In a way, they've had it. In the form of the capsule approach. A good example is the Soyuz re-entry. As much of a debacle it was, they survived.
*If* a re-usable spacecraft design proves to be useful, the crew cabin may benefit from being a capsule that could somehow explosively decouple in an incident. Of course, chances are that it's just not worth it to reuse. Even if you had such a capsule, reliable separation such that a chute could be deployed successfully would be dangerous. Add to that the cost of preparing a shuttle for re-use, *and* the limted number of uses even with those measures.
When was the last time you tried them? I have an R500 part and before that an R100 part, as well as two nVidia systems. A while ago, the difference was night and day, nVidia's drivers were much more reliable and featureful.
Over the course of 2008, that's changed for me. AMD has caught up. Meanwhile, I've started using compiz, and the nVidia systems with current drivers still corrupt the window decorations and contents when I have too many windows open. My ATI doesn't suffer from that.
nVidida does have something on flexible video decode offload and AMD is only promising something, but as it stands its horribly fragmented. nVidia has their implementation, AMD promises another incompatible one, Intel has yet another incompatible one, and all the while Xorg guys muse about a fourth strategy.
I think you don't have much luck getting them to open source notes/domino. What I wouldn't mind is for the network communication protocol to be opened up. Enable the community to at least build a client or support for Domino mail servers instead of Notes. I don't think Notes itself is a big revenue generator, but rather is there to enable Domino and domino client fees to be sold. Of course, some of the 'features' of notes require it to be a pain in the ass (preventing recipient from forwarding, copying and pasting). I think those features are misguided (a la DRM), but IBM uses them as marketing points and an open ecosystem would handily defeat those.
Selfish reason: I'm forced to use notes and want someone to build a better client for email/calendar aspect of it. When I was in charge of a Domino install, the first thing I did was enable IMAP for 99% of the email and kept Notes/Designer around for agent development and other such things. People were ecstatic to have the choice. I would suspect greater interoperability would make the Domino pill easier to swallow for organizations who have had bad experiences with notes. I had no significant gripes about Domino, just Notes.
Notes' calendar and email is horrible, but I will say a competent domino server can provide a development platform for internal applications akin to web applications that even morons can create. Web applications I whole-heartedly prefer when done well, but the skill and effort to create them is still leaps and bounds more than doing the same stuff in a notes database.
Though MS keeps teasing it, I doubt they'll actually jump to a redesigned platform. It's just too much investment and risk, given a low chance of meaningful improvement. The major flaws with Windows aren't at the lowest levels, they lie more in high-level organization and third-party activity.
In terms of manageability, to what extent has your professional experience been with at scale linux deployments? I would have said the opposite, that Linux was much easier at scale. For obscure things that MS doesn't wrap for you, MS is a pain, and Linux and others have a richer scripting environment to do those. Powershell has advanced the state of things, but that is hooribly slow and a pain. In terms of GUIs/Directory/policies, Novell may have the most comparable stack with their ZENworks stuff in SuSE. Knowing that AD was the first even moderate competitor to Novell's coveted directory management translates well to their work in SuSE.
The simple fact of the matter is that by and large, a person will learn one primary way of doing it, and be ill or misinformed about the alternatives.
MS's venture into the PDA/Smartphone realm has been problematic. It seemingly remains low on their list of priorities. A WinMo phone currently implies a further investment in third-party commercial applications to actually get suitable experience. Meanwhile, Apple and Google are getting a number of things right. Apple's out of the box experience is usable for most, and the App store is a much more well organized approach to third-party applications. Android is similar, but with the added benefit of the lowest barrier to entry for development and plenty of free apps in the 'Market'. I might have to give Google the long-term market edge in this market. We have a market that is still nearly a clean slate, fundamentally distinct from the ecosystem that MS flourished in (move the cost from developers, resellers, and users to advertisers given the ubiquitous internet access that wasn't possilbe as MS formed), and could become many people's dominant method of interaction with the internet. This would be akin to the mainframe to PC revolution you mentioned.
In terms of laziness, that also backfires too. Maybe not in marketshare, but in revenue. For example, most still run XP that was licensed long ago. Those users are not a revenue stream for MS. In terms of Linux defualt being shipped on systems, the eeepc was the best chance to date at evaluating it, and both sides of the debate find the statistics they want somehow, making it still inconclusive. Maybe Dell has some telling statistics, but if MS OS was an expensive addon to a computer, would people just get the cheap-o option?
