In a game, how do you start working on the next frame before being done with the current frame? Sure, it could be done but it would introduce input-output lag equivalent to however deep that frame pipeline might be. At 30FPS with two pipelined frames, the perceived lag would go from 1/30s to 1/20s, which is often enough to be irritating.
The bulk of game computations is game logic and physics then adjusting scene geometry accordingly, all of which is strict frame-by-frame computing. Because of this, multi-threaded games can only truly benefit from multi-threading by distributing this intra-frame computing. (Noticed how SLI and CrossFire are also designed for distributed rendering to push single frames out faster with 60-70% gain instead of alternate frame rendering where it would be easier to approach 100%?)
I am a hardware guy myself, I like VHDL and programmable logic where I am guaranteed that everything happens on every clock. Right now, I am writing an application letter with the intent of applying for jobs along the lines of "Processor Architect" - I would love to be among the people who brought SMT to Athlon64 and I have a number of ideas about how to get there. I am much more confident about thinking in parallel on an RTL basis than in a software environment where I can never know with any certainty anything about what happened/is happening/will happen.
The above paragraph should answer your mindset wondering. In short: there are much fewer uncertainties on the hardware side of things.
There is no multiprocessor advantage to games now only because most games people look at are written fundamentally single-threaded, not using threads for any meaningful/heavy things. In this common case, the only potentially significant gain from SMT/MP for current games is offloading of background processes.
Most loops where each iteration is independant from previous results can be distributed. For bulk processing an object array (say a display list for games), a shared counter can be used to distribute objects across threads running "while((nObj = _InterlockedExchangeAdd(&sharedCount, -1)) >= 0) pObj[nObj]->Process();" - I am certain many loops would be suitable for parallelizing in a similar manner when per-object processing is sufficiently non-trivial to prevent excessive bus locking.
For the costs, I did a quick check and $10k in the upper range for US/Canada patents with $20k being the upper range for WIPO (Patent Cooperation Treaty - PCT) filings.
The main advantages of filing with WIPO is savings in translation costs, no need to follow each country's specific procedures and having to redundantly defend patent challenges for each individual registration. None of the FAQs I read give any specifics about PCT pricing but they do imply additional costs less than individual filings to get actual protection in each region/country.
Last time I heard, patent application was around $10k per country and around $100k for worldwide filing.
Filing a patent in any country prevents anyone else from filing the same patent in all others but only provides legal enforcement in the countries where the patent was was filed. In most cases, inventors can file in other countries as necessary to extend legal protection but doing it incrementally can quickly cost more than an international filing.
(This is what I was told at a conference about a year or two ago.)
A basic desktop box should fall in the 100-150W range since decently priced low(-electrical)-power desktop chips are becoming a thing of the past.
My main PC is plugged on a Back-UPS XS1000 and the monitoring software tells me my idle P4-3G (with Radeon 9600XT, 1GB RAM, SB Audigy2, P4P800VM, 1xHDD) uses about 110W while idle, 160W under full CPU load and 180W under full system load. (HL2 + SETI)
My other PC with a P3-1G uses 120W while running SETI and 90W while idle... 15W for chipset/RAM, 15W for the idle CPU + VRM losses, 15W for the disk drives, 35W for fans/sound/network/video/etc., 70W sub-total, so I guess the PSU in that PC (TigerPro TP300, came with some CAN$45 case - I needed a case, the PSU was a bonus/piece-of-crap) has ~70% efficiency at 3.7A ripple current and the cheaper 'standard' ones are rated under 1A.)
On one hand, people start talking about ways of preventing solar heat from reaching Earth to reduce global warming, on the other hand, people are getting ready to create power plants that will eventually generate gigawatts.
Well, fusion plants are necessary to replace coal/diesel/fission plants but global warming initiatives should simply focus on making everything more power-efficient and low-emission instead of coming up with inconvenient/impossible ideas. (Some european minister said he was shocked at how much power standby electronics use and suggested that people should not leave TVs&all plugged in when not in use... but how many people would unplug their VCR/TV/PC/etc. to save 5-20W per appliance given the inconveniences?)
RAM densities, bus widths, working frequencies, etc. go up, this causes the current to increase and eventually cancel that gain, leading to a net increase in actual power.
