"Note that the difference between the special compute hardware ("Tesla" and "Firestream") and consumer cards tends to be that they have a little more memory, and are enormously more expensive , so the consumer cards are way ahead in terms of FLOPS per dollar. "
But if you need 64-bit performance the consumer cards are crippled - there is a factor of 2-4 difference compared to the Teslas and Quadros. Only the Teslas have ECC, which may be essential for some applications. The extra memory on the pro series cards makes a big difference for many applications, too. 6GB is a lot more than 2GB. Since accessing the main memory rather than the GPU memory is so much slower and splitting tasks across cards on different nodes is such a bitch, the extra expense for a card that can fit the whole problem in its own memory may be worth 5x the cost for 3x the the memory and 2x the 64-bit performance. In a cluster where you plan to split applications across multiple graphics cards on different nodes despite all the hassle entailed, that may not be enough of a selling point, though. If 32bit FLOPS are the only metric, then consumer cards can't be beat.
I'll assume you know more about this than me, but he did say that the nodes are going to be wired with 4x GigE. Might there be a penalty bridging from that to IB rather than 10GigE?
Anyway, to get low latency those GigE links to the nodes need to be optimized. I thought this was interesting:
High performance network technologies such as InfiniBand use a kernel by-pass method to improve performance. This capability is also available for Ethernet, but is not widely used outside of the HPC community. One such methodology is Intel® Direct Ethernet Transport (DET), which works by providing a User Direct Access Programming Library (uDAPL) interface like InfiniBand. uDAPL defines a single set of user APIs for all Remote Memory Direct Access (RDMA)-capable transports. DET includes a kernel module and an uDAPL library for Ethernet and will work on almost any Ethernet NIC. It can be linked with any software requiring a uDAPL library, such as an MPI version.
Another popular kernel by-pass effort is the Open-MX project. Open-MX is based on the Myrinet MX protocol. Essentially, any software that links to the Myricom MX library should be able to link with Open-MX. Currently, Open MPI, MPICH2, and the PVFS2 file system have all been shown to work with Open-MX. While Open-MX will work with almost all GigE and 10-GigE chip-sets without modifying drivers, it does require kernel 2.6.15 or higher to work. Depending on the chip-set Open-MX latencies as low as 10 seconds for GigE have been reported.
For 10GigE here's a recent low-latency benchmark: Audited STAC-M2 Benchmark of IBM LLM on an IBM-BNT G8264 switch, using IBM x3550 servers and Mellanox MNPH29C-XTR ConnectX®-2 EN with RoCE "Using standard Ethernet and RoCE protocols, at the base message rates set by the specs, the mean latency of the solution did not exceed 7 microseconds, while standard deviation of latency was measured at 1 microsecond. At the highest tested rate of 2.3 million messages/second, the mean latency of the solution was just 13 microseconds while the standard deviation of latency was measured at 2 microseconds."
Chelsio claims 3 microsecond latency using RDMA over 10G Ethernet on their "T4" model: "Chelsio T4 Unified Wire adapters can run iWARP RDMA, TCP, iSCSI and FCoE simultaneously with full offload and deliver full wire speed throughput and extremely low latency between the computing nodes, the storage resources, and the user and cluster management nodes in any HPC environment." Not sure how much that really costs compared to IB, though. They also say:
"Since IB lacks congestion management and adaptive routing, it quickly hits hot spots even in clusters of moderate size. iWARP over Ethernet, in contrast, achieves reliability via TCP, which results in a lower effective latency for useful applications."
*"10Gb IB link is effectively 8Gb. Furthermore, InfiniBand cards, like Ethernet cards, are limited by PCIeGen2 x8. Independently of how many 10Gb or 40Gb ports an adapter exposes, the aggregate bandwidth is limited to about 26Gbps in each direction. Therefore, Chelsio’s T4 based adapters and the leading IB adapters offer the SAME bandwidth."
*"Ethernet switch port prices have reached parity. The same can be said about adapter prices. However, an IB cluster further requires an Ethernet switch for management, a gateway for routing, and expensive IB storage available from a limited set of suppliers, as well as specialized IT personnel."