-In-house apps: I've seen a trend of web applications becoming the norm for many companies' custom applications. They already invest in web applications to lower maintenance/deployment costs. Not necessarily for platform flexibility, but it happens to help. That's not to say they are non-existant, but there certainly has been a push in this way. -Training new OS. Honestly, I don't see this as big of a deal as MS would like people to believe. The reality is companies don't have to explicitly train today and if the mass market moved, then corporations would have it as moot of a point as it is today. Of course, the distinctions between platforms are not as hard to grasp, and even MS proves users can adapt to changes and shuffling, as they do it in every major Windows release. -New office suite. Again, the amount of explicit training is small. And, frankly, Moving from Office 2k3 to 2k7 is more jarring than 2k3 to OpenOffice.org. Not saying that both are equally capable, but retraining isn't a bigger deal than MS induces all on its own. Somehow, this doesn't induce tons of panic when MS does it, wo I don't see why it would otherwise. -Critical applications, again, most applications are similar in a task. In terms of in-house applications, corporations are already proving willingness to migrate to reduce IT costs to webapps. And again, this underestimates the ability of people to adapt to relatively mild change.
Now in in IT area, there is admittedly some concerns. IT people inherently dig beneath the surface and are more impacted. Microsoft-only IT personnel may be a deal-breaker, but a large percentage of people in the field have useful Linux experience. The rest may pose a particular challenge, as a non-trivial amount of MCSE training is indoctrination, evoking personal resistance to such change.
Their approach of compiling both ('Universal Binaries') conveniently in their SDK really smoothed the transition from PPC to x86. And I haven't seen MS do something analogous, but it is a tad harder to do without re-doing their executable format or application packaging strategies. It they add a third architecture (x86_64), it certainly would be slicker. I will say we can't call in advance what they will do.
Single install media for both editions and taking that page from Apple's playbook would be interesting, but undoubtedly at least the latter is patent encumbered.
There is application and driver compatibility. Their track record with applications has been relatively decent (The notable exception was migrating Win9x market to NT kernel, where a large number of things didn't work right). Drivers they break relatively commonly.
Essentially, I would give them about equal score on compatibility since XP release that I would give glibc ("Linux" would be a misnomer, the kernel strictly speaking hasn't been perfectly consistant on syscalls and glibc has abstracted it away on occasion, and "Linux" as a term for an OS includes things like Python, that have *not* maintained good compatibility, regardless of running on Linux or Windows). Linux inherently breaks binary drivers and frequently the API to compile against, just like Windows.
Depends, if the DDR3 manufacturers ramp up, it could come down to earth. Either way, I wouldn't count on 4GB DIMMs.
On nVidia, I am using it, but my AMD system is now behaving sane (the binary drivers aren't terrible anymore, the open drivers for R500 are 'ok' for compiz and the like. The nVidia drivers on the other hand seem to have been causing me problems (compiz title bar and window corruption at scale I don't see with the ATI/AMD drivers, some weird hiccups on 3D apps).
I'm no fan of MS, but what exactly do you propose they do? They offer 64-bit variants that can run 64-bit applications of their supported platforms. They provide the platform to allow this specific thing. They provide the tools to develop for this.
What you have is commercial application providers flat-out ignoring 64-bit capability, as it is easier to target the 32-bit subset that works both on Pentium 4 and such and new. You have to make the vendors release 64-bit enabled builds. Linux suffers from this as well, to a lesser extent. In the OSS world, they rebuilt 64-bit readily. However, Acrobat, Flash, Sun JRE all took a long time or are still taking time to completely support 64-bit. The commercial world just has a hard time justifying bothering where there is backwards compatibility and 99.9% of their usage won't exceed the limit per process restrictions.
MS could have not published any 32-bit platform to accelerate ubiquity. Imagine the backlash at not supporting Core and Pentium 4, requiring those users to go to Core2 or Athlon64. Even then, it wouldn't have alleviated the issue as these vendors would still want to sell to XP users. MS could have omitted 32-bit compatibility, completely shooting backwards compatibility in the foot.
