Good idea. But you could add a few things to that "contract":
All knowledge derived from this course remains the property of the Professor. Any use ouside of class must be cleared in advance in writing with the Professor, for a licence fee to be negotiated. Any derivitive ideas must be disclosed to the Professor and such derivitive works become the sole property of the Professor. In no case shall any putative "fair use rights" extend beyond any arrangement of two successive words uttered by the Professor. Any disparagement of, or unlcenced or piratical use of any intellectual property of the Professor by the student shall be compensated by immediate liquidated damages paid by the student to the Professor equal to twice the lifetime income of the student plus interest and the title to the student's intellectual property if any, his or her genetic code, and all derivitive works.
Hmm... that seems a little soft - I think I'm leaving something out..
I read http://www.uclaprofs.com/profs/kellner.html, the "Radical of the Week" and I thought they made him seem like a great guy. They thoroughly skewered their own viewpoint and revealed how mindless their politics is.
Excuse me, but there are still professors being hired at top universities BECAUSE they are enthusiastic, fire-breathing, Joe Stalin-supporting Marxists. There are still excellent teachers and researchers who are denied tenure because they are not Marxists or deconstructionists, radical feminists or race-baiting Mugabe wannabes.
It would be fair to disparage this just effort to expose these vermin to the light of day with the allegation that they are pursuing only the leftists at UCLA, were it not for the fact that the right wing at UCLA is virtually nonexistent. If they go after just one right-winger at UCLA, they will have gotten half of them; if they go after a hundred tenured Maoist freaks, it'll just give them wriggles of joy to have been persecuted at last.
If we don't limit this to American citizens, or any particular kind of US government employee killing innocent people in any particular kind of way, then it is clear that Bush's actions in Iraq are orders of magnitude worse than the crimes of in the Clinton administration.
I don't know where you went to school, but the idea that professors are generally interested in "open intellectual inquiry" is sheer fantasy. If you have political disagreements with a humanities professor, your grade goes down in nine out of ten cases, particularly if you argue well.
"Liberals" in academia have no tolerance for dissent with the party line on racial equality of abilities, female superiority, white male culpability, or the insight of major incoherent French literary theorists. Nor will they tolerate dissent with their conservative values of professorial dominance over instructors, students and staff, within- and between- school hierarchy, and their faith that drawing a practical distinction between education and the educational establishment's credentials is seditious.
Even so-called scientists have a hard time looking at evidence objectively on controversial topics; in the humanities, the truth is whatever the professor says it is. Debunk any faulty argument or fabricated statistic that anyone uses to support any sacred-cow cause and you will be branded a heretic. Facts are irrelevant - if the opinion is "hurtful" the student will be punished one way or another - lower grades, denial of opportunities, ostracism, and discipliary measures.
Academia is really one of the most rigid cultures in the world, with many tabus, caste structures, distinct in- and out- groups, irrational rituals and shibboleths. Putting a social check on the excesses of professors by reminding them that their classes are in the public sphere of a wider culture than academia is a good idea which may in the future prevent or at least mitigate some of the abuses that professors have often committed against students in the past.
Yes there were good reasons to cut off the Japanese, but those were secondary to FDR's reasons for doing so. The main reason was to get into the war when FDR had repeatedly promised the American people that the US would not join in the war unless attacked. The only way to get around that was to invite an attack and make sure that it was not foiled.
Your other points are irrelevant in view of the information provided in the page to which I linked. You say "a link to a geocities site is not a very convincing argument ". True, but the link is not the argument, the argument is given on the page to which I linked, which you have clearly not read, apparently preferring to sneer at the label rather than take the medicine.
The facts and the law determine the answer to the factual question of guilt. Courts do not create guilt, they measure it. The presumption of innocence does not make the accused person actually innocent, it is a rule for how to measure guilt with as low a false positive rate as possible. [Oddly for such a fundamental principle, the presumption of innocence is not explicitly in the Constitution, but is inferred from amendments IV, V, VI, IX, and XIV.]That FDR died before he could be tried in no way affects the factual question of his guilt.
