I know PT does not mean only subs... but this is the only way to go when there is no open land for rails and the lines have to cross a major waterway: doing light rails would require mass expropriation for the access ramps since the rails would have to be high enough to clear cargo ships. With the vacancy rates sometimes dipping below 2% on the island and surroundings, there is no need to motivate people into moving there more than they already are when housing availability and rent are major concerns... though vacancy was up to 6% this year thanks to rent-induced mass exodus and many completed or on-going residential projects.
With nearly no land available for new developments, there is not much economic advantage to be gained from putting bilions into improving public transportation... it would only make it more accessible and convenient with only marginal economic gains on a multi-bilion expense. I bet the largest gain would end up being on property taxes... and maybe beefier supplemental vehicle registration fee for the privilege of owning a car when PT is available around the primary residence.
Every politician who tells people to use PT should have his limo service cancelled, his expense account frozen, his license suspended, his car(s) sent to storage for a year and be provided with a year's worth of country-wide bus/train/sub/etc. passes. In many cities, it would permanently shut them up until someone fixes it - or at least attempts to.
PT is like Netburst and HT: they are fundamentally good ideas but are often poorly implemented.
RAM latency is not much of an issue for proper multi-threading CPUs... having one thread stalled due to IO and instruction dependencies simply leaves more execution ressources to the other threads and as long as at least two out of four or more threads have ready-and-able instructions, the execution unit can operate near full-load most of the time and completely hide latencies as far as sustained throughput is concerned. Latency is not as much of a killer now as it was in the '80s, thanks in large part to on-chip caches.
As for the vector registers, we have small-scale but otherwise similar stuff now with SSE/AltiVec/etc. but they fall tragically short from the sort of sustained throughput and efficiency dedicated hardware can achieve. Commodity chips are inexpensive and CPUs have enough execution resources to do this stuff sufficiently fast as a loop, they are also simple to program, deploy, service and replace. In the '80s, only companies could afford anything much beyond a C64, there was no significant processing power below several thousand dollars.
For the return to architecture, architecture never really disappeared. Systems that are primarily IO-bound never stopped requiring massive architecture design efforts. Since HPC is highly latency-intolerant, you want to have as close to 0% node-to-node communications as possible: 0.1% processing spent on communications translates into a theoretical maximum of 1000X performance gain given infinite CPUs or ~641X given 1024 CPUs. Come 2008 and quad-core 16/32-ways glueless MP CPUs, we will probably have reached the limits of what can be practically packed in a single system - each CPU will have enough processing power to saturate its local RAM and IO bus. The only ways to work around this is system-scale integration and ASIC implementations - integrate data producers and consumers to reduce reliance on external bandwidth-limited busses.
Around 2010, the computing world will stop until the next breakthrough, assuming it (either the breakthrough or stop) does not happen before then.
The problem with PT is that making it more attractive is not always technically, economicaly or practically viable.
To make PT practical around here, we would need many new subway lines to link the island with the shores and extend existing ones further east/west. Since it costs over $1bn to build an underwater tunnel and something like $2bn/(10km + 5 stations) afterwards, this quickly becomes an unreasonable chunk of the local ~$75bn economy.
Bacteria will collect on the water handles/buttons... you touch those both before and after washing. (Unless they have automatic shut-off.)
IIRC, it is a scientific fact that the genitals are typically the cleanest (bacteria-wise) area on the human body. I would not be surprised if people accumulated more bacteria from stuff they touch on their way out of the bathroom than from themselves.
Depending on where you work and where you live, driving can make the difference between a 30 minutes drive and a 2-3h public transportation journey. Public transportation zealots have to keep things in perspective. Politicians say people should use public transportation... but how many do so themselves on a regular basis?
If I had a job with a ~100km round-trip, already owned a car and gasoline prices mysteriously tripled overnight, I would most likely do as you said and still keep driving - I wouldn't want to waste over 2h/day doing nearly nothing on public transportation when I have a [insert your favourite ~50MPG or better ULEV-compliant (sub-)compact car here] parked downstairs. Actually, I will be working at a place that is a ~100km round-trip starting in January and am seriously considering getting a Yaris - my previous job was in the same area and the ~2h it was taking to get there with public transportation would have driven me nuts, were it not for four-days weeks.
Instead of taxing gasoline, they should increase registration fees, tax unnecessary supersized vehicles with supersized engines and offer registration fee reductions for low emission, high efficiency, well-maintained, etc. vehicles down to (or even below) current rates. This way, people with average cars could work their way around the registration hikes/taxes by keeping their vehicles in perfect working order and by opting for more fuel-efficient and low-emission vehicles in the future. Many places already do things along those lines, some even go as far as offering subventions and tax deductions for hybrids.
Imagine a world where gasoline prices were artificially inflated to $20/gal... many people would not even be able to afford public transportation anymore, roads would be covered with soil and grass planted on them, riding horses would become popular again and this would not help the methane issue in the end. Cows and other farm animals are significant sources of atmospheric methane.