So while I'm not crazy about Windows, their x86_64 bit strategy is not any worse than other platforms, it's the commercial third-parties that cause your grief.
Don't know about affordably getting to 16GB in a single socket board. Most boards in that space have 4 dimm slots and 4 GB dimms are relatively expensive. I don't know about the shuttle cases, I don't require that small a packaging and I have historically found larger cases more convenient for maintenance and thermal management (and expense of repairs, blowing a power supply in a weird form factor sucks). If you get AMD graphics, you may have some 3D capability in BSD (nVidia too maybem but options are more open ended in AMD currently). That said, onboard ATI graphics should be more than enough for light 3d stuff.
I'm personally waiting for Phenom II DDR3 parts to come and evaluating a purchase. I'm dubious of the cost of DDR3, but wanting to see what the market does as Core i7 and AMD both ramp up to using it commonly.
-If you are right, buy a card, disable the onboard, and don't whine. Ignore it. And what the hell is your experence? Onboard video can actually be respectable, more so than you give it credit.
That aside: -Compiz/Aero/whatever indeed use GPU, but even a puny one can handle the load. It has to support certain featuresets, but it can be weak. Current Intel, nVidia, and AMD integrated graphics are well up to the task, though nVidia compiz has memory issues still. In terms of sharing your memory, DDR2 sticks are cheap for one, and for another, basic desktop functions don't mandate compiz/aero. You may disable the feature -I can't speak to CS4, but I can't help to wonder if you are basing your experience on a really old integrated part, or Unicrome or similar. I suspect 780V/G would perform surprisingly well. -For engineering apps, by all means buy a card and disable the onboard. -For development, if your specific target environment will allow for it, by all means add the card.
For all you say we need GPUs, you seem not to realize the onboard parts are still GPUs. I would make the opposite argument, that in 2001 you pretty much required a discrete video card as no good vendors took that seriously and the tech was not advanced, but things have changed a lot. Some current onboard parts could trounce the high-end parts of the GeForce4 days.
BTW, my HTPC has an integrated GeForce 6150. It's no screamer, but it is surprisingly capable. And that is a relic by today's standards.
Though still not perfect, you'd have to compare Intel TDP numbers to AMD 'ACP' numbers. Intel screwed with the point of TDP for marketing purposes and they have the leverage with OEMs to pretty much do as they damn well please, even if it makes OEM life harder.
Secondly, the Core2 isn't comparable in featureset. Notably, it requires a memory controller on motherboard which tends to have a significant amount of power in and of itself. So you'll have to jump to Core i7 parts, which currently advertise a 130W TDP, presumably largely in part due to bringing the controller in.
Thirdly, the current gen phenoms even include 65W parts at reduced clock. I would not doubt that, if not at launch, shortly thereafter lower-wattage parts will come out. Even if not, underclocking may be an option.
All said and done, AMD has something kind of compelling in various ways. I don't think it will take back the performance lead, but Intel is taking some hits in trying to get an equivalent architecture together. Will have to wait for AM3 to have complete apples-to-apples benchmarks for i7 and Phenom II, but I don't think it will be that much closer than it is today.
In fact, we deal with both AMD and Intel servers, and under benchmarking conditions, we allegedly make the Intel parts exceed TDP.
The problem is that TDP became a marketing point and thus Intel abandoned the original intent. It was supposed to allow systems vendors to plan how many CFM and how good a heat sink would be needed to dissipate all the heat in a worst case scenario. To AMD's credit, they have a metric they call 'ACP' to indicate their analogous figure to Intel's TDP. I like that better in theory than just abandoning TDP.
That said, for Intel and AMD, I wonder what they consider 'typical' loads. In the end, a breakdown of what clock speeds and what percentage of time is in C1 and such would be appreciated to know what to expect. Knowing what the CPU uses at each clock, with about 0% of time in sleep states and near 100% would let me compare a bit better than a single oversimplified number.
I'm sure Motorola isn't doing so, but I wouldn't mind a non-compete so long as the company demanding that would pay full salary equivalent during the interval they effectively demand I stay unemployed. Simple as that. Would also be fine with a severance package of exactly equivalent value (so long as the interval didn't span more than one tax year, in which case the payment would have to be spread to avoid unfair taxation).