DDR2 currently uses the same densities, comparable frequencies, etc. as DDR but DDR2's advantage will diminish as the gap between it and DDR grows. By the time we switch to DDR3, DDR2 will probably have exceeded DDR's power envelope, as DDR3 probably will by the time we move on to DDR4/FBDIMM/whatever.
So Intel's least expensive dual-core chip is less than half as expensive as AMD's. This is one of the first times where AMD substantially exceeded Intel pricing in any category.
AMD's implementation of dual cores is certainly more elegant in many ways and the A64 is usually better suited for most of what I do, my wallet says the 820 will be enough if I get dual-core itch.
Measuring wall power does not tell you much about the CPU's power. The only case where a multimeter would be useful for measuring actual CPU power is on motherboards where all CPU power is exclusively taken off the ATX12V connector. This figure would include VRM losses but at least excludes other 12V loads such as AGP/PCIE graphics.
Most of the numbers I quoted were from memory of articles and the actual spec/data-sheets a long time ago... have a look for yourself: http://www.intel.com/products/a_z/p_q.htm
Instead of being above my ideal weight like 60% of industrialized population, I am (still) under my "ideal" weight and most of the weight I have been gaining is muscle. (My primary vehicle is a mountain bicycle... 100-250 kilometers/week depending on weather.)
Somehow, two industrial heavyweights fighting each other in court seems orders of magnitude more even than some relatively defenseless hobbits finding their way through ogre armies while avoiding the riders, the freaky eye, etc.
Why would the power bill go down? P1 = 10-20W P2 = 15-35W P3 = 25-45W P4 = 35-165W
Chipset and RAM power also increases across generations so a few more watts need to be added to each upgrade... and another extra in the 10-20% range for the extra VRM and PSU losses. (PWM regulator technology and components have not changed much over the last 10 years)
But yes, having a faster CPU/RAM does make a substantial difference in firewall responsiveness and throughput. When I upgraded my router from 100MHz to 200MHz, loopback throughput roughly tripled - from 660KB/s to 2.3MB/s. (On top of being slower, a slow chip also spends more of its time processing interrupts and background stuff, leaving less time for 'useful' work, double-hit. Seems like the 100MHz chip in this case was wasting something like half of its time on house-keeping stuff.)
2.3MB/s might not seem like much but I am not expecting local ISPs to offer >20Mbps (combined up+down) for another ~10 years... at least not under CAN$50/month.
Dropping the clause would not prevent copyright/patent laws from coming back in a form very similar to what we have now.
Instead of dropping it, it would have to be reworded in a more specific and restrictive way but even that would not stop industries from buying politicians and having the constitution tweaked in their favour.
Because politicians need cash to win the popularity contest called "elections" (only getting listed already costs a small fortune so only relatively wealthy people can afford it) and lobbyists are backed by large organizations that are full of it, ordinary people are fundamentally screwed.
Re:I wish I had fibre optic on my desktop
on
Lucas's New HQ
·
· Score: 2, Interesting
That would not be terribly efficient.
Most rendering farms probably load all scene textures/geometry to allocated renderers then the workstation simply queue frame requests to each renderer, reducing the bandwidth requirement to little more than finished frames' (1920x1080x3*10/8 = 7.8MB/frame) transfer, assuming the scenes are fairly lenghty or the software is written to cache across scenes/jobs and the scenes use a substantial common texture/geometry base.
So the gigabytes of textures/geometry probably only apply while the rendering farm is doing a cold-start with clean caches. The rendering job's setup becomes even more bandwidth-trivial if the whole thing can be setup using multicast - setting up one or a hundred nodes at once takes practically no additionnal time in that case.
So far, I have two (out of two) failed HDDs from Quantum, one failure and one oddly behaving drive (erratic performance) from Maxtor, none from WD and Seagate, though my only Seagate drive (2.1GB 4500RPM) sounds like a lawn mower - I was certain it would fail but it annoyingly hummed along 24/7 until I retired the PC in which it was 5+ years later.
Thankfully, all four failures occured under warranty. Also, failure in the three total failures was progressive: after each restart, I could copy some (decreasing) amount of data before drive lock-up, usually enough for me to recover my most important files before total failure.
Well, I might have a failed Seagate drive... but I do not have a matching PCI ESDI controller card to test it with.