Mankind's influence on climate goes well beyond cars, people quickly forget that.
All this said though, I live in Canada and I would welcome an extra 5C from October through May. Since I am about 20m above sea level and live on a second floor, I am not particularly concerned about flooding:)
Just put a straight element with the same heating capacity as the magnetron that goes in this microwave heater inside a straight pipe and have the water flow along it. If you want a greater temperature gradient, use a longer pipe or put a few lower-powered elements in parallel inside a larger pipe to get a larger temperature gradient.
Now, tap water is ~15-20C, has 4.17kJ/kg/C specific heat so heating 1L/s to 40C (which is rather low) requires 4.17kJ * 25C = 83kJ/s... you need 104kW, assuming 100% transfer efficiency. A typical residential power distribution pannel is 200A/240V, that's only 48kW... barely enough to stream ~35C water from 20C tap.
WiFi works on unlicensed spectrum, companies cannot claim it as their own... that spectrum is for all to use however they please within the limits set by the relevant agencies. Trying to bury the new guy by boosting your transmitter's output would most likely violate the peak radiation limit and get your transmitter shut down if not brought back within compliance.
What would be really useful is moving WiFi towards true spread-spectrum modulation to reduce performance degradation from multiple full and partial overlaps. The main issues with this would be extra complexity, longer channel scanning/sync times and yet more bandwidth (but at a lower mW/MHz density) per useful channel.
But today, people are going mobile with PDAs, laptops, cell phones, etc.
Me, I got a wireless router at home and use wireless wherever I can simply for physical and electrical safety - I do not have to worry about a freak power incident zapping my laptop through the LAN port when I am connected wirelessly. I also do not have to worry about network cables being tripped over and ripping the Ethernet jack off my laptop's PCB.
I go wired only for large data transfers... preferably over FireWire for large transfers since my laptop and all the other PCs I use only have 100Mbps Ethernet. Also, FireWire networks are more private.
I did not need a floppy to install on my 200GB SATA drive with my i845 chipset... so I guess that requirement depends on circumstances.
Worst case, many modern BIOSes can map flash readers or a USB Flash drive as the floppy. My PC came with a multi-format reader but other than BIOS flashing, I have little to no use for flash memory so I temporarily plug a floppy drive whenever I need to flash since my desktop's case has no external 3.5" bays... I even had to buy an HDD-to-floppy power splitter because the PSU had no FDD plug. Can't have everything in an SFF-style micro-ATX case.
I thought Intel, AMD, ATI, nVidia and many others being sued by Microlinc over packet-based serial links for computer interconnects (a story from what I call the "dormant patent" department - wait for your stealth invention to become common practice then come out of hibernation to sue everybody for whatever they are worth) would be far more newsworthy... but that topic got rejected when I submitted it a few days ago and we get this instead.
HDDs have been able to burst 150+MB/s out of their on-board caches... but the actual heads-to/from-platter peak transfer rates are much less than 1Gbps (~100MB/s) even for the fastest drives. HDDs able to max out an SATA-1.5G link all alone are still many years away.
3Gbps SATA is useful for storage subsystems, like SATA-attached RAID controllers - no need to waste a PCI(-E/X) slot to get decent performance anymore and it also leaves more PCI(-X) bandwidth available to other devices. Another purpose is SATA multi-port switches where one SATA-3G port may have multiple SATA-1.5G devices attached to it. I think some people are also considering SATA for clustering networks.
There is one simple major issue with the DMCA: it is fundamentally redundant.
Copyright infringement and facilitating crimes were already illegal before the DMCA. Before the DMCA, only 'criminals' infringed on copyright and wrote software to facilitate it... and after the DMCA, only 'criminals' infringe on copyright.
The only truly new thing with DMCA its unprecedented potential for abuse... and DMCA abuses account for the majority of DMCA case I have read about. As someone else pointed out and as I also have said in the past, legal != just/fair.
As I said, I do not care for absolute performance, what I need is a system that can survive cruel and unusual workloads yet remain responsive. Try debugging a multi-threaded program with VisualStudio on non-MC/MT CPUs, if you are lucky enough to break inside a touchy "unbreakable" loop, the system locks up for several minutes before recovering.
Since I need MP/MT for my computers to survive a "normal" workload but am not generally otherwise performance-bound, the cheapest thing that gives me real concurrent execution is all I need. The P4 may be slower but it is by far the cheapest family that meets my minimum requirement, wether it be HT or dual-core. This is why I say Intel should transform Netburst into a pure multi-threading-oriented core - yes, that would kill single-threaded performance but it would do wonders for overall throughput and open many opportunities to simplify the core.
As for Itanium, AMD has proved with Opteron that the same performance and scalability levels are achievable with x86. The only thing to be gained from ditching x86 is the complex instruction set.