Treason is defined in Article III Section 3, US Constitution: "Treason against the United States, shall consist only in levying war against them, or in adhering to their enemies, giving them aid and comfort. No person shall be convicted of treason unless on the testimony of two witnesses to the same overt act, or on confession in open court."
FDR has been proven to have allowed the Japanese to attack when he had the information and means to prevent the attack or mitigate the damage of the attack. Top Secret Army Board Report, Oct, 1944: "Now let us turn to the fateful period between November 27 and December 6, 1941. In this period numerous pieces of information came to our State, War, and Navy Departments in all of their Top ranks indicating precisely the intentions of the Japanese including the probable exact hour and date of the attack. " The Army board had more information available to it than any other inquiry into the matter, and certainly met the requrement for witnesses to overt acts. It is reasonable to suppose that a court at the time, if it had access to the intercepts on which FDR was briefed twice daily (the vast majority of which are still asserted by the NSA to be too potentially damaging to national security to release even 65 years later), would have found FDR guilty of treason.
Acording to the articles, SRAM uses 6 transistors per bit and DRAM uses 1 capacitor and one transistor, while ZRAM uses just one transistor, so ZRAM is indeed supposed to be twice as dense as DRAM. The transistors scale down better than capacitors, making new process scales more easily and cheaply achievable, so ZRAM is likely to replace DRAM in main memory. Memory is a tough, low-margin business where costs per bit depend mostly on bits/area, so doubling density is something that competitive manufacturers have to do whenever it becomes practical. Memory manufacturers aren't using SOI much yet, though, so it may take a year or two to get this to market in a memory-only chip. SOI costs 15% more per raw wafer and there is a learning curve, but even so, the cost/bit reduction from ZRAM should be 40%-55% - greater than any other new tech since the current 1T 1C DRAM cell was adopted thirty years ago.
It looks like there are no speed tradeoffs and it scales even better than DRAM, proven at 45 nm and suitable for 22 nm as well. AMD says such improvements usually take two years to show up in products, so that will still be 65 nm mostly with some 45nm. At 45 nm scale, a 1 cm^2 chip that is 70%-75% ZRAM would have about 48 MB.
It's a single-transistor capacitorless memory cell using the "floating body effect" of silicon on insulator (SOI) devices. Presumably stored charge in the gate affects the operation of the transistor in a way that can be used to store and read a bit, but I didn't feel like registering to read the white paper. The new memory should be six times denser than SRAM and twice as dense as DRAM.
There is no doubt that FDR knew about the Japanese plan to attack Pearl Harbor from hundreds if not thousands of monitored Japanese communications as well as highly placed sources from several different countries. In fact he wanted war so much he actually encouraged and abetted the attack, freezing Japanese funds in the US, cutting off oil to Japan, and feeding false information to Hawaii about the location of the Japanese fleet. There is more, but it could take a book to do it justice. Take a look at http://www.geocities.com/Pentagon/6315/pearl.html . FDR was guilty of treason.
[OK, this time in the right place - sorry, GP poster]
You do know there are no binaries on Google groups, right? And you do know that there are groups beginning with letters other than "alt", right? Like "sci" and "comp" and so forth, right? Or are you really that big a flaming turd?
You do know there are no binaries on Google groups, right? And you do know that there are groups beginning with letters other than "alt", right? Like "sci" and "comp" and so forth, right? Or are you really that big a flaming turd?
Article I Section 8. The Congress shall have power...
To declare war, grant letters of marque and reprisal, and make rules concerning captures on land and water;
To raise and support armies, but no appropriation of money to that use shall be for a longer term than two years;
To provide and maintain a navy;
To make rules for the government and regulation of the land and naval forces;
To provide for calling forth the militia to execute the laws of the union, suppress insurrections and repel invasions;
To provide for organizing, arming, and disciplining, the militia, and for governing such part of them as may be employed in the service of the United States, reserving to the states respectively, the appointment of the officers, and the authority of training the militia according to the discipline prescribed by Congress;...