The ability not to install unnecessary components should be standard... people should not have to use convoluted tricks for this. With Windows 3.xx, almost everything was optionnal and the list kept shrinking with each version until it vanished with XP.
Next step would be a one-click (EULA) install where the installer unilaterally decides to fdisk the first HDD as a single NTFS partition and install everything there - this is all ok since it was agreed to by clicking through the EULA...
The EU's sanctions against M$ were somewhat more severe than the USA's but both are ultimately misguided. For M$, fines are only ghost blips on the accounting radar.
The only thing I can think of that would have been able to do a decent job of enabling fair(er) competition is full, free and open disclosure of all APIs and significant file formats, effectively allowing anyone to write replacement components and compatibility layers/wrappers for all things Windows. Such a scenario is most likely among M$ exectives' worst nightmares since it is one of the few capable of causing significant mid/long-term damage. (I vaguely remember reading a story of that nature some years ago...)
I wonder what else legislator will try against M$ on the next round, after they realize they fundamentally lost the first.
What I would be interested into is XP Pro Lite - only the OS and core GUI/management stuff with no multimedia junk beyond volume controls... I hate the way XP's setup simply installs practically everything, no questions asked. (Who actually uses the 50MB Windows Movie Maker?)
Seems like the standards for labeling stuff as "innovation" are sinking to new lows every day.
There is a simple work-around for these phishing exploits: never trust pop-ups - assuming they were not disabled in the first place. When in doubt, proceed through the official front-page.
Still does not change the fact that copyright infringement ("pirating") does not steal the copyright holders' privileges... it only infringes on them.
For copyright theft, I would have to trick the copyright owners into handing their rights over to me or impersonate them to have the registrations changed myself.
So copyright theft and copyright infringements are two very different things. Making copies does not magically transfer or duplicate copyright ownership.
The simplest form of ad-blocking is to simply disable third-party objects. Since most ads are served by off-site (advertiser) servers - most webmasters do not want to serve >500KB worth of ads per page from their own servers at their own cost - disabling third-party objects already kills most ads and "spyware" tracking cookies.
AdBlock itself is somewhat like a glorified old-school host list - it can be used to block hosts or anything else that can be defined as a regular expression.
Since simply disabling third-party objects already kills the vast majority of annoying ads (at least for me), I have not really bothered setting up AdBlock but I have it installed just in case.
How am I supposed to know in advance what kind of ads will be served by the ad servers? A given advertisement feed might contain only a few unreasonable ads but it takes only one to cause major inconvenience... and these feeds are usually dynamic, with some ads being retired and new ones inserted all the time. The only "safe" thing to do is to block them all.
To me, this is not significantly different from 30s/commercial skip on a VCR/TiVo/etc. With all the banner and pop-up training I had in the 90s, my mind can already blank off all but the most annoying ads.
Looks like you skipped the paragraph about "the only advertisements I still tolerate are google-style text-only ads and static images".
Filtering the ads is my way of telling advertisers and websites that abusive ads are unwanted. Freeloading might not be right but neither is the audio-visual abuse many advertisers use.
Example 1: a flash ad with sound popping up at 04h00 when the 100Wx2 (RMS) amplifier plugged to my PC is still on. This happened to me once and I scrapped Flash to make sure it would never happen again. (Until I learned about the likes of flashblock.)
Example 2: High-contrast, high-motion/jumpy Flash or animated GIF in or around an article make reading fairly painful - until I discovered FlashBlock&co, I used to either resize the browser and scroll to hide these or move a window on top of these ads.
Both cases are absolutely unacceptable. Silent static ads, preferably text-only, are the only ones I will tolerate - they're discrete, quiet and trying to filter them would have many undesirable side-effects.
Slashdot Mirror
In a game, how do you start working on the next frame before being done with the current frame? Sure, it could be done but it would introduce input-output lag equivalent to however deep that frame pipeline might be. At 30FPS with two pipelined frames, the perceived lag would go from 1/30s to 1/20s, which is often enough to be irritating.
The bulk of game computations is game logic and physics then adjusting scene geometry accordingly, all of which is strict frame-by-frame computing. Because of this, multi-threaded games can only truly benefit from multi-threading by distributing this intra-frame computing. (Noticed how SLI and CrossFire are also designed for distributed rendering to push single frames out faster with 60-70% gain instead of alternate frame rendering where it would be easier to approach 100%?)