If my experience with external enclosures is any indication, I would bet on the power supplies.
Most cheap/small PSUs use flyback topology and decoupling capacitors that have AC ripple ratings well below what they should be. Ordinary 1000-2200uF capacitors have AC ripple ratings under 1A and a service life of 2000-5000h at that rating. On 2-3A 5-12V rails with flyback PSU, this 1A(rms) rating is easily exceeded. This is why I have made it a standard practice to replace bulk decoupling capacitors in my storage boxes and PC PSUs by 2700uF caps rated for 7000+h at 3.6A ripple supplemented with surface-mounted 10uF MLCC caps to relieve the electrolytic caps from harmonics in the 100kHz-10MHz range wherever possible.
When output capacitors age in a flyback PSU, their impedance increase and the capacitors becomes unable to absorb high-frequency energy. This causes spikes in the output voltage and if the PSU does not have a proper shunt regulator or over-voltage crowbar circuitry, those spikes can definitely kill electronics - I have seen/measured microsecond-scale spikes go as high as 15V on a 5V rail and 20V on a 12V rail.
At least this will solve M$'s supply problem after the news makes the news and front pages. Remaining early adopters will want to wait until M$ announces its take on things before thinking of buying again.
On the XBox-360 poll, my answer was Never / X-Mas 2006. Launch prices are too high, I do not care about the coolness factor of having 0-dayz new stuff and I had doubts about how well the initial hardware would work. Quitting the bleeding-edge to stick with mature mainstream stuff has saved me quite a bit of cash and trouble, I'll stick to that.
This article has responses from a few game devs stating that their launch titles are multi-threaded and a M$ threading person said: "Six months ago, we had only looked at a handful of games. Most of those games were single-threaded. Today, we've evaluated most launch titles and the majority are using multiple threads."
I did state that AMD had the performance and power advantages in my first paragraph, bringing that point back is redundant.
If Joe Sixpack wants a dual-core chip and he does not mind the heat and does not care too much about absolute performance, the $100 difference between the two sides' entry-level dual-core offering will settle it.
In my case, I do not care for performance much as long as the the system does not crumble under my moderate-to-heavy multitasking habbits... and here, my 3GHz P4-NW/HT performs beatifully where my Athlon64-3000+ chokes to death. An X2 would certainly be better but the improvements compared to HT or P4D from neither would be nowhere near as dramatic as the price difference.
As much as I like AMD's CPU (a little) and system architectures better, I still prefer Intel where responsiveness/throughput/price is concerned. At the moment and for the foreseeable future, AMD has no low-end parts that can compete with P4/HT and P4D on heavy multithreaded/multitasking workloads.
I wish Intel would back-port Prescott's improvements to the Northwood pipeline and integrate a memory controller... with a few other tweaks, they could save Netburst's face.
AMD cannot outsell Intel because it lacks the mindshare and fab capacity... but this is changing: AMD is up to nearly 30% market share (desktop+server) while Intel is down to about 70%, this is +10% for AMD, -10% for Intel and -20% for the gap between the two compared to last year. If AMD cracks Dell, things could even out fairly fast. AMD's upcoming glueless (non-Horus&all) 16-ways and 32-ways (as in sockets) chips will put immense pressure on Dell's server business... and any other vendor who currently relies primarily on Xeons.
As for IBM making an AMD64-compatible PPC, this will never happen. The AMD64 instruction set is the good old x86 instruction set we all know and love/hate, it only adds prefixes to modify register/memory/operand addressing modes. The instruction set structure dictates many rules about how code can be efficiently executed so the PPC would certainly require extensive modifications to deal with x86-64 well beyond the instruction decoders.
As for IBM's involvement in AMD's business, AFAIK it is only fabbing-related stuff like process tweaks, AMD would still have access to these through licensing deals. This partnership came to be back when AMD could not afford to build a fab for Athlon production and partnered with Motorola and IBM to split costs. Now that AMD can fly on its own again, the IBM partnership is primarily an historic convenience.
With the upcoming 16/32-ways Opteron chips, AMD will take another step up in the HPC major leagues and should be able to make IBM's Power5 run for its money. Opterons are already starting to eat into the Top500's top-100 with 12 systems there... and now that Opteron has been selected for the next ASCI, many supercomputer sites are now considering Opterons for their next build. I am expecting a mass invasion of the top-20 in 2007-8. Clock-for-clock and chip-for-chip, Opteron fares fairly well against Power5 - keep in mind that Power5 is a multi-threading chip.
Actually, the R520 only has 16... R580 might upgrade that to 24.