A lot of people buy boxed distros, even when they could download. It's a lot to download, and in the case of a bootleg, who knows what hacks the uploader added? Anyway, a certain amount of piracy broadens the market, and they run a high risk anyway of getting their DRM cracked and OSX uploaded anyway. Once that happens they might as well accept the money of the people who would prefer to get Intel-version OSX from Apple directly.
Yep. Brilliant. Beyond that, if they relaease a boxed Intel OSX that will run on non-Apple hardware (no Apple support, obviously), they could sell perhaps 4x - 8x as many units as they currently sell computers. Each of those units would be 90%+ profit, and at say $100 - $200 each would likely give them more profit than the rest of their computer business. Each such "unauthorized" unit would be like an ad for their hardware, too, and those using OSX on generic hardware would want to move up to the social status, reliability, clean design and support of having a real Mac. Those who already have Mac hardware are mostly too loyal and hassle-avoiding to move to the PC hardware, and even if some do, the potential new market is so much bigger that overall it's a huge win for Apple. Five years out, Apple could be on a third of computers or more, but since many of those will also have a Microsoft OS, it won't take nearly that much business away from Microsoft.
So having a lower market share than Dell in the slowly-growing computer market is actually better for forward-looking Apple investors - much more room for growth than Dell, particularly when Apple is selling machines which have superior engineering and marketing to differentiate them in the market, while Dell is still just a big PC clone-maker. Also Apple is the leader in the quickly-growing portable media player market, which gives them even better growth prospects.
Er.. price edge, not value. How much to get OSX on that Dell laptop? OTOH, you can run (well,walk;) XP on the Mac. The amount of hassle-free work that one can get done over the life of the computer must be far higher on the Mac. Macs are way more fun to use, too, and owning one makes the buyer feel like a superior life-form, which is worth quite a bit by itself. pleasure + productivity - hassle = value
The first product based on IBM/Toshiba/Sony's Cell processor has shipped, reports Mercury Computer Systems. Mercury's Cell Technology Evaluation System (CTES) is a 470-pound behemoth with one or two dual-Cell blades running Linux. It targets defense, medical, and industrial inspection markets.
The CTES system is available with one or two of Mercury's Dual Cell-based Blade units. Each Blade features two Cell processors clocked at 2.4GHz, and running Linux in SMP (symmetric multi-processing) mode. Each Blade also has 512MB of "XDR" SDRAM, a 40GB hard drive, and dual gigabit Ethernet interfaces (dual PCIe Infiniband HCA add-in cards will be available in Q2). The Blades run a net-bootable Yellow Dog Linux variant called "Y-HPC" that was developed by Terra Soft Solutions, one of Mercury's VARs (value-added resellers).
The CTES system is housed in a 19-inch, 11U IBM Blade-Server chassis with a Web-based management module, dual gigabit Ethernet switches, and an "acoustic attenuation module." Additional components include a 17-inch flat-panel display with integrated keyboard and touchpad, 2000-Watt power supply, and an Intel Xeon-powered IBM xSeries 336 PC Server development and simulation system (a dual-PowerPC alternative will also be available) running RedHat Fedora Core 4 Linux. The system measures 34.4 x 20.5 x 24 inches, and weighs 470 pounds.
On the software side, an included IBM SDK offers compilers and gdb's (GNU debuggers) for the Cell processor's PPE (Power processor elements) and SPE (synergistic processor elements), along with a Cell simulator, and PPE and SPE libraries that support 32-bit PPE applications.
CTES additionally integrates a variety of Eclipse-integrated Mercury middleware, including its MultiCore Framework (MCF), aimed at managing the distribution of data across multiple computing elements working in tandem; its Scientific Algorithm Library (SAL); its Parallel Acceleration System (PAS); and its Trace Analysis Tool and Library (TATL).