I am a hardware guy myself, I like VHDL and programmable logic where I am guaranteed that everything happens on every clock. Right now, I am writing an application letter with the intent of applying for jobs along the lines of "Processor Architect" - I would love to be among the people who brought SMT to Athlon64 and I have a number of ideas about how to get there. I am much more confident about thinking in parallel on an RTL basis than in a software environment where I can never know with any certainty anything about what happened/is happening/will happen.
The above paragraph should answer your mindset wondering. In short: there are much fewer uncertainties on the hardware side of things.
There is no multiprocessor advantage to games now only because most games people look at are written fundamentally single-threaded, not using threads for any meaningful/heavy things. In this common case, the only potentially significant gain from SMT/MP for current games is offloading of background processes.
Most loops where each iteration is independant from previous results can be distributed. For bulk processing an object array (say a display list for games), a shared counter can be used to distribute objects across threads running "while((nObj = _InterlockedExchangeAdd(&sharedCount, -1)) >= 0) pObj[nObj]->Process();" - I am certain many loops would be suitable for parallelizing in a similar manner when per-object processing is sufficiently non-trivial to prevent excessive bus locking.
For the costs, I did a quick check and $10k in the upper range for US/Canada patents with $20k being the upper range for WIPO (Patent Cooperation Treaty - PCT) filings.
The main advantages of filing with WIPO is savings in translation costs, no need to follow each country's specific procedures and having to redundantly defend patent challenges for each individual registration. None of the FAQs I read give any specifics about PCT pricing but they do imply additional costs less than individual filings to get actual protection in each region/country.
Last time I heard, patent application was around $10k per country and around $100k for worldwide filing.
Filing a patent in any country prevents anyone else from filing the same patent in all others but only provides legal enforcement in the countries where the patent was was filed. In most cases, inventors can file in other countries as necessary to extend legal protection but doing it incrementally can quickly cost more than an international filing.
(This is what I was told at a conference about a year or two ago.)
In that case, a multimeter is all you need.
A basic desktop box should fall in the 100-150W range since decently priced low(-electrical)-power desktop chips are becoming a thing of the past.
My main PC is plugged on a Back-UPS XS1000 and the monitoring software tells me my idle P4-3G (with Radeon 9600XT, 1GB RAM, SB Audigy2, P4P800VM, 1xHDD) uses about 110W while idle, 160W under full CPU load and 180W under full system load. (HL2 + SETI)
My other PC with a P3-1G uses 120W while running SETI and 90W while idle... 15W for chipset/RAM, 15W for the idle CPU + VRM losses, 15W for the disk drives, 35W for fans/sound/network/video/etc., 70W sub-total, so I guess the PSU in that PC (TigerPro TP300, came with some CAN$45 case - I needed a case, the PSU was a bonus/piece-of-crap) has ~70% efficiency at 3.7A ripple current and the cheaper 'standard' ones are rated under 1A.)
Wow, that last bit was miles off-topic.
On one hand, people start talking about ways of preventing solar heat from reaching Earth to reduce global warming, on the other hand, people are getting ready to create power plants that will eventually generate gigawatts.
Well, fusion plants are necessary to replace coal/diesel/fission plants but global warming initiatives should simply focus on making everything more power-efficient and low-emission instead of coming up with inconvenient/impossible ideas. (Some european minister said he was shocked at how much power standby electronics use and suggested that people should not leave TVs&all plugged in when not in use... but how many people would unplug their VCR/TV/PC/etc. to save 5-20W per appliance given the inconveniences?)
10A at 10V and 100A at 1V is still 100W.
RAM densities, bus widths, working frequencies, etc. go up, this causes the current to increase and eventually cancel that gain, leading to a net increase in actual power.
DDR2 currently uses the same densities, comparable frequencies, etc. as DDR but DDR2's advantage will diminish as the gap between it and DDR grows. By the time we switch to DDR3, DDR2 will probably have exceeded DDR's power envelope, as DDR3 probably will by the time we move on to DDR4/FBDIMM/whatever.