Right. And the Korean Internet is essentially a closed system anyway - if you want to see pages in Korean, they're almost always going to be made and hosted in Korea by a powerful local conglomerate, and closely regulated by a clueless authoritarian bureaucracy.
It takes several years for the sites that are big in the US to catch on in Korea, with Koreans scoffing all the way, supporting local sites as a matter of national pride, until suddenly the market shifts - and seemingly in a matter of weeks, hardly anyone uses the Korean site and the US site is the only cool thing to use.
I remember people back then laughing at Google's increased vigor in making headway in the Korean market, since Naver will NEVER lose its dominance, right? People laughed and said Google overly-American dedication to simplicity would NEVER go over in Korea, and its little attempts at coloring up the Google Korea site with cute little animations was something akin to pitiful. People also laughed when YouTube Korea rolled out, swearing up and down that they'd be crushed by the better resolution of the domestic sites. Those techheads missed the point.
I've been saying, since that post, as well as in "The Mis-execution of Korean UCC" in April 2007, when I continued criticizing the utter lack of content on Korean UCC, that the Korean Internet is woefully devoid of ideas, and pitifully cordoned off from the rest of the world. Like the Korean economy, this is not just a side-effect, but totally intentional and a key part of how it succeeds.
In the 2006 post, I called the Korean Internet a "jaebeol" system, which I very much think it is. Inherent in the system isn't a core of intensely creative people and ideas that find expression in a myriad different ways, that come together and combine in unpredictable ways.
For instance, who knew how the synergy between blogging, embeddable video, other social media such as Twitter, social bookmarking, and other things would come together? It's led to new forms of media, business models, and ways of disseminating information itself. It's changed pop culture, politics, and so many other fields in a real -- not gimmicky -- way.
The Korean Internet? With broadband that leaves the US in the dust, no real problem of a "digital divide", computers everywhere, a high degree of technical skill with all kinds of programs and electronic devices, and a youth culture that is deeply socially invested in the Internet -- where's the beef? Meaning -- where's the content?
Where are the new ideas? Why didn't Koreans invent YouTube, Digg, or Twitter, or even the concepts of blogging, podcasting, or social bookmarking? Where are the funky new business ideas, new revenue models, or even (and especially) THE CONTENT?
Another good article of his on the subject: Facebook Taking Over Korea, as Predicted!"I made a bet with my 'Korean Wave and Media' class that I teach at Myongji University that Facebook would inevitably take over the Korean Internet. Whoa, they said. That's crazy. Nothing can beat Cyworld."
"is a Turing test valid if the human is an idiot?"
What about the humans in the control group who failed the test? Maybe some of them were flunked by idiots making the judgement, but likely many of them really were indistinguishable from bots. Given that this test was done at a tech convention in India, I personally suspect that most of the 36.7% of humans who flunked the test work in call centers. I've certainly had a few on the line that were indistinguishable from a chatbot running on a Speak & Spell, and were certainly quite as useless as a very useless thing indeed.
The journalist was a damned fool for not realizing that a file on the web could be downloaded by anybody, and in the case of a Wikileaks server, any file there for more than a minute or two was virtually certain to be downloaded by a number of intelligence agencies and shady characters on the off chance that the key would eventually turn up. Or, in little words for little minds: data can be copied. Publicly posted data can be copied by anybody. When potential enemies might have the encrypted text, releasing the password is the same as releasing the data.
The reporter should have known this, yet he published the exact password, with absolutely no reason to do so. He could have just described it in general terms, but he and he alone intentionally ran what he should have known was a substantial risk of releasing the cables. This release happened, and it would not have happened but for his reckless,stupid, pointless action. The reporter alone is responsible for this release and any consequences it may have.
"...are *still* way above what was managed with the $200 machine" No, they aren't. TFA price did not include display, OS, any way of obtaining or loading the OS, keyboard, mouse, camera, microphone,... it wasn't really a working computer, and worse, with only 2GB RAM, it's barely even web-capable. My 2-year-old Lenovo dual-core laptop was $400 with all that listed above, DVD-RW, SD slot, 3GB RAM, Wi-Fi, multi-hour UPS, etc. (though admittedly only 250GB HDD rather than the 500GB of TFA.) You would have to spend at least another $200 over the TFA's setup to get just the stuff that I use every day on my laptop.