Anyway, I know companies aren't up for it. The most generous severance package I got was three months pay on top of untaken vacation, and that was without any non-compete criteria attached.
Putting aside debate on whether government has integrity, we do know private industry tends to not have integrity. Integrity in and of itself isn't a profit driving concept. Vision can be rewarding, but not always.
In terms on current failings of private industry and the internet, no one will provide significant throughput to some people I know. They can get electricity and phone (and dial-up) because of government interference, but the CO is too far for DSL and cable companies aren't going to bother with this area. The cellular providers that in theory could provide coverage won't do so without limits that make it impractical.
Now this isn't exactly life and death, but it does preclude them from participating in a number of internet based industries, as consumers and as potential entrepreneurs. If fast internet connectivity is a significant prerequisite for enabling more of the constituents to participate and compete in the global market, I could see it as being worth taxpayer dollar to some extent given the precedent of the Interstate system. The 2009 budget for that seems to be about 35 billion for maintenance and development. 44 billion depending on the scope and duration of the spending may be a realistic number.
Too many good companies attract sharks known as shareholders and executives that just want to pimp out the name, and not maintain efforts to lead their original advances. As you say, Cisco is letting their networking slide as they chase whatever flavor of the month executives want to slap the name onto. Guess this is good in a way, provides a bizarre, but eerily ubiquitous mechanism of preventing monopolies, self-destruct through greed.
If lucky, they will be like other companies that at least come to their senses once in a while (i.e. like HP did after the exit of Carly). But right now and for a significant amount of time they've let their leadership slip. Even if I like their firmware features, I can't exactly ignore the hardware advances the competition makes...
All of them have integrated switches (the Dell M-series, IBM from its inception). Hell, Cisco even makes a switch to put into the IBM product, probably the others too. Will note that none put them into the backplane, but in modular form factors, as it should be. Backplane should be as simple and reliable as can be (as passive as possible, redundant power and data traces, etc etc.)
They run their FC business much like their ethernet business;) If you want high performance, Cisco is definitely not the brand even in ethernet. In ethernet, at least, they live on the strength of their manageability, not much more.
I know Cisco has some ethernet switches than can handle line rate on all their ports, but it's more rare than it should be. For example, I don't think any vendor other than Cisco has an entry level 48 port gigabit switch that doesn't have the fabric to handle it all concurrently (I speak of the 2960 specifically). I know a lot of places don't need that performance, but it is an odd omission given the current state of competitor technology.
I will add that I'm not impressed with their hardware, but am impressed with their *firmare*. In terms of price performance, speed of switching, density, etc etc, they have competitors that beat them handily, without compromising fundamental reliability.
Now in terms of managing complex networks and chasing problems down, I don't know of a competitor that provides compelling features in the firmware to detect and locate all the various conditions I've seen Cisco make easy. It's not fundamental to the hardware, all the required data to identify this is by its nature is resident in most all ethernet switches, but it is the firmware that does the requisite analysis and identification of troublesome situations.
It is a blade server offering. Just because it will be sold as a full stack with bundles ESX and other prescribed software doesn't mean the technology isn't at its core another server offering, no more than a processor ceases to become a processor because it is sold only in a whole system by anyone. This is important in evaluating the fundamental ability of others to compete.
In terms of "changing the game in ways others can't compete", it sounds like some great pep rally morale speak, but in the end, full stack solutions for various intents can be seen across the vendors, as it seems to be all the rage among them now. I personally think the competency is useful, but pre-done vendor solutions will fall short of a company's needs compared with a customized one done by the company itself (or at least partner with a vendor rather than just buy it). Of course, in the commodity market, that fat profit margin in full stacks is appealing to the companies, hence the interest.
In any event, it will be interesting to see if anything particularly keen comes out of it. However, I don't expect much out of the servers just like I wouldn't expect much out of managed switching from a Dell badged switch. As a company, I would stick to having your own architects listening to the vendors and picking the most appropriate parts from each. There may be a simplicity of support, but in aggregate you should be able to ensure enterprise grade support even if you have something more tailored to your needs.
Slashdot Mirror
In a way, they've had it. In the form of the capsule approach. A good example is the Soyuz re-entry. As much of a debacle it was, they survived.