And to complete this price comparison:
$600 - Athlon64 X2-4200
$250 - Pentium D-820 (2.8GHz)
So Intel's least expensive dual-core chip is less than half as expensive as AMD's. This is one of the first times where AMD substantially exceeded Intel pricing in any category.
AMD's implementation of dual cores is certainly more elegant in many ways and the A64 is usually better suited for most of what I do, my wallet says the 820 will be enough if I get dual-core itch.
Measuring wall power does not tell you much about the CPU's power. The only case where a multimeter would be useful for measuring actual CPU power is on motherboards where all CPU power is exclusively taken off the ATX12V connector. This figure would include VRM losses but at least excludes other 12V loads such as AGP/PCIE graphics.
Most of the numbers I quoted were from memory of articles and the actual spec/data-sheets a long time ago... have a look for yourself: http://www.intel.com/products/a_z/p_q.htm
Look at static current (no clock) figures, current chips have higher no-clock power than older chips under full load.
t s/24526408.pdf p.32)
My P233MMX runs fine at full-load without CPU fan, my P3-1G with its larger heatsink eventually crashes at idle. The P54C-233 has a max TDP of 17W (http://www.intel.com/design/intarch/mmx/mmx.htm) while the P3's SLEEP/HALT power is 10-12W. (Islp = 6.9A / Idslp = 6.6A @ 1.65-1.75V - ftp://download.intel.com/design/PentiumIII/datash
So, any P3 uses almost as much power while _(deep)sleeping_ as the hottest P54C does under full-load.
I cannot.
Instead of being above my ideal weight like 60% of industrialized population, I am (still) under my "ideal" weight and most of the weight I have been gaining is muscle. (My primary vehicle is a mountain bicycle... 100-250 kilometers/week depending on weather.)
Somehow, two industrial heavyweights fighting each other in court seems orders of magnitude more even than some relatively defenseless hobbits finding their way through ogre armies while avoiding the riders, the freaky eye, etc.
Why would the power bill go down?
P1 = 10-20W
P2 = 15-35W
P3 = 25-45W
P4 = 35-165W
Chipset and RAM power also increases across generations so a few more watts need to be added to each upgrade... and another extra in the 10-20% range for the extra VRM and PSU losses. (PWM regulator technology and components have not changed much over the last 10 years)
But yes, having a faster CPU/RAM does make a substantial difference in firewall responsiveness and throughput. When I upgraded my router from 100MHz to 200MHz, loopback throughput roughly tripled - from 660KB/s to 2.3MB/s. (On top of being slower, a slow chip also spends more of its time processing interrupts and background stuff, leaving less time for 'useful' work, double-hit. Seems like the 100MHz chip in this case was wasting something like half of its time on house-keeping stuff.)
2.3MB/s might not seem like much but I am not expecting local ISPs to offer >20Mbps (combined up+down) for another ~10 years... at least not under CAN$50/month.
Dropping the clause would not prevent copyright/patent laws from coming back in a form very similar to what we have now.
Instead of dropping it, it would have to be reworded in a more specific and restrictive way but even that would not stop industries from buying politicians and having the constitution tweaked in their favour.
Because politicians need cash to win the popularity contest called "elections" (only getting listed already costs a small fortune so only relatively wealthy people can afford it) and lobbyists are backed by large organizations that are full of it, ordinary people are fundamentally screwed.
That would not be terribly efficient.
Most rendering farms probably load all scene textures/geometry to allocated renderers then the workstation simply queue frame requests to each renderer, reducing the bandwidth requirement to little more than finished frames' (1920x1080x3*10/8 = 7.8MB/frame) transfer, assuming the scenes are fairly lenghty or the software is written to cache across scenes/jobs and the scenes use a substantial common texture/geometry base.
So the gigabytes of textures/geometry probably only apply while the rendering farm is doing a cold-start with clean caches. The rendering job's setup becomes even more bandwidth-trivial if the whole thing can be setup using multicast - setting up one or a hundred nodes at once takes practically no additionnal time in that case.
Only a month of testing/burn-in?
All of my three failed HDDs so far died after 10-18 months of 24/7 service.