The base Raspberry Pi is not a working computer in any sense of the word. It has no mass storage at all (does not come with an SD card), no OS, no power supply, no cables, no I/O devices, no case, no documentation. All it is is a motherboard with no slots and an insufficient number of USB ports. Getting what is needed to make it work at all will cost at least another $35 in money or time, effort and travel. A full setup could easily cost $275, buying the lowest-cost new components, plus $250/year+ for internet service. A decent used setup would still run well over $100. (But for that same money one could get a comparable used x86 computer system.)
No, it's bad but not that bad. PAL is 625 by about 700 or so. That's reduced a bit by imprecision in the electronics and at the edges by the frame, but it's still at least 550 lines. NTSC is about 15% less after deductions. 640x480 should work reasonably well on either. Both are interlaced, though.
A decent ~20" 1920x1080p LCD TV can be had for less than $150, shipping included, though.
No, there are several other, more selective IQ clubs, including the Mega Society, (the supposedly 1 in a million IQ club) with a substantial fraction of members who can't/won't use their brains and don't get on well with anybody. (K.L., I'm thinking of you!)
The RAM is package on package, and the bigger memory parts aren't available yet. Yes, the $25 version is pointless when the $35 version is so much more capable, but the real minimum price needs to include shipping, power supply, SD card with loaded OS, keyboard, mouse and video cord at an absolute minimum. That about doubles the price to $70 even without a case, network, display, or other peripherals.
The truth is that it is hard to round up all the things you need to really have a functioning Raspberry Pi system. The total system cost if you buy the cheapest new gear is about $150, or $250-$275 with a 1080p TV. (Old TVs would be much inferior, and often unsuitable due to size and weight.) One could easily spend 40-80% more at a local big box store or on padded shipping fees for the same quality of gear and much more even than that for better quality gear. Some of that you could find used, but it wouldn't be much cheaper at a used computer store (if you even have one where you live), and if you try to get it from individuals with used extra equipment, you will have to somehow have to find those individuals and likely make several trips over a period of time to pick the gear up. This will take several hours minimum of hunting, calling and driving. If a school starts requiring them for classes for hundreds of students, these resources will be depleted.
None of the following is supplied:
Needed: Decent TV USB mouse, keyboard, hub SD card, min. 4 GB, preferably 8GB, loaded with compatible OS HDMI and/or component video cable Power supply - 6-20V, 300ma+, with the right connector Plastic custom enclosure (not yet made as the dimensions and port locations of the final card have not been released yet)
(Not to mention a good place to put the computer in the home, preferably with a desk and chair but not interfering with other family uses.)
Desirable: Ethernet hub or WiFi hub/bridge and/or USB WiFi receiver Ethernet cables Internet service, internet access device with an Ethernet or WiFi bridge Headphones and/or speakers USB microphone Spare/backup SD card
Plus shipping for all of the above, and the Raspberry Pi itself.
All that adds up - the cost of the Raspberry Pi itself is only a small fraction of what is needed. The minimum expenses to a school system to set up a classroom would be much higher - likely in the $500/seat range.
Their own internal documents show AT&T does not need T-Mobile to expand service, and that AT&T intends to raise prices. This is a deal that should not happen. At last the DOJ does something right on the merger front.
By making more until their parts inventory runs out, they can get rid of their otherwise useless parts inventory for only a 70% loss instead of a 95% loss. The cost of the parts to HP was likely more than half of the initial $300 price for the tablet. At the same time it makes a bunch of new customers happy who are now much more likely to buy other products on which HP actually makes money. These new customers are disproportionately of the type that has a great deal of influence over tech buying decisions both at work and socially. It's really one of the best targeted marketing campaigns HP could have devised, though I doubt it was intentional.
Thing is, this device just can't be sold for anything less than $200, even at a nominal parts + labor + packaging accounting of cost. How big is the market at that price? A lot smaller than at $100. And they really need more than that to break even after design, support etc. are covered. HP needs to re-run the numbers again to see if the loss-leader inadvertent investment in market share has changed the playing field enough to try doing a low-end tablet, but I wouldn't be surprised if the answer turned out to be "no".