*If* a re-usable spacecraft design proves to be useful, the crew cabin may benefit from being a capsule that could somehow explosively decouple in an incident. Of course, chances are that it's just not worth it to reuse. Even if you had such a capsule, reliable separation such that a chute could be deployed successfully would be dangerous. Add to that the cost of preparing a shuttle for re-use, *and* the limted number of uses even with those measures.
When was the last time you tried them? I have an R500 part and before that an R100 part, as well as two nVidia systems. A while ago, the difference was night and day, nVidia's drivers were much more reliable and featureful.
Over the course of 2008, that's changed for me. AMD has caught up. Meanwhile, I've started using compiz, and the nVidia systems with current drivers still corrupt the window decorations and contents when I have too many windows open. My ATI doesn't suffer from that.
nVidida does have something on flexible video decode offload and AMD is only promising something, but as it stands its horribly fragmented. nVidia has their implementation, AMD promises another incompatible one, Intel has yet another incompatible one, and all the while Xorg guys muse about a fourth strategy.
There is a difference between an unexpected surprise (which many users are trained to automatically suspect malware) and an announced change.
I think you don't have much luck getting them to open source notes/domino. What I wouldn't mind is for the network communication protocol to be opened up. Enable the community to at least build a client or support for Domino mail servers instead of Notes. I don't think Notes itself is a big revenue generator, but rather is there to enable Domino and domino client fees to be sold. Of course, some of the 'features' of notes require it to be a pain in the ass (preventing recipient from forwarding, copying and pasting). I think those features are misguided (a la DRM), but IBM uses them as marketing points and an open ecosystem would handily defeat those.
Selfish reason: I'm forced to use notes and want someone to build a better client for email/calendar aspect of it. When I was in charge of a Domino install, the first thing I did was enable IMAP for 99% of the email and kept Notes/Designer around for agent development and other such things. People were ecstatic to have the choice. I would suspect greater interoperability would make the Domino pill easier to swallow for organizations who have had bad experiences with notes. I had no significant gripes about Domino, just Notes.
Notes' calendar and email is horrible, but I will say a competent domino server can provide a development platform for internal applications akin to web applications that even morons can create. Web applications I whole-heartedly prefer when done well, but the skill and effort to create them is still leaps and bounds more than doing the same stuff in a notes database.
Though MS keeps teasing it, I doubt they'll actually jump to a redesigned platform. It's just too much investment and risk, given a low chance of meaningful improvement. The major flaws with Windows aren't at the lowest levels, they lie more in high-level organization and third-party activity.
In terms of manageability, to what extent has your professional experience been with at scale linux deployments? I would have said the opposite, that Linux was much easier at scale. For obscure things that MS doesn't wrap for you, MS is a pain, and Linux and others have a richer scripting environment to do those. Powershell has advanced the state of things, but that is hooribly slow and a pain. In terms of GUIs/Directory/policies, Novell may have the most comparable stack with their ZENworks stuff in SuSE. Knowing that AD was the first even moderate competitor to Novell's coveted directory management translates well to their work in SuSE.
The simple fact of the matter is that by and large, a person will learn one primary way of doing it, and be ill or misinformed about the alternatives.
MS's venture into the PDA/Smartphone realm has been problematic. It seemingly remains low on their list of priorities. A WinMo phone currently implies a further investment in third-party commercial applications to actually get suitable experience. Meanwhile, Apple and Google are getting a number of things right. Apple's out of the box experience is usable for most, and the App store is a much more well organized approach to third-party applications. Android is similar, but with the added benefit of the lowest barrier to entry for development and plenty of free apps in the 'Market'. I might have to give Google the long-term market edge in this market. We have a market that is still nearly a clean slate, fundamentally distinct from the ecosystem that MS flourished in (move the cost from developers, resellers, and users to advertisers given the ubiquitous internet access that wasn't possilbe as MS formed), and could become many people's dominant method of interaction with the internet. This would be akin to the mainframe to PC revolution you mentioned.
In terms of laziness, that also backfires too. Maybe not in marketshare, but in revenue. For example, most still run XP that was licensed long ago. Those users are not a revenue stream for MS. In terms of Linux defualt being shipped on systems, the eeepc was the best chance to date at evaluating it, and both sides of the debate find the statistics they want somehow, making it still inconclusive. Maybe Dell has some telling statistics, but if MS OS was an expensive addon to a computer, would people just get the cheap-o option?