Having multiple GPUs works well (at least in theory) because graphics rendering is implicitly massively parallel. But GPUs usually choke on RAM bandwidth so going multi-core is pointless unless RAM bandwidth can keep up... and if able to feed twice as much bandwidth to a chip, it would probably be simpler and more efficient to simply double the number of pipelines - we're back on choking upon RAM bandwidth. There is no point in scaling the number of pipelines beyond the point where RAM bandwidth is routinely exhausted: in (otherwise ideal) parallel processing, if 5% of the processing ends up serialized (RAM/IO bottleneck for example), the optimal speed-up is 20X when given infinite processors. So, to fully exploit 24 pipelines, memory bottlenecking must happen less than 1% of the time. (Even at 1% the maximum speed is still under 22X that of a single pipeline. For a concrete example of how a 2-3% improvement in memory contention can make major differences in throughput, look at ATI's 5.11 Catalyst drivers, 10-30% extra performance from memory controller/scheduler tweaking.)
As for the number of simultaneous operations, it depends on the definition. Many Pixel/Vertex Shader instructions (like everything that involves 4x4 matrix products) expand into large multiplication sets and additions (16 muls + 12 adds for 4x4) once they reach the hardware pipeline.
You're right for the power and performance... but price-wise, Athlon FX and X2 are not particularly cheaper than their closest Pentium equivalents - for example, the cheapest X2 costs nearly $100 more than the cheapest Pentium-D and both companies' top chips bear the same $1000-ish price tag. Below the upper-mid-range, things do become very much one-sided in AMD's favour though.
As for why they need marketing, that is a matter of mindshare. It does not matter how good your platform is if you are unable to get big buyers to place orders because they do not know about your products or are not aware of them being potentially superior to the stuff they usually order. We know some CEOs&others from large companies have become aware of the performance, scalability and power advantages of AMD64 chips by the fact that Dell has started making calls to notify its suppliers that it may start building Opteron boxes. A little more marketing could go a long way towards cracking tough nuts like Dell.
You forgot:
* Non-Windows (or OSX) OSes
* Non-x86 (or PPC for OSX) CPUs
* User accounts that do not have software and driver install privileges
* Disabling autorun (gpedit.msc -> Admin Templates -> Turn Off Autoplay = Enabled for all)
Gaming console releases are on 4-5 years cycles with often completely new architectures, this is very different from the PC world where every individual component class is standard, each class has multiple incremental improvements from various vendors each year and each class usually has multiple price/performance points to choose from.
The PS2 was released in 2000, the PS3 is coming some time next year so the PS2 will have a commercial life of 6+ years. The PS3's release will not lead to overnight discontinuation of ongoing PS2 projects and PS2 console sales so the PS3 commercial life might exceed seven years. Console and PC product cycles are two fundamentally different things simply because a console's specs are set in stone from design to end-of-life, no PC-style "hardware of the day" syndrome. There is also the matter of exclusive games that make all other game consoles irrelevant if one of the games you really want to play exists on only one platform.
I borrowed a crypto book yesterday, it had a little data flow graph for the cipher that showed no reason why AES could not scale to larger key and block sizes. The 256bits AES "limit" seems purely artificial and Wikipedia confirmed my suspicion: Rijndael can use arbitrarily large blocks and keys. The NSA arbitrarily set the maximum AES key size to 256bits and had TwoFish or Serpent won the AES competition instead of Rijndael, their keys and block sizes would have been set to the same limits for the official AES implementation.
Yes, I was initially talking through my hat but it turns out there was nothing wrong with my suggestion. Nothing prevents anybody from taking the Rijndael code and scale it to 1KB keys and blocks if someone wanted to. Only problem is that Rijndael's (AES) performance scales worst than the two other main AES finalists beyond 192bits so a 1KB (8kbits) keys with supersized blocks could be awfully slow.
But I suspect more than 60% of Celerons are fully functionnal P4s that were lobotomized to avoid flooding the market with high-end parts that would kill ASPs. If we look at AMD that does market-specific core respins, we know yields are good enough to make area-optimized all-or-nothing cores more desirable than one-size-fits-all lobotomizable designs when production capacity is somewhat limited.
AFAIK, if you put water in a kettle and only come back once it is singing steady and loud, the water should be boiling near 100C... I usually have more important/useful things to do than waiting after it.
Automatic brewers (at least the cheap ones) do roughly the same thing but in small increments. Water drips from the tank into a heating chamber with an intake floater valve, is heated up to the boiling point where the chamber's contents gets shot into the ground coffee by the rising pressure and the subsequent drop in chamber pressure draws in the next drip of fresh water along with whatever did not exit the dump pipe. Once the chamber's heating element has reached its cruise temperature, water dripping in boils instantaneously and the process becomes swift and noisy... and sometimes makes a steamy overflow mess when the boil&draw action pumps water faster than it can get through the coffee and filter. These spills were hot enough to ruin the counter's melamine finish. (The heat softened the melamine enough to let gases trapped under it cause large blister-like deformations. The melamine must have been misapplied or the glue improperly mixed/cured.)