Yes, they cut losses on components in inventory with no other use to HP than the tablet (surplus component markdowns are often 90-99%) and HP gets some really good advertising value and some desperately needed goodwill as a bonus.
Geometric algebra, that is, real-valued Clifford algebras. It has some similarities to differential forms, but has advantages. See my post lower on this page with links to resources.
The four states of matter (as we phrase it today) are solid, liquid, gas and plasma, which map directly to the ancient conceptions of earth, water, air and fire. Just because we changed the meaning of "element" doesn't make the ancients wrong.
I agree that imaginary/complex numbers are not needed per se, they mostly represent rotation or oscillation, but I don't think they are all that good as simplifying notation; rather, they obscure physical intuition. The same is true of most applications of matrices. (The Pauli matrices are particularly bad notation.) Geometric algebra can replace both complex numbers and matrices (and quaternions and tensors) in most situations while giving much better physical intuition. (Writing a library to handle geometric algebras invariably reverts to lots of matrix math under the hood, though.) (See the long post I made earlier today for links to GA references and resources.)
I forgot to mention why the idea that velocity = rotation gives insight. Imagine rotating a stick from vertical to horizontal. As the height decreases, the length increases. Rotating in the xt plane, as the length decreases, the duration increases. So the phenomena of relativistic length contraction and time slowing are really exactly the same thing. As length contracts as you go faster, time slows down proportionally.
Slashdot Mirror
"Note that the difference between the special compute hardware ("Tesla" and "Firestream") and consumer cards tends to be that they have a little more memory, and are enormously more expensive , so the consumer cards are way ahead in terms of FLOPS per dollar. "
But if you need 64-bit performance the consumer cards are crippled - there is a factor of 2-4 difference compared to the Teslas and Quadros. Only the Teslas have ECC, which may be essential for some applications. The extra memory on the pro series cards makes a big difference for many applications, too. 6GB is a lot more than 2GB. Since accessing the main memory rather than the GPU memory is so much slower and splitting tasks across cards on different nodes is such a bitch, the extra expense for a card that can fit the whole problem in its own memory may be worth 5x the cost for 3x the the memory and 2x the 64-bit performance. In a cluster where you plan to split applications across multiple graphics cards on different nodes despite all the hassle entailed, that may not be enough of a selling point, though. If 32bit FLOPS are the only metric, then consumer cards can't be beat.
I'll assume you know more about this than me, but he did say that the nodes are going to be wired with 4x GigE. Might there be a penalty bridging from that to IB rather than 10GigE?
Anyway, to get low latency those GigE links to the nodes need to be optimized. I thought this was interesting:
(From The Ethernet Cluster
For 10GigE here's a recent low-latency benchmark:
Audited STAC-M2 Benchmark of IBM LLM on an IBM-BNT G8264 switch, using IBM x3550 servers and Mellanox MNPH29C-XTR ConnectX®-2 EN with RoCE
"Using standard Ethernet and RoCE protocols, at the base message rates set by the specs, the mean latency of the solution did not exceed 7 microseconds, while standard deviation of latency was measured at 1 microsecond. At the highest tested rate of 2.3 million messages/second, the mean latency of the solution was just 13 microseconds while the standard deviation of latency was measured at 2 microseconds."
Chelsio claims 3 microsecond latency using RDMA over 10G Ethernet on their "T4" model: "Chelsio T4 Unified Wire adapters can run iWARP RDMA, TCP, iSCSI and FCoE simultaneously with full offload and deliver full wire speed throughput and extremely low latency between the computing nodes, the storage resources, and the user and cluster management nodes in any HPC environment." Not sure how much that really costs compared to IB, though.
They also say:
Right. And the Korean Internet is essentially a closed system anyway - if you want to see pages in Korean, they're almost always going to be made and hosted in Korea by a powerful local conglomerate, and closely regulated by a clueless authoritarian bureaucracy.
It takes several years for the sites that are big in the US to catch on in Korea, with Koreans scoffing all the way, supporting local sites as a matter of national pride, until suddenly the market shifts - and seemingly in a matter of weeks, hardly anyone uses the Korean site and the US site is the only cool thing to use.