Randy Dean, Mercury's VP of business and technology development, stated, "Our customers have expressed high expectations with the implementation of the Cell Technology into their application development."
Availability
The CTES is available now to "early access" customers at an undisclosed price.
There are scientific applications where single-precision is sufficient that do not need denormalized values, or use them infrequently enough that the special cases can be handled in software.
A single Cell, essentially a Network on Chip, offers up to 256 GFlop single precision floating point performance.
A prototype was produced with 90nm silicon on insulator (SOI) technology with 8 copper layers (wiring). It consists of 241 Million Transistors on 235 mm^2 and consumes 60-80W....
The Power Processing Element (PPE) [1 per chip] offers the normal PowerPC (PPC) ISA. It is a dual threaded 64 bit power processor which includes VMX (aka Altivec which is comparable to SSE). Its architecture is very simple to guarantee high clock rates. Thus, it uses only in order execution with a deep super scalar 2-way pipeline with more than 20 stages. It offers a 2x32kB L1 split cache, a 512kB L2 cache and virtualization....
Synergistic Processing Element [7 for game console (IIRC), or 8 per chip]- The SPE is essentially a full blown vector CPU with own RAM. Its ISA is not compatible to VMX and has a fixed length of 32 Bit. Current SPEs have about 21 Million Transistors where 2/3 of them are dedicated to the SRAM (memory). The processor has no branch prediction or scheduling logic, and relies on the program- mer/compiler to find parallelism in the code. As the PPE, it uses two independent pipelines and issues two instructions per cycle, one SIMD computation operation and one memory access operation. All instructions are processed strictly in-order and each instruction works with 128 Bit compound data items. 4 single precision floating point units and 4 integer units offer up to 32GOps each. The single precision floating point units are not IEEE754 compliant in terms of rounding and special val- ues. [denormalized numbers - +0, -0, +/-inf, etc.] The single precision units can also be used to compute double precision floating point numbers which are compliant to the IEEE754 standard. But their computa- tion is rather slow (3-4GFlops)....each SPE has it's own 256kB RAM which is called Local Storage (LS). This SRAM storage can be accessed extremely fast in 128 bit lines. Additionally, each SPE has a large register file of 128 128 bit registers which store all available data types. There is no cache, virtual memory support or coherency for the Local Storage...
The Element Interconnect Bus [1 per chip] - The EIB is the central communication channel inside a Cell processor, it consists of four 128 bit wide concentric rings. The ring uses buffered point to point commu- nication to transfer the data and is therewith scalable. It can move 96 bytes per cycle and is optimized for 1024 bit data blocks. Additional nodes (e.g. SPEs) can be added easily and increase only the maximal latency of the ring.
The I/O Interconnect - FlexIO The I/O Interconnect connects the Cell processor (the EIB) to the external world, e.g. other cell processors:). It offers 12 uni-directional byte-lanes which are 96 wires. Each lane may transport up to 6.4GB/s, which make 76.8 GB accumulated bandwidth. 7 lanes are outgoing (44.8 GB/s) and 5 lanes incoming (32 GB/s). There are cache coherent (CPU interconnect) and non coherent links (device interconnect) and two cell processors can be connected glueless....
The Memory Interface Controller The MIC connects the EIB to the main DRAM memory, which is in this case Rambus XDR memory which offers a bandwidth of 25.2 GB/s. The memory is ECC protected...
No, it could go the other way - current Mac users won't use anything else and Apple will make a mint from people who want a real OS. Huge numbers of people will try it out because friends recommended it and they can still keep XP just in case something goes wrong. They'll fall in love but will be frustrated by their old hardware and will make a real Mac their next computer purchase. Something similar happened with the iPod and its quality outshines the competition far less than OSX does XP.
