCI or PCIe based system) is that part of "running the device" is telling the device where to DMA to.
When devices can easilly crash the system without help from a driver it seems unlikely any driver verifier could prevent a driver intentionally or unintentionally crashing the system.
Dynamic range is the ratio of the smallest signal you can detect (under a given set of settings) and the largest.
If your dynamic range is too small for the scene then no matter what exposure you chose you will lose detail in some parts of the image.
Cameras have a much smaller dynamic range than the human eye so scenes that our eyes deal with no problem can pose a problem for cameras.
High dynamic range imaging gets around this by taking two or more exposures. Longer exposures (or wider apetures or higher sensor sensitivity) for the dark parts of the image and shorter exposures (or equivilent) for the bright parts. Software can then combine these images to produce one high dynamic range image.
For stills of static scenes this is easy however for video (or worse stills of fast moving objects) there is a problem. If you try and use one camera your images with different exposures will be taken at different times. If you try and use two cameras with seperate lenses the different viewpoints will cause an issue. The soloution to this is to use beam splitting optics.
and I think it has to do with the limitations of the final media indeed, a normal monitor has a limited dynamic range. With many modern LCDs each channel is only 6 bit!
So if you want to make both the shadow and highlight detail in a a high dynamic range image visible on a normal monitor you will have to compress the dynamic range down.
so you have to adjust the schedules for all trains on that line. Fast. No you just have to leave the red signal behind the train red until the train leaves the section of track it is on.
Is that locked flight deck door thing that came in after 9-11 US specific or do other countries require it too (afaict ryanair don't fly to the US)
Imagine trying to keep the plane pointed forward and on-speed while checking your map chart, dealing with ATC and the radios, and any number of other little things that come up. Though small planes often don't have either an autopilot or a copilot.
Unlike me that will look how big the numbers are, how big the screen is, how heavy it is, does it support Linux, and how cheap it is.... I also tend to read the service manual (and consider it a major minus point if I can't get the service manual). How easy is it to upgrade the ram? how easy is it to pull the hard drive etc are important to me when buying a laptop.
leaving pretty much no mess. Unfortunately you can often see a section of unworn plastic that sticks out like a sore thumb from the worn plastic surrounding it.
Which is why we need to stop burning it as quickly as possible. Petroleum is such useful stuff, it's just crazy to burn it.
While we do have some control over the process it's not as though petroleum is one compound that we can either make into plastics or make into fuels.
In particular plastics are mainly made from ethene and propene. We get those molecules by splitting bits off the end of long chain alkanes giving us shorter chain alkanes (which are generally more valuable fuels than the long chain ones) along with short alkenes like ethene and propene.
the chances of oil rigs exploding would be lessened by the fact that there would be far less oil rigs in the first place. I have my doubts on this, afaict nuclear is primerally a source of electricity and very little electricty comes from OIL since other fuels are much cheaper.
The other big advantage of torx is it either fits or it doesn't.
There are at least two types of cross head screws common in the west (philpps and pozi) and afaict there is a japanese standard that is different again. Bits from one family will sort of fit in the others and bits the wrong size from the same range will also generall sort of fit.
Screwdrivers that appear to fit but don't fit very well are a BAD thing, they lead to damaged screwheads which in turn can make it virtually impossible to dismantle a device.
From the article you linked "Bungie released the source code of Marathon 2 in 1999 shortly before being acquired by Microsoft, which led to the advent of the Marathon Open Source Project, more commonly known as Aleph One by fans. The project, which is still active as of 2010[update], is committed to adding enhancements to the Marathon 2 engine. In 2002, a project to port Marathon to the Marathon 2 engine, called M1A1, was completed. Several of the game's music tracks have been remixed and enhanced multiple times by different people. In early 2004, Bungie released the entire Marathon Trilogy as freeware, allowing it to be downloaded free of charge. "
If ISPs were just ISPs (as they were back in the dialup days) i'd be inclined to agree with you, people fed up with a sucky ISP could just move to a different one and ISPs would want to offer good service to keep thier customers.