Michael Hurt is one of the most insightful analysts of the Korean media and cultural scene. His blog "Scribblings of The Metropolitician" is well worth reading. Here's a brief section from one of his main posts about the Internet in Korea:
Some More Thoughts On the Korean Internet and Cultural Barriers to Content Production
Another good article of his on the subject:
Facebook Taking Over Korea, as Predicted!"I made a bet with my 'Korean Wave and Media' class that I teach at Myongji University that Facebook would inevitably take over the Korean Internet. Whoa, they said. That's crazy. Nothing can beat Cyworld."
"is a Turing test valid if the human is an idiot?"
What about the humans in the control group who failed the test? Maybe some of them were flunked by idiots making the judgement, but likely many of them really were indistinguishable from bots. Given that this test was done at a tech convention in India, I personally suspect that most of the 36.7% of humans who flunked the test work in call centers. I've certainly had a few on the line that were indistinguishable from a chatbot running on a Speak & Spell, and were certainly quite as useless as a very useless thing indeed.
The journalist was a damned fool for not realizing that a file on the web could be downloaded by anybody, and in the case of a Wikileaks server, any file there for more than a minute or two was virtually certain to be downloaded by a number of intelligence agencies and shady characters on the off chance that the key would eventually turn up. Or, in little words for little minds: data can be copied. Publicly posted data can be copied by anybody. When potential enemies might have the encrypted text, releasing the password is the same as releasing the data.
The reporter should have known this, yet he published the exact password, with absolutely no reason to do so. He could have just described it in general terms, but he and he alone intentionally ran what he should have known was a substantial risk of releasing the cables. This release happened, and it would not have happened but for his reckless,stupid, pointless action. The reporter alone is responsible for this release and any consequences it may have.
Didn't 6 digit UIDs start around 1999? 7 digit about 3 or four years ago?
"...are *still* way above what was managed with the $200 machine"
No, they aren't. TFA price did not include display, OS, any way of obtaining or loading the OS, keyboard, mouse, camera, microphone,... it wasn't really a working computer, and worse, with only 2GB RAM, it's barely even web-capable. My 2-year-old Lenovo dual-core laptop was $400 with all that listed above, DVD-RW, SD slot, 3GB RAM, Wi-Fi, multi-hour UPS, etc. (though admittedly only 250GB HDD rather than the 500GB of TFA.) You would have to spend at least another $200 over the TFA's setup to get just the stuff that I use every day on my laptop.
The base Raspberry Pi is not a working computer in any sense of the word. It has no mass storage at all (does not come with an SD card), no OS, no power supply, no cables, no I/O devices, no case, no documentation. All it is is a motherboard with no slots and an insufficient number of USB ports. Getting what is needed to make it work at all will cost at least another $35 in money or time, effort and travel. A full setup could easily cost $275, buying the lowest-cost new components, plus $250/year+ for internet service. A decent used setup would still run well over $100. (But for that same money one could get a comparable used x86 computer system.)
No, it's bad but not that bad. PAL is 625 by about 700 or so. That's reduced a bit by imprecision in the electronics and at the edges by the frame, but it's still at least 550 lines. NTSC is about 15% less after deductions. 640x480 should work reasonably well on either. Both are interlaced, though.
A decent ~20" 1920x1080p LCD TV can be had for less than $150, shipping included, though.
Sorry, that should be composite video, not component video.
The Raspberry Pi does not have a VGA port. HDMI or component (RCA jack) video only.
"They're the World's smartest idiots IMO."
No, there are several other, more selective IQ clubs, including the Mega Society, (the supposedly 1 in a million IQ club) with a substantial fraction of members who can't/won't use their brains and don't get on well with anybody. (K.L., I'm thinking of you!)
The RAM is package on package, and the bigger memory parts aren't available yet. Yes, the $25 version is pointless when the $35 version is so much more capable, but the real minimum price needs to include shipping, power supply, SD card with loaded OS, keyboard, mouse and video cord at an absolute minimum. That about doubles the price to $70 even without a case, network, display, or other peripherals.
With what? The computer you already have all to yourself as a poor 8th grader? I think you are kind of missing the point of this device.
