Cell server maker promises pint-sized powerhouse

[PC-Engine]

What is a supercomputer?

PCE, you mean to say the machine at 501th place is not a Supercomputer?

Technically I'd say no. In fact nobody even knows what the 501st place would be in terms of performance. It could be 1 TFLOPS or it could be 500GFLOPS.

Would you consider the 1000th place a supercomputer? What about the 1millionth place? Is my computer a supercomputer too even if it falls to the 400 millionth place?

Get the point?

 

[Ty]

What is the established definition of a supercomputer?

The rank-based system that PCE suggests? Does it have to be DP? Wikipedia didn't mention one whit about precision.

http://nhse.cs.rice.edu/NHSEreview/CMS/Chapter6.html

Just mentions, "A time dependent term which refers to the class of most powerful computer systems world-wide at the time of reference." Which seems to be a rank based system void of precision - but I don't know if this is reference quality.

 

[pcchen]

Actually I think a supercomputer has to be able to do something very fast that normal computers (including specially designed computers) can't even touch. At least 100X faster is a good start. I prefer 1000X faster.

For example, Deep Crack is a supercomputer, although it can't even compute any floating point numbers. However, it is able to process DES algorithm much faster than any computer one can buy off the shelf (I think this still applies). IMHO that makes it a supercomputer.

So is this CELL based computer a supercomputer? For signal processing jobs, I'd say yes. It's probably able to process sonar signals or radar signals much faster than anything else. Furthermore, it's probably still ok for bio-informatics workloads. It may not be so hot for some other works, but that's not what it's designed for.

 

[karlotta]

Actually I think a supercomputer has to be able to do something very fast that normal computers (including specially designed computers) can't even touch. At least 100X faster is a good start. I prefer 1000X faster.

For example, Deep Crack is a supercomputer, although it can't even compute any floating point numbers. However, it is able to process DES algorithm much faster than any computer one can buy off the shelf (I think this still applies). IMHO that makes it a supercomputer.

So is this CELL based computer a supercomputer? For signal processing jobs, I'd say yes. It's probably able to process sonar signals or radar signals much faster than anything else. Furthermore, it's probably still ok for bio-informatics workloads. It may not be so hot for some other works, but that's not what it's designed for.

My thoughts too.Reason is a good thing. To get all anal about the exact def. of "supercomputer" is like arguing about your dick size in the artic ocean.

 

[PC-Engine]

For example, Deep Crack is a supercomputer, although it can't even compute any floating point numbers. However, it is able to process DES algorithm much faster than any computer one can buy off the shelf (I think this still applies). IMHO that makes it a supercomputer.

You still have to compare it to other supercomputers. I think it's kinda naive to compare it to something you could by off the shelf. So how does it compare to supercomputers in the top500 running DES algorithm?

 

[pcchen]

You still have to compare it to other supercomputers. I think it's kinda naive to compare it to something you could by off the shelf. So how does it compare to supercomputers in the top500 running DES algorithm?

Since different computers are designed for different jobs, it sometimes pointless to compare between them. For example, in DES there is a pointless (really) bit-level permutation stage which is easy to implement on hardware but quite slow on software.

Anyway, Deep Crack in 1998 cracked a encrypted message by brute force in 56 hours. I don't know how fast a PPC 440 0.7GHz used in BlueGene/L can do on decrypting DES. A fast DES algorithm by Biham runs about 2M blocks per second on a 300MHz Alpha 8400 processor, so a PPC 440 0.7GHz is probably able to process about 5M blocks per second. It now has 131,072 processors so it's able to process 0.65 T blocks per second. So on average it should be able to crack a DES message in about 15 hours. Of course, Deep Crack is a 1998 machine, a more recent design should be much faster.

 

[PC-Engine]

Since different computers are designed for different jobs, it sometimes pointless to compare between them. For example, in DES there is a pointless (really) bit-level permutation stage which is easy to implement on hardware but quite slow on software.

Anyway, Deep Crack in 1998 cracked a encrypted message by brute force in 56 hours. I don't know how fast a PPC 440 0.7GHz used in BlueGene/L can do on decrypting DES. A fast DES algorithm by Biham runs about 2M blocks per second on a 300MHz Alpha 8400 processor, so a PPC 440 0.7GHz is probably able to process about 5M blocks per second. It now has 131,072 processors so it's able to process 0.65 T blocks per second. So on average it should be able to crack a DES message in about 15 hours. Of course, Deep Crack is a 1998 machine, a more recent design should be much faster.

What was available in the top500 back in 98?

 

[pcchen]

I think the #1 computer in the TOP500 listing in 1998 is the ASCI Red, which comprises of more than 9000 PPro processors. It's linpack benchmark is about 1.3TFLOPS. It's still in the latest list (#276) because it's upgraded to P3 Xeon in 1999.

[EDIT] I should clarify that Deep Crack was built to crack a DES message in 4.5 days on average. So BlueGene/L is probably 7x or faster than Deep Crack at cracking DES.

 

[pcchen]

My friend sent me a ilnk which has some benchmarks of CELL processor:

CELL performance view

It said that CELL with 8 SPU running at 3.2GHz can do 9.46 GFLOPS in LINPACK 1kx1k (peak 14.63 GFLOPS). Not bad for a single processor. The best single processor in LINPACK 1kx1k remains to be NEC SX-8 2GHz, which can do 14.96 GFLOPS (peak 16 GFLOPS).

When running SP code, 8 SPU can do 73.04 GFLOPS in LINPACK 1kx1k, and 155.5 GFLOPS in 4kx4k.

 

[mckmas8808]

Yeah we've been talking about that here for a few days now. Thanks for sharing though.