Slashdot Mirror
I know PT does not mean only subs... but this is the only way to go when there is no open land for rails and the lines have to cross a major waterway: doing light rails would require mass expropriation for the access ramps since the rails would have to be high enough to clear cargo ships. With the vacancy rates sometimes dipping below 2% on the island and surroundings, there is no need to motivate people into moving there more than they already are when housing availability and rent are major concerns... though vacancy was up to 6% this year thanks to rent-induced mass exodus and many completed or on-going residential projects.
With nearly no land available for new developments, there is not much economic advantage to be gained from putting bilions into improving public transportation... it would only make it more accessible and convenient with only marginal economic gains on a multi-bilion expense. I bet the largest gain would end up being on property taxes... and maybe beefier supplemental vehicle registration fee for the privilege of owning a car when PT is available around the primary residence.
Every politician who tells people to use PT should have his limo service cancelled, his expense account frozen, his license suspended, his car(s) sent to storage for a year and be provided with a year's worth of country-wide bus/train/sub/etc. passes. In many cities, it would permanently shut them up until someone fixes it - or at least attempts to.
PT is like Netburst and HT: they are fundamentally good ideas but are often poorly implemented.
RAM latency is not much of an issue for proper multi-threading CPUs... having one thread stalled due to IO and instruction dependencies simply leaves more execution ressources to the other threads and as long as at least two out of four or more threads have ready-and-able instructions, the execution unit can operate near full-load most of the time and completely hide latencies as far as sustained throughput is concerned. Latency is not as much of a killer now as it was in the '80s, thanks in large part to on-chip caches.
As for the vector registers, we have small-scale but otherwise similar stuff now with SSE/AltiVec/etc. but they fall tragically short from the sort of sustained throughput and efficiency dedicated hardware can achieve. Commodity chips are inexpensive and CPUs have enough execution resources to do this stuff sufficiently fast as a loop, they are also simple to program, deploy, service and replace. In the '80s, only companies could afford anything much beyond a C64, there was no significant processing power below several thousand dollars.
For the return to architecture, architecture never really disappeared. Systems that are primarily IO-bound never stopped requiring massive architecture design efforts. Since HPC is highly latency-intolerant, you want to have as close to 0% node-to-node communications as possible: 0.1% processing spent on communications translates into a theoretical maximum of 1000X performance gain given infinite CPUs or ~641X given 1024 CPUs. Come 2008 and quad-core 16/32-ways glueless MP CPUs, we will probably have reached the limits of what can be practically packed in a single system - each CPU will have enough processing power to saturate its local RAM and IO bus. The only ways to work around this is system-scale integration and ASIC implementations - integrate data producers and consumers to reduce reliance on external bandwidth-limited busses.
Around 2010, the computing world will stop until the next breakthrough, assuming it (either the breakthrough or stop) does not happen before then.
The problem with PT is that making it more attractive is not always technically, economicaly or practically viable.
To make PT practical around here, we would need many new subway lines to link the island with the shores and extend existing ones further east/west. Since it costs over $1bn to build an underwater tunnel and something like $2bn/(10km + 5 stations) afterwards, this quickly becomes an unreasonable chunk of the local ~$75bn economy.
Bacteria will collect on the water handles/buttons... you touch those both before and after washing. (Unless they have automatic shut-off.)
IIRC, it is a scientific fact that the genitals are typically the cleanest (bacteria-wise) area on the human body. I would not be surprised if people accumulated more bacteria from stuff they touch on their way out of the bathroom than from themselves.
Depending on where you work and where you live, driving can make the difference between a 30 minutes drive and a 2-3h public transportation journey. Public transportation zealots have to keep things in perspective. Politicians say people should use public transportation... but how many do so themselves on a regular basis?
:)
If I had a job with a ~100km round-trip, already owned a car and gasoline prices mysteriously tripled overnight, I would most likely do as you said and still keep driving - I wouldn't want to waste over 2h/day doing nearly nothing on public transportation when I have a [insert your favourite ~50MPG or better ULEV-compliant (sub-)compact car here] parked downstairs. Actually, I will be working at a place that is a ~100km round-trip starting in January and am seriously considering getting a Yaris - my previous job was in the same area and the ~2h it was taking to get there with public transportation would have driven me nuts, were it not for four-days weeks.
Instead of taxing gasoline, they should increase registration fees, tax unnecessary supersized vehicles with supersized engines and offer registration fee reductions for low emission, high efficiency, well-maintained, etc. vehicles down to (or even below) current rates. This way, people with average cars could work their way around the registration hikes/taxes by keeping their vehicles in perfect working order and by opting for more fuel-efficient and low-emission vehicles in the future. Many places already do things along those lines, some even go as far as offering subventions and tax deductions for hybrids.
Imagine a world where gasoline prices were artificially inflated to $20/gal... many people would not even be able to afford public transportation anymore, roads would be covered with soil and grass planted on them, riding horses would become popular again and this would not help the methane issue in the end. Cows and other farm animals are significant sources of atmospheric methane.