It all depends...
-In-house apps: I've seen a trend of web applications becoming the norm for many companies' custom applications. They already invest in web applications to lower maintenance/deployment costs. Not necessarily for platform flexibility, but it happens to help. That's not to say they are non-existant, but there certainly has been a push in this way.
-Training new OS. Honestly, I don't see this as big of a deal as MS would like people to believe. The reality is companies don't have to explicitly train today and if the mass market moved, then corporations would have it as moot of a point as it is today. Of course, the distinctions between platforms are not as hard to grasp, and even MS proves users can adapt to changes and shuffling, as they do it in every major Windows release.
-New office suite. Again, the amount of explicit training is small. And, frankly, Moving from Office 2k3 to 2k7 is more jarring than 2k3 to OpenOffice.org. Not saying that both are equally capable, but retraining isn't a bigger deal than MS induces all on its own. Somehow, this doesn't induce tons of panic when MS does it, wo I don't see why it would otherwise.
-Critical applications, again, most applications are similar in a task. In terms of in-house applications, corporations are already proving willingness to migrate to reduce IT costs to webapps. And again, this underestimates the ability of people to adapt to relatively mild change.
Now in in IT area, there is admittedly some concerns. IT people inherently dig beneath the surface and are more impacted. Microsoft-only IT personnel may be a deal-breaker, but a large percentage of people in the field have useful Linux experience. The rest may pose a particular challenge, as a non-trivial amount of MCSE training is indoctrination, evoking personal resistance to such change.
Their approach of compiling both ('Universal Binaries') conveniently in their SDK really smoothed the transition from PPC to x86. And I haven't seen MS do something analogous, but it is a tad harder to do without re-doing their executable format or application packaging strategies. It they add a third architecture (x86_64), it certainly would be slicker. I will say we can't call in advance what they will do.
Single install media for both editions and taking that page from Apple's playbook would be interesting, but undoubtedly at least the latter is patent encumbered.
There is application and driver compatibility. Their track record with applications has been relatively decent (The notable exception was migrating Win9x market to NT kernel, where a large number of things didn't work right). Drivers they break relatively commonly.
Essentially, I would give them about equal score on compatibility since XP release that I would give glibc ("Linux" would be a misnomer, the kernel strictly speaking hasn't been perfectly consistant on syscalls and glibc has abstracted it away on occasion, and "Linux" as a term for an OS includes things like Python, that have *not* maintained good compatibility, regardless of running on Linux or Windows). Linux inherently breaks binary drivers and frequently the API to compile against, just like Windows.
Depends, if the DDR3 manufacturers ramp up, it could come down to earth. Either way, I wouldn't count on 4GB DIMMs.
On nVidia, I am using it, but my AMD system is now behaving sane (the binary drivers aren't terrible anymore, the open drivers for R500 are 'ok' for compiz and the like. The nVidia drivers on the other hand seem to have been causing me problems (compiz title bar and window corruption at scale I don't see with the ATI/AMD drivers, some weird hiccups on 3D apps).
I'm no fan of MS, but what exactly do you propose they do? They offer 64-bit variants that can run 64-bit applications of their supported platforms. They provide the platform to allow this specific thing. They provide the tools to develop for this.
What you have is commercial application providers flat-out ignoring 64-bit capability, as it is easier to target the 32-bit subset that works both on Pentium 4 and such and new. You have to make the vendors release 64-bit enabled builds. Linux suffers from this as well, to a lesser extent. In the OSS world, they rebuilt 64-bit readily. However, Acrobat, Flash, Sun JRE all took a long time or are still taking time to completely support 64-bit. The commercial world just has a hard time justifying bothering where there is backwards compatibility and 99.9% of their usage won't exceed the limit per process restrictions.
MS could have not published any 32-bit platform to accelerate ubiquity. Imagine the backlash at not supporting Core and Pentium 4, requiring those users to go to Core2 or Athlon64. Even then, it wouldn't have alleviated the issue as these vendors would still want to sell to XP users. MS could have omitted 32-bit compatibility, completely shooting backwards compatibility in the foot.
So while I'm not crazy about Windows, their x86_64 bit strategy is not any worse than other platforms, it's the commercial third-parties that cause your grief.
