From my own benchmarks, the dual threaded PPC64 is very similar to my desktop AMD64 @1.8GHz (one core). In addition, the 6 SPE units of the Cell are beasts, not just doing SIMD, but also because of massive bandwidth between their 256KB per SPE local memories (25GB/s of simultaneous data transfer between *many* SPEs at same time (!)).
You're right but only for postgraduate students, as it is because doesn't count as work (in Spain, the used term is "beca", in the same way of "scholarship" finance aid in other countries), also, there are almost no taxes (because legally does not count as work, although it may change in the future, to encourage investigation, etc.).
Needless to say that engineering salary (not scolarship) it is not fixed, in no way.
I bet for a case of misunderstanding. Not by a law, but it is frequent that 23-24 year old engineering workers make 14-18k euro/year (before taxes), reaching 28-30k (before taxes) with 5 years of experience. It is difficult to make more than 30k in Spain by just programming, despite your briliantness (that's why, unfortunately for Spain -as we lose most brilliant people-, so many people run away from the country, looking for a better standard of living across Europe).
I'm also spanish, engineer, and I don't know what are you talking about. What law? Do you have a BOE link? Or are you talking about a "de facto law" or about your experience about some specific context?
You're right. It's so sad to se a powerful 520MHz ARM XScale processor in the Toshiba Portege G900 running Windows Mobile 7... it is slower than my old 8MHz 8086 running GEM (from Digital Research) !!!
Unfortunately, I found no way of loading Linux in it (some other smartphone devices such as the ones made by HTC are easier to hack).
P.S. I did not bought the smartphone, it was a present from the company for which I work, for my personal use, and also to be able to test some nice remote desktop access software done by us (as upgrade from a Nokia 3410, in my opinion outperforms the G900 as phone, despite the lack of bells and whistles).
Intel has to buy Nvidia or AMD to be competitive in the long run. There is no try, as if AMD survive, it could hurt badly Intel with CPU+IGP solutions.
Well, still being true, it is not enough as excuse. NVidia has power enough to "communicate" the "importance" of moving to the open source arena... capisci?
VxWorks is the only OS I've played with so far that allows this (...)
May be you need to play more:-) . LynxOS is a full-fledged POSIX RTOS, with memory protection, far better than VxWorks if disk and network is involved (I was developing toll systems real time applications for 4 years). I hope someday Linux reach RTOS habilities without having to use "cokernels", as although I'm pretty happy with current kernel preemptiveness, I would love to get true RTOS Linux (echo 1 >/proc/rtos).
Well, yes, but for parallel code the best synchronization is the one that can be delayed or completely avoided. I would rather prefer to use an algorithm that requires 100 critical region access per second using the OS primitives rather than a 100,000 accesses using "optimized" critical region code (i.e. favor better algorithms rather than optimize a low frequency operation).
I posted a benchmark that without using special instructions, just some unrolling, gives about 2x speed-up on Pentium Pro and above, without losing backward compatibility. For extra speed-up, you can use SSE2 prefetch, but it is not necessary.
That kind of unrolling has to be handwritten in assembly, as despite current compilers are good, are not good enough to give the last bit of performance.
I know you were joking, but these days, in both in-order and out-of-order CPUs, there is very little code that it is worth to handwrite in assembly (only vector code, and it may be matter of time to get good enough vectorized code). Good C/C++ handmade optimization, helping the compiler with the strength reduction and constant propagation phases, with some explicit/manual unrolling, trying to avoid data cache misses, is almost as fast as writing in assembly (one data cache miss is worth more than any superb assembly optimization, as current RAMs have huge latencies -from 40 to 200 cycles-).
You're right. These instruction became useless -because a lot faster implementation was possible- since the Pentium Pro, being introduced out-of-order execution and enhanced branch prediction. I'm not sure about unrolling can be actually much faster on the original Pentium, but I'm convinced that you could be able to get a notable speed-up, if not in the string scan case, at least in the memcpy case using the FPU for 64-bit transfers.
