• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

ibm reclaims supercomputer lead

ElFenix

ElFenix

Elite Member
Super Moderator
<a target=_blank class=ftalternatingbarlinklarge href="http://www.reuters.com/newsArticle.jhtml?type=technologyNews&amp;storyID=6367372">Blue Gene/L machine... will occupy 64 full racks, with a peak performance of 360 teraflops.
:Q</a>
 
what is a flop? and what is a teraflop? I know its a large amoutn but Iv never heard of it before.

also any idea what processors they used in this comp? AMD 64's?
 
a teraflop is 1 trillion calculations per second. on comparison, nowadays athlon 64s and P4s get 7000 megaflops(million operations per second) or less.

so 7000 Megaflops compared to 360 Teraflops.



 
Originally posted by: DEMO24
what is a flop? and what is a teraflop? I know its a large amoutn but Iv never heard of it before.

also any idea what processors they used in this comp? AMD 64's?
Click to expand...
I believe a flop is: floating point operation per second.
 
Actually, not really.

The benchmark used is probably linpack, so blue gene will actually perform better in applications that are similar to linpack. I'd bet that it would actually be SLOWER than earth simulator on certain important programs.

Earth Simulator is a vector processor machine, the same kind of processor design Cray machines used to be made of, if I'm not mistaken. This technology is very much interesting and quite undevelopped. Some knowledgeable people believe vector processors could achieve incredible performance in certain types of applications due to how they exploit parallelism.

The fact that Blue Gene overthrew Earth Simulator could be bad news for the computing industry if this means that less interest will be given to vector processors.

Let me make an hypothetic comparison. First Guy invents a way to appropriately harness magnetic propulsion as an alternative to combustion engines, then Second Guy just packs more explosive fuel in his combustion engine and shows that his engine goes faster than First Guy's magnetic engine. Hence, more dollars go into researching more explosive fuels, and money is withdrawn from magnetic propulsion....

🙁
 
Originally posted by: DEMO24
what is a flop? and what is a teraflop? I know its a large amoutn but Iv never heard of it before.

also any idea what processors they used in this comp? AMD 64's?
Click to expand...


flop = Floating Point Operation
and they probably used mutiple Power 5 cpus
 
Originally posted by: nghtdvl
Actually, not really.

The benchmark used is probably linpack, so blue gene will actually perform better in applications that are similar to linpack. I'd bet that it would actually be SLOWER than earth simulator on certain important programs.

Earth Simulator is a vector processor machine, the same kind of processor design Cray machines used to be made of, if I'm not mistaken. This technology is very much interesting and quite undevelopped. Some knowledgeable people believe vector processors could achieve incredible performance in certain types of applications due to how they exploit parallelism.

The fact that Blue Gene overthrew Earth Simulator could be bad news for the computing industry if this means that less interest will be given to vector processors.

Let me make an hypothetic comparison. First Guy invents a way to appropriately harness magnetic propulsion as an alternative to combustion engines, then Second Guy just packs more explosive fuel in his combustion engine and shows that his engine goes faster than First Guy's magnetic engine. Hence, more dollars go into researching more explosive fuels, and money is withdrawn from magnetic propulsion....

🙁
Click to expand...

Nearly all modern processors have vector instructions, they're hardly something new. As far as I know, vector instructions are just SIMD instructions. SSE/SSE2/SSE3, Altivec, and 3Dnow are all SIMD instructions that can be performed on run-of-the-mill CPUs.
 
Originally posted by: rh71
Originally posted by: Yossarian
skynet here we come.
Click to expand...
LOL. No... IBM doesn't use that control-everything mentality. You're confusing them with Microsoft.
Click to expand...

IBM's always been shady. Just because microsoft is the demon of the day, doesn't mean IBM won't return to the top hated spot. 😉
 
What a do not get is that since the system is a cluster, what is keeping them from adding more nodes instead of building a new system?
 
Originally posted by: RaynorWolfcastle
[Nearly all modern processors have vector instructions, they're hardly something new. As far as I know, vector instructions are just SIMD instructions. SSE/SSE2/SSE3, Altivec, and 3Dnow are all SIMD instructions that can be performed on run-of-the-mill CPUs.
Click to expand...

Actually, it is not only a question of having vector instructions, but rather having the complete processor architecture built around vectors.... vector registers, vector functional units, vector load/store units, etc. I know that SSE2 does somewhat emulate this on scalar cpus, but, for example, where AltiVec and SSE2 have a vector length of 128 bits, a VMIPS vector processor uses 4096 bits. This kind of processor will often perform exceptionnally better in certain applications that kill scalar cpus (usually those of the scatter-gather type over huge data sets... weather prediction, climate change, even car crash simulations). I'm not denying that it's not the other way around in lots of cases, but I'd like to have all avenues explored, including "pure" vector architectures. I guess the main problem with this architecture is the cost, which really kills the deal.
 
Originally posted by: nghtdvl
Originally posted by: RaynorWolfcastle
[Nearly all modern processors have vector instructions, they're hardly something new. As far as I know, vector instructions are just SIMD instructions. SSE/SSE2/SSE3, Altivec, and 3Dnow are all SIMD instructions that can be performed on run-of-the-mill CPUs.
Click to expand...

Actually, it is not only a question of having vector instructions, but rather having the complete processor architecture built around vectors.... vector registers, vector functional units, vector load/store units, etc. I know that SSE2 does somewhat emulate this on scalar cpus, but, for example, where AltiVec and SSE2 have a vector length of 128 bits, a VMIPS vector processor uses 4096 bits. This kind of processor will often perform exceptionnally better in certain applications that kill scalar cpus (usually those of the scatter-gather type over huge data sets... weather prediction, climate change, even car crash simulations). I'm not denying that it's not the other way around in lots of cases, but I'd like to have all avenues explored, including "pure" vector architectures. I guess the main problem with this architecture is the cost, which really kills the deal.
Click to expand...

Invest in SGI.
 
Originally posted by: n0cmonkey
Originally posted by: nghtdvl
Originally posted by: RaynorWolfcastle
[Nearly all modern processors have vector instructions, they're hardly something new. As far as I know, vector instructions are just SIMD instructions. SSE/SSE2/SSE3, Altivec, and 3Dnow are all SIMD instructions that can be performed on run-of-the-mill CPUs.
Click to expand...

Actually, it is not only a question of having vector instructions, but rather having the complete processor architecture built around vectors.... vector registers, vector functional units, vector load/store units, etc. I know that SSE2 does somewhat emulate this on scalar cpus, but, for example, where AltiVec and SSE2 have a vector length of 128 bits, a VMIPS vector processor uses 4096 bits. This kind of processor will often perform exceptionnally better in certain applications that kill scalar cpus (usually those of the scatter-gather type over huge data sets... weather prediction, climate change, even car crash simulations). I'm not denying that it's not the other way around in lots of cases, but I'd like to have all avenues explored, including "pure" vector architectures. I guess the main problem with this architecture is the cost, which really kills the deal.
Click to expand...

Invest in SGI.
Click to expand...

LOL...i need to write that one down...😀 ...lol
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top