Mankind's influence on climate goes well beyond cars, people quickly forget that.
All this said though, I live in Canada and I would welcome an extra 5C from October through May. Since I am about 20m above sea level and live on a second floor, I am not particularly concerned about flooding
Just put a straight element with the same heating capacity as the magnetron that goes in this microwave heater inside a straight pipe and have the water flow along it. If you want a greater temperature gradient, use a longer pipe or put a few lower-powered elements in parallel inside a larger pipe to get a larger temperature gradient.
Now, tap water is ~15-20C, has 4.17kJ/kg/C specific heat so heating 1L/s to 40C (which is rather low) requires 4.17kJ * 25C = 83kJ/s... you need 104kW, assuming 100% transfer efficiency. A typical residential power distribution pannel is 200A/240V, that's only 48kW... barely enough to stream ~35C water from 20C tap.
This does not work.
WiFi works on unlicensed spectrum, companies cannot claim it as their own... that spectrum is for all to use however they please within the limits set by the relevant agencies. Trying to bury the new guy by boosting your transmitter's output would most likely violate the peak radiation limit and get your transmitter shut down if not brought back within compliance.
What would be really useful is moving WiFi towards true spread-spectrum modulation to reduce performance degradation from multiple full and partial overlaps. The main issues with this would be extra complexity, longer channel scanning/sync times and yet more bandwidth (but at a lower mW/MHz density) per useful channel.
But today, people are going mobile with PDAs, laptops, cell phones, etc.
Me, I got a wireless router at home and use wireless wherever I can simply for physical and electrical safety - I do not have to worry about a freak power incident zapping my laptop through the LAN port when I am connected wirelessly. I also do not have to worry about network cables being tripped over and ripping the Ethernet jack off my laptop's PCB.
I go wired only for large data transfers... preferably over FireWire for large transfers since my laptop and all the other PCs I use only have 100Mbps Ethernet. Also, FireWire networks are more private.
I did not need a floppy to install on my 200GB SATA drive with my i845 chipset... so I guess that requirement depends on circumstances.
Worst case, many modern BIOSes can map flash readers or a USB Flash drive as the floppy. My PC came with a multi-format reader but other than BIOS flashing, I have little to no use for flash memory so I temporarily plug a floppy drive whenever I need to flash since my desktop's case has no external 3.5" bays... I even had to buy an HDD-to-floppy power splitter because the PSU had no FDD plug. Can't have everything in an SFF-style micro-ATX case.
I thought Intel, AMD, ATI, nVidia and many others being sued by Microlinc over packet-based serial links for computer interconnects (a story from what I call the "dormant patent" department - wait for your stealth invention to become common practice then come out of hibernation to sue everybody for whatever they are worth) would be far more newsworthy... but that topic got rejected when I submitted it a few days ago and we get this instead.
HDDs have been able to burst 150+MB/s out of their on-board caches... but the actual heads-to/from-platter peak transfer rates are much less than 1Gbps (~100MB/s) even for the fastest drives. HDDs able to max out an SATA-1.5G link all alone are still many years away.
3Gbps SATA is useful for storage subsystems, like SATA-attached RAID controllers - no need to waste a PCI(-E/X) slot to get decent performance anymore and it also leaves more PCI(-X) bandwidth available to other devices. Another purpose is SATA multi-port switches where one SATA-3G port may have multiple SATA-1.5G devices attached to it. I think some people are also considering SATA for clustering networks.
There is one simple major issue with the DMCA: it is fundamentally redundant.
Copyright infringement and facilitating crimes were already illegal before the DMCA. Before the DMCA, only 'criminals' infringed on copyright and wrote software to facilitate it... and after the DMCA, only 'criminals' infringe on copyright.
The only truly new thing with DMCA its unprecedented potential for abuse... and DMCA abuses account for the majority of DMCA case I have read about. As someone else pointed out and as I also have said in the past, legal != just/fair.
As I said, I do not care for absolute performance, what I need is a system that can survive cruel and unusual workloads yet remain responsive. Try debugging a multi-threaded program with VisualStudio on non-MC/MT CPUs, if you are lucky enough to break inside a touchy "unbreakable" loop, the system locks up for several minutes before recovering.
Since I need MP/MT for my computers to survive a "normal" workload but am not generally otherwise performance-bound, the cheapest thing that gives me real concurrent execution is all I need. The P4 may be slower but it is by far the cheapest family that meets my minimum requirement, wether it be HT or dual-core. This is why I say Intel should transform Netburst into a pure multi-threading-oriented core - yes, that would kill single-threaded performance but it would do wonders for overall throughput and open many opportunities to simplify the core.
As for Itanium, AMD has proved with Opteron that the same performance and scalability levels are achievable with x86. The only thing to be gained from ditching x86 is the complex instruction set.