Some examples, actual bencharks (2 years old, but are pretty the same with K8 and Core2Duo):
REPNE SCASD: (look element into sequential dword vector)
Pentium II @300MHz: 133 MB/s (100MHz FSB, 100MHz SDRAM)
Pentium IV @3GHz: 2.3 GB/s (800MHz FSB, 400MHz DDR SDRAM)
256-bit uprolling: (process 8 elements in a row)
Pentium II @300MHz: 233MB/s (100MHz FSB, 100MHz SDRAM)
Pentium IV @3GHz: 3.3 GB/s (800MHz FSB, 400MHz DDR SDRAM)
256-bit uprolling w/ SSE2 prefetch to increase data cache hit: (process 8 elements in a row)
Pentium II @300MHz: -no SSE2- (100MHz FSB, 100MHz SDRAM)
Pentium IV @3GHz: 4.0 GB/s (800MHz FSB, 400MHz DDR SDRAM)
P.S. Both REP MOVSB and REP MOVSD are slow: the performance per clock is between 1/8 and 1/16 in the first and between 1/2 and 1/4 in the second. The is no reason for using the microcoded instructions other than backwards compatibility, but it seems nonsense to me to save 16KB to write unrolled and/or prefetched memcpy/memmove/scan variants.
These "quick operations" are not quick anymore; on modern -out of order- x86 procesors (P4, PentiumM/Core, K7, K8), explicit string search (still without using SIMD tricks) is from 2x to 3x -using SSE2 prefetch- faster than the microprogrammed code, as you can unroll loops without conditional jump penalty.
Bertrand Russell's specialization was not the philosophy of religion, but rather other fields within philosophy.
It is your opinion.
Why don't you quote people who actually know what they are talking about, non-theist philosophers of religion like (pre-conversion) Flew or Mackie?
We'll, I was also thinking about quoting Philip J. Fry, but thought that Russell had a bit more of relevance in the mankind history.
Please check your constitution, I hope that no civil law can override constitutional rights in Sweden (I know Jönköping and Stockholm, and so civilized country deserves a civilized carta magna), however, unconstitutional laws can be runned until some civil association action tries to stop the madness.
Exiting times indeed in Sweden with growing fears about privacy, the influence of swedish and american record companies and the impending trial of The Pirate Bay.
Same fears here, in Spain. Goverments could use the media way as an *excuse* to cut civil rights. We must stop it, it is not about not paying for copyrights, its about individual freedom (!).
Probably not. The Spanish law doesn't require telcos to disclose the requested information (actually they would get in serious trouble if they provided it to third parties without a judge involved), and the EU said that said law is ok, but other countries could have a different law and that would be ok as well.
In Spain, judges play the "God mode" role, there are almost no limits for them, and are the only ones that are able to request *private* held information. As user, Telefonica/Orange/Jazztel/ONO/BT/ya.com/COLT/whatever can not, in no case, provide private information to third parties other than to the case instructing judge. Both "Promusicae" and "SGAE" are RIAA-like private associations. In the case of "SGAE" (stands for "Sociedad General de Autores y Editores", that can be poorly translated as "Authors and Editors General Association" -theorethically a non profit association-), they get money as "canon" -as euphemism, because in Spain a private company can not emit taxes!- for every suitable media (CD/DVD-R*, Hard Disks, flash memory, etc.). In my opinion, Al Capone had, much better gang taste.
As side effect of the undercover taxing, most people buy media to other countries, not only avoid the "canon" (Al-Capone-like tax, as is being issued by a *private* organization), but also VAT and other taxes (!). That stupid measure is generating a huge black market of media, and hurting badly the little computer shops.
Media associations (RIAA an others) are pushing without thinking about that who make the laws are the same that buys the media. The boomerang is already going back, por borregos y avariciosos.