The truth is that it is hard to round up all the things you need to really have a functioning Raspberry Pi system. The total system cost if you buy the cheapest new gear is about $150, or $250-$275 with a 1080p TV. (Old TVs would be much inferior, and often unsuitable due to size and weight.) One could easily spend 40-80% more at a local big box store or on padded shipping fees for the same quality of gear and much more even than that for better quality gear. Some of that you could find used, but it wouldn't be much cheaper at a used computer store (if you even have one where you live), and if you try to get it from individuals with used extra equipment, you will have to somehow have to find those individuals and likely make several trips over a period of time to pick the gear up. This will take several hours minimum of hunting, calling and driving. If a school starts requiring them for classes for hundreds of students, these resources will be depleted.
None of the following is supplied:
Needed:
Decent TV
USB mouse, keyboard, hub
SD card, min. 4 GB, preferably 8GB, loaded with compatible OS
HDMI and/or component video cable
Power supply - 6-20V, 300ma+, with the right connector
Plastic custom enclosure (not yet made as the dimensions and port locations of the final card have not been released yet)
(Not to mention a good place to put the computer in the home, preferably with a desk and chair but not interfering with other family uses.)
Desirable:
Ethernet hub or WiFi hub/bridge and/or USB WiFi receiver
Ethernet cables
Internet service, internet access device with an Ethernet or WiFi bridge
Headphones and/or speakers
USB microphone
Spare/backup SD card
Plus shipping for all of the above, and the Raspberry Pi itself.
All that adds up - the cost of the Raspberry Pi itself is only a small fraction of what is needed. The minimum expenses to a school system to set up a classroom would be much higher - likely in the $500/seat range.
Their own internal documents show AT&T does not need T-Mobile to expand service, and that AT&T intends to raise prices. This is a deal that should not happen. At last the DOJ does something right on the merger front.
By making more until their parts inventory runs out, they can get rid of their otherwise useless parts inventory for only a 70% loss instead of a 95% loss. The cost of the parts to HP was likely more than half of the initial $300 price for the tablet. At the same time it makes a bunch of new customers happy who are now much more likely to buy other products on which HP actually makes money. These new customers are disproportionately of the type that has a great deal of influence over tech buying decisions both at work and socially. It's really one of the best targeted marketing campaigns HP could have devised, though I doubt it was intentional.
Thing is, this device just can't be sold for anything less than $200, even at a nominal parts + labor + packaging accounting of cost. How big is the market at that price? A lot smaller than at $100. And they really need more than that to break even after design, support etc. are covered. HP needs to re-run the numbers again to see if the loss-leader inadvertent investment in market share has changed the playing field enough to try doing a low-end tablet, but I wouldn't be surprised if the answer turned out to be "no".
Yes, they cut losses on components in inventory with no other use to HP than the tablet (surplus component markdowns are often 90-99%) and HP gets some really good advertising value and some desperately needed goodwill as a bonus.
Geometric algebra, that is, real-valued Clifford algebras. It has some similarities to differential forms, but has advantages. See my post lower on this page with links to resources.
I'm not saying no math, just not that formalism. The symbols are just a shorthand for the geometry, which is the actual content of GR.
The four states of matter (as we phrase it today) are solid, liquid, gas and plasma, which map directly to the ancient conceptions of earth, water, air and fire. Just because we changed the meaning of "element" doesn't make the ancients wrong.
I agree that imaginary/complex numbers are not needed per se, they mostly represent rotation or oscillation, but I don't think they are all that good as simplifying notation; rather, they obscure physical intuition. The same is true of most applications of matrices. (The Pauli matrices are particularly bad notation.) Geometric algebra can replace both complex numbers and matrices (and quaternions and tensors) in most situations while giving much better physical intuition. (Writing a library to handle geometric algebras invariably reverts to lots of matrix math under the hood, though.)
(See the long post I made earlier today for links to GA references and resources.)
Ha, I linked to that above, and to your GA survey. Nice to see you on /. !
I forgot to mention why the idea that velocity = rotation gives insight. Imagine rotating a stick from vertical to horizontal. As the height decreases, the length increases. Rotating in the xt plane, as the length decreases, the duration increases. So the phenomena of relativistic length contraction and time slowing are really exactly the same thing. As length contracts as you go faster, time slows down proportionally.