If my experience with external enclosures is any indication, I would bet on the power supplies.
Most cheap/small PSUs use flyback topology and decoupling capacitors that have AC ripple ratings well below what they should be. Ordinary 1000-2200uF capacitors have AC ripple ratings under 1A and a service life of 2000-5000h at that rating. On 2-3A 5-12V rails with flyback PSU, this 1A(rms) rating is easily exceeded. This is why I have made it a standard practice to replace bulk decoupling capacitors in my storage boxes and PC PSUs by 2700uF caps rated for 7000+h at 3.6A ripple supplemented with surface-mounted 10uF MLCC caps to relieve the electrolytic caps from harmonics in the 100kHz-10MHz range wherever possible.
When output capacitors age in a flyback PSU, their impedance increase and the capacitors becomes unable to absorb high-frequency energy. This causes spikes in the output voltage and if the PSU does not have a proper shunt regulator or over-voltage crowbar circuitry, those spikes can definitely kill electronics - I have seen/measured microsecond-scale spikes go as high as 15V on a 5V rail and 20V on a 12V rail.
At least this will solve M$'s supply problem after the news makes the news and front pages. Remaining early adopters will want to wait until M$ announces its take on things before thinking of buying again.
On the XBox-360 poll, my answer was Never / X-Mas 2006. Launch prices are too high, I do not care about the coolness factor of having 0-dayz new stuff and I had doubts about how well the initial hardware would work. Quitting the bleeding-edge to stick with mature mainstream stuff has saved me quite a bit of cash and trouble, I'll stick to that.
This is a myth.
This article has responses from a few game devs stating that their launch titles are multi-threaded and a M$ threading person said: "Six months ago, we had only looked at a handful of games. Most of those games were single-threaded. Today, we've evaluated most launch titles and the majority are using multiple threads."
I did state that AMD had the performance and power advantages in my first paragraph, bringing that point back is redundant.
If Joe Sixpack wants a dual-core chip and he does not mind the heat and does not care too much about absolute performance, the $100 difference between the two sides' entry-level dual-core offering will settle it.
In my case, I do not care for performance much as long as the the system does not crumble under my moderate-to-heavy multitasking habbits... and here, my 3GHz P4-NW/HT performs beatifully where my Athlon64-3000+ chokes to death. An X2 would certainly be better but the improvements compared to HT or P4D from neither would be nowhere near as dramatic as the price difference.
As much as I like AMD's CPU (a little) and system architectures better, I still prefer Intel where responsiveness/throughput/price is concerned. At the moment and for the foreseeable future, AMD has no low-end parts that can compete with P4/HT and P4D on heavy multithreaded/multitasking workloads.
I wish Intel would back-port Prescott's improvements to the Northwood pipeline and integrate a memory controller... with a few other tweaks, they could save Netburst's face.
AMD cannot outsell Intel because it lacks the mindshare and fab capacity... but this is changing: AMD is up to nearly 30% market share (desktop+server) while Intel is down to about 70%, this is +10% for AMD, -10% for Intel and -20% for the gap between the two compared to last year. If AMD cracks Dell, things could even out fairly fast. AMD's upcoming glueless (non-Horus&all) 16-ways and 32-ways (as in sockets) chips will put immense pressure on Dell's server business... and any other vendor who currently relies primarily on Xeons.
As for IBM making an AMD64-compatible PPC, this will never happen. The AMD64 instruction set is the good old x86 instruction set we all know and love/hate, it only adds prefixes to modify register/memory/operand addressing modes. The instruction set structure dictates many rules about how code can be efficiently executed so the PPC would certainly require extensive modifications to deal with x86-64 well beyond the instruction decoders.
As for IBM's involvement in AMD's business, AFAIK it is only fabbing-related stuff like process tweaks, AMD would still have access to these through licensing deals. This partnership came to be back when AMD could not afford to build a fab for Athlon production and partnered with Motorola and IBM to split costs. Now that AMD can fly on its own again, the IBM partnership is primarily an historic convenience.
With the upcoming 16/32-ways Opteron chips, AMD will take another step up in the HPC major leagues and should be able to make IBM's Power5 run for its money. Opterons are already starting to eat into the Top500's top-100 with 12 systems there... and now that Opteron has been selected for the next ASCI, many supercomputer sites are now considering Opterons for their next build. I am expecting a mass invasion of the top-20 in 2007-8. Clock-for-clock and chip-for-chip, Opteron fares fairly well against Power5 - keep in mind that Power5 is a multi-threading chip.
Actually, the R520 only has 16... R580 might upgrade that to 24.