Slashdot Mirror
From my own benchmarks, the dual threaded PPC64 is very similar to my desktop AMD64 @1.8GHz (one core). In addition, the 6 SPE units of the Cell are beasts, not just doing SIMD, but also because of massive bandwidth between their 256KB per SPE local memories (25GB/s of simultaneous data transfer between *many* SPEs at same time (!)).
You're right. Havok can be in the same case as GNU C library (LGPL), which can be linked statically to GPL code.
You're right but only for postgraduate students, as it is because doesn't count as work (in Spain, the used term is "beca", in the same way of "scholarship" finance aid in other countries), also, there are almost no taxes (because legally does not count as work, although it may change in the future, to encourage investigation, etc.).
Needless to say that engineering salary (not scolarship) it is not fixed, in no way.
I bet for a case of misunderstanding. Not by a law, but it is frequent that 23-24 year old engineering workers make 14-18k euro/year (before taxes), reaching 28-30k (before taxes) with 5 years of experience. It is difficult to make more than 30k in Spain by just programming, despite your briliantness (that's why, unfortunately for Spain -as we lose most brilliant people-, so many people run away from the country, looking for a better standard of living across Europe).
I'm also spanish, engineer, and I don't know what are you talking about. What law? Do you have a BOE link? Or are you talking about a "de facto law" or about your experience about some specific context?
323233331112132 33323132212331 2111331132312233 333212123213112 311333313331111 211333323232211 232313331121231 33231312
323233331112132 33323132212331 2111331132312233 333212123213113 311333313331111 211333323232211 232313331121231 33231312
You're right. It's so sad to se a powerful 520MHz ARM XScale processor in the Toshiba Portege G900 running Windows Mobile 7... it is slower than my old 8MHz 8086 running GEM (from Digital Research) !!!
Unfortunately, I found no way of loading Linux in it (some other smartphone devices such as the ones made by HTC are easier to hack).
P.S. I did not bought the smartphone, it was a present from the company for which I work, for my personal use, and also to be able to test some nice remote desktop access software done by us (as upgrade from a Nokia 3410, in my opinion outperforms the G900 as phone, despite the lack of bells and whistles).
Intel has to buy Nvidia or AMD to be competitive in the long run. There is no try, as if AMD survive, it could hurt badly Intel with CPU+IGP solutions.
Well, still being true, it is not enough as excuse. NVidia has power enough to "communicate" the "importance" of moving to the open source arena... capisci?
Click here to read the original in spanish (for completeness' sake).
Thank you.
May be you need to play more
Well, you can check for string length before applying one or another snippet.
Well, yes, but for parallel code the best synchronization is the one that can be delayed or completely avoided. I would rather prefer to use an algorithm that requires 100 critical region access per second using the OS primitives rather than a 100,000 accesses using "optimized" critical region code (i.e. favor better algorithms rather than optimize a low frequency operation).
I posted a benchmark that without using special instructions, just some unrolling, gives about 2x speed-up on Pentium Pro and above, without losing backward compatibility. For extra speed-up, you can use SSE2 prefetch, but it is not necessary.
That kind of unrolling has to be handwritten in assembly, as despite current compilers are good, are not good enough to give the last bit of performance.
I know you were joking, but these days, in both in-order and out-of-order CPUs, there is very little code that it is worth to handwrite in assembly (only vector code, and it may be matter of time to get good enough vectorized code). Good C/C++ handmade optimization, helping the compiler with the strength reduction and constant propagation phases, with some explicit/manual unrolling, trying to avoid data cache misses, is almost as fast as writing in assembly (one data cache miss is worth more than any superb assembly optimization, as current RAMs have huge latencies -from 40 to 200 cycles-).
You're right. These instruction became useless -because a lot faster implementation was possible- since the Pentium Pro, being introduced out-of-order execution and enhanced branch prediction. I'm not sure about unrolling can be actually much faster on the original Pentium, but I'm convinced that you could be able to get a notable speed-up, if not in the string scan case, at least in the memcpy case using the FPU for 64-bit transfers.