Slashdot Mirror
Good idea. But you could add a few things to that "contract":
All knowledge derived from this course remains the property of the Professor. Any use ouside of class must be cleared in advance in writing with the Professor, for a licence fee to be negotiated. Any derivitive ideas must be disclosed to the Professor and such derivitive works become the sole property of the Professor. In no case shall any putative "fair use rights" extend beyond any arrangement of two successive words uttered by the Professor. Any disparagement of, or unlcenced or piratical use of any intellectual property of the Professor by the student shall be compensated by immediate liquidated damages paid by the student to the Professor equal to twice the lifetime income of the student plus interest and the title to the student's intellectual property if any, his or her genetic code, and all derivitive works.
Hmm... that seems a little soft - I think I'm leaving something out..
I read http://www.uclaprofs.com/profs/kellner.html, the "Radical of the Week" and I thought they made him seem like a great guy. They thoroughly skewered their own viewpoint and revealed how mindless their politics is.
Excuse me, but there are still professors being hired at top universities BECAUSE they are enthusiastic, fire-breathing, Joe Stalin-supporting Marxists. There are still excellent teachers and researchers who are denied tenure because they are not Marxists or deconstructionists, radical feminists or race-baiting Mugabe wannabes.
It would be fair to disparage this just effort to expose these vermin to the light of day with the allegation that they are pursuing only the leftists at UCLA, were it not for the fact that the right wing at UCLA is virtually nonexistent. If they go after just one right-winger at UCLA, they will have gotten half of them; if they go after a hundred tenured Maoist freaks, it'll just give them wriggles of joy to have been persecuted at last.
If we don't limit this to American citizens, or any particular kind of US government employee killing innocent people in any particular kind of way, then it is clear that Bush's actions in Iraq are orders of magnitude worse than the crimes of in the Clinton administration.
I don't know where you went to school, but the idea that professors are generally interested in "open intellectual inquiry" is sheer fantasy. If you have political disagreements with a humanities professor, your grade goes down in nine out of ten cases, particularly if you argue well.
"Liberals" in academia have no tolerance for dissent with the party line on racial equality of abilities, female superiority, white male culpability, or the insight of major incoherent French literary theorists. Nor will they tolerate dissent with their conservative values of professorial dominance over instructors, students and staff, within- and between- school hierarchy, and their faith that drawing a practical distinction between education and the educational establishment's credentials is seditious.
Even so-called scientists have a hard time looking at evidence objectively on controversial topics; in the humanities, the truth is whatever the professor says it is. Debunk any faulty argument or fabricated statistic that anyone uses to support any sacred-cow cause and you will be branded a heretic. Facts are irrelevant - if the opinion is "hurtful" the student will be punished one way or another - lower grades, denial of opportunities, ostracism, and discipliary measures.
Academia is really one of the most rigid cultures in the world, with many tabus, caste structures, distinct in- and out- groups, irrational rituals and shibboleths. Putting a social check on the excesses of professors by reminding them that their classes are in the public sphere of a wider culture than academia is a good idea which may in the future prevent or at least mitigate some of the abuses that professors have often committed against students in the past.
Yes there were good reasons to cut off the Japanese, but those were secondary to FDR's reasons for doing so. The main reason was to get into the war when FDR had repeatedly promised the American people that the US would not join in the war unless attacked. The only way to get around that was to invite an attack and make sure that it was not foiled.
Your other points are irrelevant in view of the information provided in the page to which I linked. You say "a link to a geocities site is not a very convincing argument ". True, but the link is not the argument, the argument is given on the page to which I linked, which you have clearly not read, apparently preferring to sneer at the label rather than take the medicine.
The facts and the law determine the answer to the factual question of guilt. Courts do not create guilt, they measure it. The presumption of innocence does not make the accused person actually innocent, it is a rule for how to measure guilt with as low a false positive rate as possible. [Oddly for such a fundamental principle, the presumption of innocence is not explicitly in the Constitution, but is inferred from amendments IV, V, VI, IX, and XIV.]That FDR died before he could be tried in no way affects the factual question of his guilt.
Treason is defined in Article III Section 3, US Constitution: "Treason against the United States, shall consist only in levying war against them, or in adhering to their enemies, giving them aid and comfort. No person shall be convicted of treason unless on the testimony of two witnesses to the same overt act, or on confession in open court."