Having multiple GPUs works well (at least in theory) because graphics rendering is implicitly massively parallel. But GPUs usually choke on RAM bandwidth so going multi-core is pointless unless RAM bandwidth can keep up... and if able to feed twice as much bandwidth to a chip, it would probably be simpler and more efficient to simply double the number of pipelines - we're back on choking upon RAM bandwidth. There is no point in scaling the number of pipelines beyond the point where RAM bandwidth is routinely exhausted: in (otherwise ideal) parallel processing, if 5% of the processing ends up serialized (RAM/IO bottleneck for example), the optimal speed-up is 20X when given infinite processors. So, to fully exploit 24 pipelines, memory bottlenecking must happen less than 1% of the time. (Even at 1% the maximum speed is still under 22X that of a single pipeline. For a concrete example of how a 2-3% improvement in memory contention can make major differences in throughput, look at ATI's 5.11 Catalyst drivers, 10-30% extra performance from memory controller/scheduler tweaking.)
As for the number of simultaneous operations, it depends on the definition. Many Pixel/Vertex Shader instructions (like everything that involves 4x4 matrix products) expand into large multiplication sets and additions (16 muls + 12 adds for 4x4) once they reach the hardware pipeline.
You're right for the power and performance... but price-wise, Athlon FX and X2 are not particularly cheaper than their closest Pentium equivalents - for example, the cheapest X2 costs nearly $100 more than the cheapest Pentium-D and both companies' top chips bear the same $1000-ish price tag. Below the upper-mid-range, things do become very much one-sided in AMD's favour though.
As for why they need marketing, that is a matter of mindshare. It does not matter how good your platform is if you are unable to get big buyers to place orders because they do not know about your products or are not aware of them being potentially superior to the stuff they usually order. We know some CEOs&others from large companies have become aware of the performance, scalability and power advantages of AMD64 chips by the fact that Dell has started making calls to notify its suppliers that it may start building Opteron boxes. A little more marketing could go a long way towards cracking tough nuts like Dell.
You forgot:
* Non-Windows (or OSX) OSes
* Non-x86 (or PPC for OSX) CPUs
* User accounts that do not have software and driver install privileges
* Disabling autorun (gpedit.msc -> Admin Templates -> Turn Off Autoplay = Enabled for all)
Gaming console releases are on 4-5 years cycles with often completely new architectures, this is very different from the PC world where every individual component class is standard, each class has multiple incremental improvements from various vendors each year and each class usually has multiple price/performance points to choose from.
The PS2 was released in 2000, the PS3 is coming some time next year so the PS2 will have a commercial life of 6+ years. The PS3's release will not lead to overnight discontinuation of ongoing PS2 projects and PS2 console sales so the PS3 commercial life might exceed seven years. Console and PC product cycles are two fundamentally different things simply because a console's specs are set in stone from design to end-of-life, no PC-style "hardware of the day" syndrome. There is also the matter of exclusive games that make all other game consoles irrelevant if one of the games you really want to play exists on only one platform.
I borrowed a crypto book yesterday, it had a little data flow graph for the cipher that showed no reason why AES could not scale to larger key and block sizes. The 256bits AES "limit" seems purely artificial and Wikipedia confirmed my suspicion: Rijndael can use arbitrarily large blocks and keys. The NSA arbitrarily set the maximum AES key size to 256bits and had TwoFish or Serpent won the AES competition instead of Rijndael, their keys and block sizes would have been set to the same limits for the official AES implementation.
Yes, I was initially talking through my hat but it turns out there was nothing wrong with my suggestion. Nothing prevents anybody from taking the Rijndael code and scale it to 1KB keys and blocks if someone wanted to. Only problem is that Rijndael's (AES) performance scales worst than the two other main AES finalists beyond 192bits so a 1KB (8kbits) keys with supersized blocks could be awfully slow.
But I suspect more than 60% of Celerons are fully functionnal P4s that were lobotomized to avoid flooding the market with high-end parts that would kill ASPs. If we look at AMD that does market-specific core respins, we know yields are good enough to make area-optimized all-or-nothing cores more desirable than one-size-fits-all lobotomizable designs when production capacity is somewhat limited.
AFAIK, if you put water in a kettle and only come back once it is singing steady and loud, the water should be boiling near 100C... I usually have more important/useful things to do than waiting after it.
Automatic brewers (at least the cheap ones) do roughly the same thing but in small increments. Water drips from the tank into a heating chamber with an intake floater valve, is heated up to the boiling point where the chamber's contents gets shot into the ground coffee by the rising pressure and the subsequent drop in chamber pressure draws in the next drip of fresh water along with whatever did not exit the dump pipe. Once the chamber's heating element has reached its cruise temperature, water dripping in boils instantaneously and the process becomes swift and noisy... and sometimes makes a steamy overflow mess when the boil&draw action pumps water faster than it can get through the coffee and filter. These spills were hot enough to ruin the counter's melamine finish. (The heat softened the melamine enough to let gases trapped under it cause large blister-like deformations. The melamine must have been misapplied or the glue improperly mixed/cured.)