Some examples, actual bencharks (2 years old, but are pretty the same with K8 and Core2Duo):
REPNE SCASD: (look element into sequential dword vector)
Pentium II @300MHz: 133 MB/s (100MHz FSB, 100MHz SDRAM)
Pentium IV @3GHz: 2.3 GB/s (800MHz FSB, 400MHz DDR SDRAM)
256-bit uprolling: (process 8 elements in a row)
Pentium II @300MHz: 233MB/s (100MHz FSB, 100MHz SDRAM)
Pentium IV @3GHz: 3.3 GB/s (800MHz FSB, 400MHz DDR SDRAM)
256-bit uprolling w/ SSE2 prefetch to increase data cache hit: (process 8 elements in a row)
Pentium II @300MHz: -no SSE2- (100MHz FSB, 100MHz SDRAM)
Pentium IV @3GHz: 4.0 GB/s (800MHz FSB, 400MHz DDR SDRAM)
P.S. Both REP MOVSB and REP MOVSD are slow: the performance per clock is between 1/8 and 1/16 in the first and between 1/2 and 1/4 in the second. The is no reason for using the microcoded instructions other than backwards compatibility, but it seems nonsense to me to save 16KB to write unrolled and/or prefetched memcpy/memmove/scan variants.
These "quick operations" are not quick anymore; on modern -out of order- x86 procesors (P4, PentiumM/Core, K7, K8), explicit string search (still without using SIMD tricks) is from 2x to 3x -using SSE2 prefetch- faster than the microprogrammed code, as you can unroll loops without conditional jump penalty.
[fucking possible patent by prior act: on]
No problem, let's go program "hard disk comfort noise" as done in crystal-clear audio communications!!
In my opinion, Russell was as qualified as Kurt Gödel in the "philosophy of religion" field, years light beyond Descartes or Saint Thomas.
It is your opinion.
Why don't you quote people who actually know what they are talking about, non-theist philosophers of religion like (pre-conversion) Flew or Mackie?
We'll, I was also thinking about quoting Philip J. Fry, but thought that Russell had a bit more of relevance in the mankind history.
"...The knowledge exists by which universal happiness can be secured; the chief obstacle to its utilization for that purpose is the teaching of religion. Religion prevents our children from having a rational education; religion prevents us from removing the fundamental causes of war; religion prevents us from teaching the ethic of scientific co-operation in place of the old fierce doctrines of sin and punishment. It is possible that mankind is on the threshold of a golden age; but, if so, it will be necessary first to slay the dragon that guards the door, and this dragon is religion." (Bertrand Russell)
Exiting times indeed in Sweden with growing fears about privacy, the influence of swedish and american record companies and the impending trial of The Pirate Bay.
Same fears here, in Spain. Goverments could use the media way as an *excuse* to cut civil rights. We must stop it, it is not about not paying for copyrights, its about individual freedom (!).
Hej då!
In Spain, judges play the "God mode" role, there are almost no limits for them, and are the only ones that are able to request *private* held information. As user, Telefonica/Orange/Jazztel/ONO/BT/ya.com/COLT/whatever can not, in no case, provide private information to third parties other than to the case instructing judge. Both "Promusicae" and "SGAE" are RIAA-like private associations. In the case of "SGAE" (stands for "Sociedad General de Autores y Editores", that can be poorly translated as "Authors and Editors General Association" -theorethically a non profit association-), they get money as "canon" -as euphemism, because in Spain a private company can not emit taxes!- for every suitable media (CD/DVD-R*, Hard Disks, flash memory, etc.). In my opinion, Al Capone had, much better gang taste.
As side effect of the undercover taxing, most people buy media to other countries, not only avoid the "canon" (Al-Capone-like tax, as is being issued by a *private* organization), but also VAT and other taxes (!). That stupid measure is generating a huge black market of media, and hurting badly the little computer shops.
Media associations (RIAA an others) are pushing without thinking about that who make the laws are the same that buys the media. The boomerang is already going back, por borregos y avariciosos.