FDR has been proven to have allowed the Japanese to attack when he had the information and means to prevent the attack or mitigate the damage of the attack. Top Secret Army Board Report, Oct, 1944: "Now let us turn to the fateful period between November 27 and December 6, 1941. In this period numerous pieces of information came to our State, War, and Navy Departments in all of their Top ranks indicating precisely the intentions of the Japanese including the probable exact hour and date of the attack. " The Army board had more information available to it than any other inquiry into the matter, and certainly met the requrement for witnesses to overt acts. It is reasonable to suppose that a court at the time, if it had access to the intercepts on which FDR was briefed twice daily (the vast majority of which are still asserted by the NSA to be too potentially damaging to national security to release even 65 years later), would have found FDR guilty of treason.
Acording to the articles, SRAM uses 6 transistors per bit and DRAM uses 1 capacitor and one transistor, while ZRAM uses just one transistor, so ZRAM is indeed supposed to be twice as dense as DRAM. The transistors scale down better than capacitors, making new process scales more easily and cheaply achievable, so ZRAM is likely to replace DRAM in main memory. Memory is a tough, low-margin business where costs per bit depend mostly on bits/area, so doubling density is something that competitive manufacturers have to do whenever it becomes practical. Memory manufacturers aren't using SOI much yet, though, so it may take a year or two to get this to market in a memory-only chip. SOI costs 15% more per raw wafer and there is a learning curve, but even so, the cost/bit reduction from ZRAM should be 40%-55% - greater than any other new tech since the current 1T 1C DRAM cell was adopted thirty years ago.
Good article. The parent should have been modded up, not down.
Good article, thanks.
It looks like there are no speed tradeoffs and it scales even better than DRAM, proven at 45 nm and suitable for 22 nm as well. AMD says such improvements usually take two years to show up in products, so that will still be 65 nm mostly with some 45nm. At 45 nm scale, a 1 cm^2 chip that is 70%-75% ZRAM would have about 48 MB.
That should mean about 25 bits per square micron in a 65um process or 3.25 megabytes of memory per square millimeter. Pretty cool.
It's a single-transistor capacitorless memory cell using the "floating body effect" of silicon on insulator (SOI) devices. Presumably stored charge in the gate affects the operation of the transistor in a way that can be used to store and read a bit, but I didn't feel like registering to read the white paper. The new memory should be six times denser than SRAM and twice as dense as DRAM.
There is no doubt that FDR knew about the Japanese plan to attack Pearl Harbor from hundreds if not thousands of monitored Japanese communications as well as highly placed sources from several different countries. In fact he wanted war so much he actually encouraged and abetted the attack, freezing Japanese funds in the US, cutting off oil to Japan, and feeding false information to Hawaii about the location of the Japanese fleet. There is more, but it could take a book to do it justice. Take a look at http://www.geocities.com/Pentagon/6315/pearl.html . FDR was guilty of treason.
[OK, this time in the right place - sorry, GP poster]
You do know there are no binaries on Google groups, right? And you do know that there are groups beginning with letters other than "alt", right? Like "sci" and "comp" and so forth, right? Or are you really that big a flaming turd?
You do know there are no binaries on Google groups, right? And you do know that there are groups beginning with letters other than "alt", right? Like "sci" and "comp" and so forth, right? Or are you really that big a flaming turd?
Now I feel like I have to take a shower.
The Constitution disagrees with you.
...
...
Article I Section 8. The Congress shall have power
To declare war, grant letters of marque and reprisal, and make rules concerning captures on land and water;
To raise and support armies, but no appropriation of money to that use shall be for a longer term than two years;
To provide and maintain a navy;
To make rules for the government and regulation of the land and naval forces;
To provide for calling forth the militia to execute the laws of the union, suppress insurrections and repel invasions;
To provide for organizing, arming, and disciplining, the militia, and for governing such part of them as may be employed in the service of the United States, reserving to the states respectively, the appointment of the officers, and the authority of training the militia according to the discipline prescribed by Congress;
A lot of people buy boxed distros, even when they could download. It's a lot to download, and in the case of a bootleg, who knows what hacks the uploader added? Anyway, a certain amount of piracy broadens the market, and they run a high risk anyway of getting their DRM cracked and OSX uploaded anyway. Once that happens they might as well accept the money of the people who would prefer to get Intel-version OSX from Apple directly.