Don't know about affordably getting to 16GB in a single socket board. Most boards in that space have 4 dimm slots and 4 GB dimms are relatively expensive. I don't know about the shuttle cases, I don't require that small a packaging and I have historically found larger cases more convenient for maintenance and thermal management (and expense of repairs, blowing a power supply in a weird form factor sucks). If you get AMD graphics, you may have some 3D capability in BSD (nVidia too maybem but options are more open ended in AMD currently). That said, onboard ATI graphics should be more than enough for light 3d stuff.
I'm personally waiting for Phenom II DDR3 parts to come and evaluating a purchase. I'm dubious of the cost of DDR3, but wanting to see what the market does as Core i7 and AMD both ramp up to using it commonly.
-If you are right, buy a card, disable the onboard, and don't whine. Ignore it. And what the hell is your experence? Onboard video can actually be respectable, more so than you give it credit.
That aside:
-Compiz/Aero/whatever indeed use GPU, but even a puny one can handle the load. It has to support certain featuresets, but it can be weak. Current Intel, nVidia, and AMD integrated graphics are well up to the task, though nVidia compiz has memory issues still. In terms of sharing your memory, DDR2 sticks are cheap for one, and for another, basic desktop functions don't mandate compiz/aero. You may disable the feature
-I can't speak to CS4, but I can't help to wonder if you are basing your experience on a really old integrated part, or Unicrome or similar. I suspect 780V/G would perform surprisingly well.
-For engineering apps, by all means buy a card and disable the onboard.
-For development, if your specific target environment will allow for it, by all means add the card.
For all you say we need GPUs, you seem not to realize the onboard parts are still GPUs. I would make the opposite argument, that in 2001 you pretty much required a discrete video card as no good vendors took that seriously and the tech was not advanced, but things have changed a lot. Some current onboard parts could trounce the high-end parts of the GeForce4 days.
BTW, my HTPC has an integrated GeForce 6150. It's no screamer, but it is surprisingly capable. And that is a relic by today's standards.
Though still not perfect, you'd have to compare Intel TDP numbers to AMD 'ACP' numbers. Intel screwed with the point of TDP for marketing purposes and they have the leverage with OEMs to pretty much do as they damn well please, even if it makes OEM life harder.
Secondly, the Core2 isn't comparable in featureset. Notably, it requires a memory controller on motherboard which tends to have a significant amount of power in and of itself. So you'll have to jump to Core i7 parts, which currently advertise a 130W TDP, presumably largely in part due to bringing the controller in.
Thirdly, the current gen phenoms even include 65W parts at reduced clock. I would not doubt that, if not at launch, shortly thereafter lower-wattage parts will come out. Even if not, underclocking may be an option.
All said and done, AMD has something kind of compelling in various ways. I don't think it will take back the performance lead, but Intel is taking some hits in trying to get an equivalent architecture together. Will have to wait for AM3 to have complete apples-to-apples benchmarks for i7 and Phenom II, but I don't think it will be that much closer than it is today.
In fact, we deal with both AMD and Intel servers, and under benchmarking conditions, we allegedly make the Intel parts exceed TDP.
The problem is that TDP became a marketing point and thus Intel abandoned the original intent. It was supposed to allow systems vendors to plan how many CFM and how good a heat sink would be needed to dissipate all the heat in a worst case scenario. To AMD's credit, they have a metric they call 'ACP' to indicate their analogous figure to Intel's TDP. I like that better in theory than just abandoning TDP.
That said, for Intel and AMD, I wonder what they consider 'typical' loads. In the end, a breakdown of what clock speeds and what percentage of time is in C1 and such would be appreciated to know what to expect. Knowing what the CPU uses at each clock, with about 0% of time in sleep states and near 100% would let me compare a bit better than a single oversimplified number.
I'm sure Motorola isn't doing so, but I wouldn't mind a non-compete so long as the company demanding that would pay full salary equivalent during the interval they effectively demand I stay unemployed. Simple as that. Would also be fine with a severance package of exactly equivalent value (so long as the interval didn't span more than one tax year, in which case the payment would have to be spread to avoid unfair taxation).
Anyway, I know companies aren't up for it. The most generous severance package I got was three months pay on top of untaken vacation, and that was without any non-compete criteria attached.