Unfortunately in many places that isn't the case. The company that owns the lines can often either refuse to give access to other ISPs at all or can set up pricing structures to either price other ISPs out of the market completely or at least make it difficult for them to offer a better service than the line owner.
here in the UK local loop unbundling has provided some relief but there has been a worryingly large ammount of consolidation in the LLU market and it wouldn't surprise me if the LLU companies end up merging down into one or two players. Also LLU is really only a short term soloution as it's only practical on legacy copper phone networks. Our cable monopoly is already offering speeds way higher than any company can reach on a single ADSL line and bonded ADSL lines are a prohibitively expensive option (not to mention that most areas simply don't have the phone wires spare to support a significant fraction of customers being on bonded ADSL).
For low power but high transistor (or transistor substitute) count stuff like memory i'm inclined to agree with you.
For processors afaict the limiting factor is more how much power can we get rid of than how many transistors can we pack in.
Also (unless there is a radical change in how we make chips) going 3D is going to be expensive since each layer of tranistors (or transistor substitutes) will require seperate deposition, masking and etching steps.
Slashdot Mirror
IF it's really as low as $250K that doesn't actually sound like too bad a deal. it's "only" $125 per gigabyte.
Comparatively I built a 48GB dual quad box a while ago for about £3000 (including VAT but also including our discounts from the supplier) which at current exchange rates works out to just under $100 per gigabyte. I could have got that price lower with slower processors.
Not as large a difference as I expected really but I still think in a ram dominated virtualisation situation the dual socket boxes are likely to win out.
s/only supported 16 modules/only supported 32 modules/
and the Nehalem-EX with 8 sockets go as high as 2TB of RAM per server.
IIRC in theory you get four memory buffers per processor each with two channels. With two modules per channel, two channels per expander, four expanders and 8 processors you would get 128 modules. With 16GB modules you would indeed have 2TB of ram.
But does anyone actually sell a board that will take that much ram (the only nahelm ex board i've seen for sale was from supermicro and only had four CPUs and one module per channel so only supported 16 modules)? and how much would the bloody thing cost? (last I checked there was a huge premium on cost per gigabyte for 8GB modules, I dread to think what 16GB is like).
And still people want more RAM per box.
Some people do but for many workloads i'd think that beyond a certain point the TCO of adding more boxes would be lower than the TCO of trying to stuff more ram into one box (depending on how much your cost to admin and possiblly license an extra box is of course).
There are at least two seperate issues that need to come together to cause this denial of service flaw.
1: systems that process untrusted data (and yes that means web browsers) need resource limits. If some aspect of loading a page sucks up unacceptablly large ammounts of memory then the browser should stop and give an error. Firefox clearly doesn't.
2: The animiation should not take this much memory to handle. Yes for a short loop of a few frames storing everything uncompressed in memory makes sense. However for long loops and one-shot animations it makes far more sense to work in a streaming manner.
working around an arbitrary limitation in a not-too-far-in-the-future legacy 32-bit platform.
Actually on a 64-bit system this security vulerability is potentially MUCH worse. On a modern 32-bit system it likely just means the browser crashes. On a 64-bit system the exploit file it will push the entire system deep into swap potentially bringing the entire system to it's knees.
Really there is a deeper issue, crashing or eating up insane ammounts of swap is not an appropriate response to an item causing excessive memory use in an animation decoder/displayer whether that is the result of a legitimate codepath (in this case converting an animated gif to a sequance of uncompressed images) or a leak or an exploit. Systems like browsers that are exposed to malicious remote entities need resource limits.
3th party chip sets
Less high power high speed chips means less power spent on interfacing between them and less complex system cooling. For most systems that is probablly worth sacrificing the ability to choose northbridges.
also apple. They can't stay core2 for ever on the mini / some of there laptops and intel video does not fit in there gpu api.
If true that sucks for them but ultimately I don't think in.
It appears (see http://slashdot.org/comments.pl?sid=1786182&cid=33570924 for caveats) that with sandy bridge intel is now beating the best nvidia chipsets for the C2D so it won't be a downgrade on general graphics performance for laptop builders to move from C2D+nvidia chipset to sandy bridge.
With consoles each console has a distinct pool of games. Most games are either released for a single console or for a group of consoles with similar capbilities. Sometimes with a PC release as well but the PC release is often either crap, late or both.
If you want to play recent Mario games you have to buy a Wii (or maybe screw arround with emulation on a PC but most people won't bother to go to that much trouble). If you want to play GTA4 you have to buy a PS3, an xbox 360 or a high end gaming PC. If you want to play recent halo you have to buy an xbox 360. If you want to play recent ratchet and clank you have to buy a PS3.