Yep. Brilliant.
Beyond that, if they relaease a boxed Intel OSX that will run on non-Apple hardware (no Apple support, obviously), they could sell perhaps 4x - 8x as many units as they currently sell computers. Each of those units would be 90%+ profit, and at say $100 - $200 each would likely give them more profit than the rest of their computer business. Each such "unauthorized" unit would be like an ad for their hardware, too, and those using OSX on generic hardware would want to move up to the social status, reliability, clean design and support of having a real Mac. Those who already have Mac hardware are mostly too loyal and hassle-avoiding to move to the PC hardware, and even if some do, the potential new market is so much bigger that overall it's a huge win for Apple. Five years out, Apple could be on a third of computers or more, but since many of those will also have a Microsoft OS, it won't take nearly that much business away from Microsoft.
So having a lower market share than Dell in the slowly-growing computer market is actually better for forward-looking Apple investors - much more room for growth than Dell, particularly when Apple is selling machines which have superior engineering and marketing to differentiate them in the market, while Dell is still just a big PC clone-maker. Also Apple is the leader in the quickly-growing portable media player market, which gives them even better growth prospects.
"So Dell does still have the value edge. "
;) XP on the Mac. The amount of hassle-free work that one can get done over the life of the computer must be far higher on the Mac. Macs are way more fun to use, too, and owning one makes the buyer feel like a superior life-form, which is worth quite a bit by itself. pleasure + productivity - hassle = value
Er.. price edge, not value. How much to get OSX on that Dell laptop? OTOH, you can run (well,walk
http://www.linuxdevices.com/news/NS3591350722.html
Jan. 10, 2006
The first product based on IBM/Toshiba/Sony's Cell processor has shipped, reports Mercury Computer Systems. Mercury's Cell Technology Evaluation System (CTES) is a 470-pound behemoth with one or two dual-Cell blades running Linux. It targets defense, medical, and industrial inspection markets.
The CTES system is available with one or two of Mercury's Dual Cell-based Blade units. Each Blade features two Cell processors clocked at 2.4GHz, and running Linux in SMP (symmetric multi-processing) mode. Each Blade also has 512MB of "XDR" SDRAM, a 40GB hard drive, and dual gigabit Ethernet interfaces (dual PCIe Infiniband HCA add-in cards will be available in Q2). The Blades run a net-bootable Yellow Dog Linux variant called "Y-HPC" that was developed by Terra Soft Solutions, one of Mercury's VARs (value-added resellers).
The CTES system is housed in a 19-inch, 11U IBM Blade-Server chassis with a Web-based management module, dual gigabit Ethernet switches, and an "acoustic attenuation module." Additional components include a 17-inch flat-panel display with integrated keyboard and touchpad, 2000-Watt power supply, and an Intel Xeon-powered IBM xSeries 336 PC Server development and simulation system (a dual-PowerPC alternative will also be available) running RedHat Fedora Core 4 Linux. The system measures 34.4 x 20.5 x 24 inches, and weighs 470 pounds.
On the software side, an included IBM SDK offers compilers and gdb's (GNU debuggers) for the Cell processor's PPE (Power processor elements) and SPE (synergistic processor elements), along with a Cell simulator, and PPE and SPE libraries that support 32-bit PPE applications.
CTES additionally integrates a variety of Eclipse-integrated Mercury middleware, including its MultiCore Framework (MCF), aimed at managing the distribution of data across multiple computing elements working in tandem; its Scientific Algorithm Library (SAL); its Parallel Acceleration System (PAS); and its Trace Analysis Tool and Library (TATL).