So far, I have two (out of two) failed HDDs from Quantum, one failure and one oddly behaving drive (erratic performance) from Maxtor, none from WD and Seagate, though my only Seagate drive (2.1GB 4500RPM) sounds like a lawn mower - I was certain it would fail but it annoyingly hummed along 24/7 until I retired the PC in which it was 5+ years later.
Thankfully, all four failures occured under warranty. Also, failure in the three total failures was progressive: after each restart, I could copy some (decreasing) amount of data before drive lock-up, usually enough for me to recover my most important files before total failure.
Well, I might have a failed Seagate drive... but I do not have a matching PCI ESDI controller card to test it with.
The ability not to install unnecessary components should be standard... people should not have to use convoluted tricks for this. With Windows 3.xx, almost everything was optionnal and the list kept shrinking with each version until it vanished with XP.
Next step would be a one-click (EULA) install where the installer unilaterally decides to fdisk the first HDD as a single NTFS partition and install everything there - this is all ok since it was agreed to by clicking through the EULA...
The EU's sanctions against M$ were somewhat more severe than the USA's but both are ultimately misguided. For M$, fines are only ghost blips on the accounting radar.
The only thing I can think of that would have been able to do a decent job of enabling fair(er) competition is full, free and open disclosure of all APIs and significant file formats, effectively allowing anyone to write replacement components and compatibility layers/wrappers for all things Windows. Such a scenario is most likely among M$ exectives' worst nightmares since it is one of the few capable of causing significant mid/long-term damage. (I vaguely remember reading a story of that nature some years ago...)
I wonder what else legislator will try against M$ on the next round, after they realize they fundamentally lost the first.
XP Home vs XP Pro is all the degradation I need.
What I would be interested into is XP Pro Lite - only the OS and core GUI/management stuff with no multimedia junk beyond volume controls... I hate the way XP's setup simply installs practically everything, no questions asked. (Who actually uses the 50MB Windows Movie Maker?)
Seems like the standards for labeling stuff as "innovation" are sinking to new lows every day.
There is a simple work-around for these phishing exploits: never trust pop-ups - assuming they were not disabled in the first place. When in doubt, proceed through the official front-page.
Still does not change the fact that copyright infringement ("pirating") does not steal the copyright holders' privileges... it only infringes on them.
For copyright theft, I would have to trick the copyright owners into handing their rights over to me or impersonate them to have the registrations changed myself.
So copyright theft and copyright infringements are two very different things. Making copies does not magically transfer or duplicate copyright ownership.
The simplest form of ad-blocking is to simply disable third-party objects. Since most ads are served by off-site (advertiser) servers - most webmasters do not want to serve >500KB worth of ads per page from their own servers at their own cost - disabling third-party objects already kills most ads and "spyware" tracking cookies.
AdBlock itself is somewhat like a glorified old-school host list - it can be used to block hosts or anything else that can be defined as a regular expression.
Since simply disabling third-party objects already kills the vast majority of annoying ads (at least for me), I have not really bothered setting up AdBlock but I have it installed just in case.
How am I supposed to know in advance what kind of ads will be served by the ad servers? A given advertisement feed might contain only a few unreasonable ads but it takes only one to cause major inconvenience... and these feeds are usually dynamic, with some ads being retired and new ones inserted all the time. The only "safe" thing to do is to block them all.
To me, this is not significantly different from 30s/commercial skip on a VCR/TiVo/etc. With all the banner and pop-up training I had in the 90s, my mind can already blank off all but the most annoying ads.
Looks like you skipped the paragraph about "the only advertisements I still tolerate are google-style text-only ads and static images".
Filtering the ads is my way of telling advertisers and websites that abusive ads are unwanted. Freeloading might not be right but neither is the audio-visual abuse many advertisers use.
Example 1: a flash ad with sound popping up at 04h00 when the 100Wx2 (RMS) amplifier plugged to my PC is still on. This happened to me once and I scrapped Flash to make sure it would never happen again. (Until I learned about the likes of flashblock.)
Example 2: High-contrast, high-motion/jumpy Flash or animated GIF in or around an article make reading fairly painful - until I discovered FlashBlock&co, I used to either resize the browser and scroll to hide these or move a window on top of these ads.
Both cases are absolutely unacceptable. Silent static ads, preferably text-only, are the only ones I will tolerate - they're discrete, quiet and trying to filter them would have many undesirable side-effects.