OTOH PCs all draw from the same pool of games. With a low end PC some games are unplayable and others require you to turn down the settings and/or tolerate slowdowns. With a high end PC you should be able to run all games well (other than driver bugs which can hit both low and high end systems :( ).
My question would be are those wanting a laptop that is too small for a discrete GPU and who care about graphics performance going to be better off staying with a core 2 duo with a nvidia chipset? or will they be better off with sandy bridge?
Unfortunately anandtech ( http://www.anandtech.com/show/3871/the-sandy-bridge-preview-three-wins-in-a-row/7 ) didn't include nvidia integrated graphics in their comparison. Also they were using a desktop not a laptop chip afaict (though they don't seem to know for sure exactly what they had from reading the comments). Integrated graphics performance is far less relevant for desktops than for laptops since it's so easy to add dedicated graphics to a desktop.
Further it seems anandtech can't seem to be consistent in their benchmarking. I found an article where they benchmarked a 320M ( http://www.anandtech.com/show/3762/apples-13inch-macbook-pro-early-2010-reviewed-shaking-the-cpugpu-balance/2 ) but WOW was at lower settings and the other games didn't match up at all. Still told me that sandy bridge was probablly better than the 320M at least for WOW.
Currently with intel stuff (it's a while since i've looked at the AMD side) the laptop and low end desktop platforms have 2 channels and at least with the boards i've seen the max configuration supported is 4x4GB for a total of 16GB.
The current intel high end desktop platform has 3 channels and at least with the boards i've seen the max configuration supported is 4x4GB for a total of 16GB
Workstation/server platforms go much higher, with the right board and a big enough budget you can get 18x8GB (maybe more now if larger modules have appeared) on a dual socket platform and more on the 4/8 socket platform (though all the boards i've seen for that platform have not come even close to maxing out the chipsets memory capability)
All the details i've seen about sandy bridge seem to reffer to a laptop/low end desktop platform. I've seen a mention of a high end variant but no real details of it.
Ultimately for laptops and low end desktops moving all the high speed logic (graphics, CPU, memory controller) into one chip makes a lot of sense from both a cost and a power point of view.
Yes it's annoying that the option of a nvidia chipset with integrated graphics that were better than intel's while being cheaper and lower power than a dedicated graphics chip with it's own memory has been frozen out by this change.
Yes it's annoying that you can no longer use a low end CPU with a high end platform or vice versa like you could in the core 2 days.
But while the former of those is annoying for laptop gamers and the latter is annoying for those whose CPU and IO requirements are mismatched that does NOT mean that the decision to go for greater integration was the wrong one for intel.
in fact the 4x4GB DDR2 is mostly replaced by 6x2GB DDR3 as the "top of the line" at mortal prices.
Hmm, afaict only the last generation of core 2 boards supported 4x4GB whereas nearly every i series board i've seen supported that configuration and ddr2 and ddr3 seem similar in price.
Your right though there doesn't seem to be much pressure for more ram from desktop apps. I have 8GB in my desktop at uni and it's more than enough for all the normal desktop stuff (I also use an app that can drive a 48GB box into swapping but I defaintely wouldn't consider such an app typical.
It doesn't guarantee that the driver will run the device properly, but it does guarantee that the driver won't crash the OS.
I don't see how thats possible. My understanding of a bus master DMA based system (e.g. pretty much any PCI or PCIe based system) is that part of "running the device" is telling the device where to DMA to.
When devices can easilly crash the system without help from a driver it seems unlikely any driver verifier could prevent a driver intentionally or unintentionally crashing the system.
Dynamic range is the ratio of the smallest signal you can detect (under a given set of settings) and the largest.
If your dynamic range is too small for the scene then no matter what exposure you chose you will lose detail in some parts of the image.
Cameras have a much smaller dynamic range than the human eye so scenes that our eyes deal with no problem can pose a problem for cameras.
High dynamic range imaging gets around this by taking two or more exposures. Longer exposures (or wider apetures or higher sensor sensitivity) for the dark parts of the image and shorter exposures (or equivilent) for the bright parts. Software can then combine these images to produce one high dynamic range image.