Putting aside debate on whether government has integrity, we do know private industry tends to not have integrity. Integrity in and of itself isn't a profit driving concept. Vision can be rewarding, but not always.
In terms on current failings of private industry and the internet, no one will provide significant throughput to some people I know. They can get electricity and phone (and dial-up) because of government interference, but the CO is too far for DSL and cable companies aren't going to bother with this area. The cellular providers that in theory could provide coverage won't do so without limits that make it impractical.
Now this isn't exactly life and death, but it does preclude them from participating in a number of internet based industries, as consumers and as potential entrepreneurs. If fast internet connectivity is a significant prerequisite for enabling more of the constituents to participate and compete in the global market, I could see it as being worth taxpayer dollar to some extent given the precedent of the Interstate system. The 2009 budget for that seems to be about 35 billion for maintenance and development. 44 billion depending on the scope and duration of the spending may be a realistic number.
Is your point that broandband is fundamentally distinct in this regard from interstates, or that the interstate system is unconstitutional?
Too many good companies attract sharks known as shareholders and executives that just want to pimp out the name, and not maintain efforts to lead their original advances. As you say, Cisco is letting their networking slide as they chase whatever flavor of the month executives want to slap the name onto. Guess this is good in a way, provides a bizarre, but eerily ubiquitous mechanism of preventing monopolies, self-destruct through greed.
If lucky, they will be like other companies that at least come to their senses once in a while (i.e. like HP did after the exit of Carly). But right now and for a significant amount of time they've let their leadership slip. Even if I like their firmware features, I can't exactly ignore the hardware advances the competition makes...
In this play, the x86 servers are a necessary evil to push the fat profit margin product, the 'solution', including fat software margins.
All of them have integrated switches (the Dell M-series, IBM from its inception). Hell, Cisco even makes a switch to put into the IBM product, probably the others too. Will note that none put them into the backplane, but in modular form factors, as it should be. Backplane should be as simple and reliable as can be (as passive as possible, redundant power and data traces, etc etc.)
They run their FC business much like their ethernet business ;) If you want high performance, Cisco is definitely not the brand even in ethernet. In ethernet, at least, they live on the strength of their manageability, not much more.
I know Cisco has some ethernet switches than can handle line rate on all their ports, but it's more rare than it should be. For example, I don't think any vendor other than Cisco has an entry level 48 port gigabit switch that doesn't have the fabric to handle it all concurrently (I speak of the 2960 specifically). I know a lot of places don't need that performance, but it is an odd omission given the current state of competitor technology.
They would rebadge whatever Dell is rebadging, skip the middle man.
I will add that I'm not impressed with their hardware, but am impressed with their *firmare*. In terms of price performance, speed of switching, density, etc etc, they have competitors that beat them handily, without compromising fundamental reliability.
Now in terms of managing complex networks and chasing problems down, I don't know of a competitor that provides compelling features in the firmware to detect and locate all the various conditions I've seen Cisco make easy. It's not fundamental to the hardware, all the required data to identify this is by its nature is resident in most all ethernet switches, but it is the firmware that does the requisite analysis and identification of troublesome situations.
It is a blade server offering. Just because it will be sold as a full stack with bundles ESX and other prescribed software doesn't mean the technology isn't at its core another server offering, no more than a processor ceases to become a processor because it is sold only in a whole system by anyone. This is important in evaluating the fundamental ability of others to compete.
In terms of "changing the game in ways others can't compete", it sounds like some great pep rally morale speak, but in the end, full stack solutions for various intents can be seen across the vendors, as it seems to be all the rage among them now. I personally think the competency is useful, but pre-done vendor solutions will fall short of a company's needs compared with a customized one done by the company itself (or at least partner with a vendor rather than just buy it). Of course, in the commodity market, that fat profit margin in full stacks is appealing to the companies, hence the interest.
In any event, it will be interesting to see if anything particularly keen comes out of it. However, I don't expect much out of the servers just like I wouldn't expect much out of managed switching from a Dell badged switch. As a company, I would stick to having your own architects listening to the vendors and picking the most appropriate parts from each. There may be a simplicity of support, but in aggregate you should be able to ensure enterprise grade support even if you have something more tailored to your needs.