Randy Dean, Mercury's VP of business and technology development, stated, "Our customers have expressed high expectations with the implementation of the Cell Technology into their application development."
Availability
The CTES is available now to "early access" customers at an undisclosed price.
There are scientific applications where single-precision is sufficient that do not need denormalized values, or use them infrequently enough that the special cases can be handled in software.
Great article - here are some of the hardware details:
...
...
:). It offers 12 uni-directional byte-lanes which are 96 wires. Each lane may transport up to 6.4GB/s, which make 76.8 GB accumulated bandwidth. 7 lanes are outgoing (44.8 GB/s) and 5 lanes incoming (32 GB/s). There are cache coherent (CPU interconnect) and non coherent links (device interconnect) and two cell processors can be connected glueless. ...
[Quotes from "The Cell Processor - A short Introduction" by Torsten Hoefler - bracketed comments are mine.]
A single Cell, essentially a Network on Chip, offers up to 256 GFlop single precision floating point performance.
A prototype was produced with 90nm silicon on insulator (SOI) technology with 8 copper layers (wiring). It consists of 241 Million Transistors on 235 mm^2 and consumes 60-80W.
The Power Processing Element (PPE) [1 per chip] offers the normal PowerPC (PPC) ISA. It is a dual threaded 64 bit power processor which includes VMX (aka Altivec which is comparable to SSE). Its architecture is very simple to guarantee high clock rates. Thus, it uses only in order execution with a deep super scalar 2-way pipeline with more than 20 stages. It offers a 2x32kB L1 split cache, a 512kB L2 cache and virtualization.
Synergistic Processing Element [7 for game console (IIRC), or 8 per chip]- The SPE is essentially a full blown vector CPU with own RAM. Its ISA is not compatible to VMX and has a fixed length of 32 Bit. Current SPEs have about 21 Million Transistors where 2/3 of them are dedicated to the SRAM (memory). The processor has no branch prediction or scheduling logic, and relies on the program- mer/compiler to find parallelism in the code. As the PPE, it uses two independent pipelines and issues two instructions per cycle, one SIMD computation operation and one memory access operation. All instructions are processed strictly in-order and each instruction works with 128 Bit compound data items. 4 single precision floating point units and 4 integer units offer up to 32GOps each. The single precision floating point units are not IEEE754 compliant in terms of rounding and special val- ues. [denormalized numbers - +0, -0, +/-inf, etc.] The single precision units can also be used to compute double precision floating
point numbers which are compliant to the IEEE754 standard. But their computa-
tion is rather slow (3-4GFlops)....each SPE has it's own 256kB RAM which is called Local Storage (LS). This SRAM storage can be accessed extremely fast in 128 bit lines. Additionally, each SPE has a large register file of 128 128 bit registers which store all available data types. There is no cache, virtual memory support or coherency for the Local Storage...
The Element Interconnect Bus [1 per chip] - The EIB is the central communication channel inside a Cell processor, it consists of four 128 bit wide concentric rings. The ring uses buffered point to point commu- nication to transfer the data and is therewith scalable. It can move 96 bytes per cycle and is optimized for 1024 bit data blocks. Additional nodes (e.g. SPEs) can be added easily and increase only the maximal latency of the ring.
The I/O Interconnect - FlexIO The I/O Interconnect connects the Cell processor (the EIB) to the external world, e.g. other cell processors
The Memory Interface Controller The MIC connects the EIB to the main DRAM memory, which is in this case Rambus XDR memory which offers a bandwidth of 25.2 GB/s. The memory is ECC protected...
No, it could go the other way - current Mac users won't use anything else and Apple will make a mint from people who want a real OS. Huge numbers of people will try it out because friends recommended it and they can still keep XP just in case something goes wrong. They'll fall in love but will be frustrated by their old hardware and will make a real Mac their next computer purchase. Something similar happened with the iPod and its quality outshines the competition far less than OSX does XP.