For stills of static scenes this is easy however for video (or worse stills of fast moving objects) there is a problem. If you try and use one camera your images with different exposures will be taken at different times. If you try and use two cameras with seperate lenses the different viewpoints will cause an issue. The soloution to this is to use beam splitting optics.
and I think it has to do with the limitations of the final media
indeed, a normal monitor has a limited dynamic range. With many modern LCDs each channel is only 6 bit!
So if you want to make both the shadow and highlight detail in a a high dynamic range image visible on a normal monitor you will have to compress the dynamic range down.
so you have to adjust the schedules for all trains on that line. Fast.
No you just have to leave the red signal behind the train red until the train leaves the section of track it is on.
Is that locked flight deck door thing that came in after 9-11 US specific or do other countries require it too (afaict ryanair don't fly to the US)
Imagine trying to keep the plane pointed forward and on-speed while checking your map chart, dealing with ATC and the radios, and any number of other little things that come up.
Though small planes often don't have either an autopilot or a copilot.
Unlike me that will look how big the numbers are, how big the screen is, how heavy it is, does it support Linux, and how cheap it is....
I also tend to read the service manual (and consider it a major minus point if I can't get the service manual). How easy is it to upgrade the ram? how easy is it to pull the hard drive etc are important to me when buying a laptop.
leaving pretty much no mess.
Unfortunately you can often see a section of unworn plastic that sticks out like a sore thumb from the worn plastic surrounding it.
Which is why we need to stop burning it as quickly as possible. Petroleum is such useful stuff, it's just crazy to burn it.
While we do have some control over the process it's not as though petroleum is one compound that we can either make into plastics or make into fuels.
In particular plastics are mainly made from ethene and propene. We get those molecules by splitting bits off the end of long chain alkanes giving us shorter chain alkanes (which are generally more valuable fuels than the long chain ones) along with short alkenes like ethene and propene.
the chances of oil rigs exploding would be lessened by the fact that there would be far less oil rigs in the first place.
I have my doubts on this, afaict nuclear is primerally a source of electricity and very little electricty comes from OIL since other fuels are much cheaper.
The other big advantage of torx is it either fits or it doesn't.
There are at least two types of cross head screws common in the west (philpps and pozi) and afaict there is a japanese standard that is different again. Bits from one family will sort of fit in the others and bits the wrong size from the same range will also generall sort of fit.
Screwdrivers that appear to fit but don't fit very well are a BAD thing, they lead to damaged screwheads which in turn can make it virtually impossible to dismantle a device.
From the article you linked
"Bungie released the source code of Marathon 2 in 1999 shortly before being acquired by Microsoft, which led to the advent of the Marathon Open Source Project, more commonly known as Aleph One by fans. The project, which is still active as of 2010[update], is committed to adding enhancements to the Marathon 2 engine. In 2002, a project to port Marathon to the Marathon 2 engine, called M1A1, was completed. Several of the game's music tracks have been remixed and enhanced multiple times by different people. In early 2004, Bungie released the entire Marathon Trilogy as freeware, allowing it to be downloaded free of charge. "
So it seems that one has already been revived :)
If ISPs were just ISPs (as they were back in the dialup days) i'd be inclined to agree with you, people fed up with a sucky ISP could just move to a different one and ISPs would want to offer good service to keep thier customers.
Unfortunately in many places that isn't the case. The company that owns the lines can often either refuse to give access to other ISPs at all or can set up pricing structures to either price other ISPs out of the market completely or at least make it difficult for them to offer a better service than the line owner.
here in the UK local loop unbundling has provided some relief but there has been a worryingly large ammount of consolidation in the LLU market and it wouldn't surprise me if the LLU companies end up merging down into one or two players. Also LLU is really only a short term soloution as it's only practical on legacy copper phone networks. Our cable monopoly is already offering speeds way higher than any company can reach on a single ADSL line and bonded ADSL lines are a prohibitively expensive option (not to mention that most areas simply don't have the phone wires spare to support a significant fraction of customers being on bonded ADSL).
For low power but high transistor (or transistor substitute) count stuff like memory i'm inclined to agree with you.
For processors afaict the limiting factor is more how much power can we get rid of than how many transistors can we pack in.
Also (unless there is a radical change in how we make chips) going 3D is going to be expensive since each layer of tranistors (or transistor substitutes) will require seperate deposition, masking and etching steps.