Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Vtech G5 Supercomputer Linpack Performance Revealed
- Thread starter Cubeboy
- Start date
- Sort by reaction score
Firstly, I really don't know if it did well when compared to the other machines since I don't recognize them.
Secondly, its on page 55. The computers were listed in order from fastest to slowest, I think.
Thirdly, if Apple really did well for a desktop, then sweet!!!
Secondly, its on page 55. The computers were listed in order from fastest to slowest, I think.
Thirdly, if Apple really did well for a desktop, then sweet!!!
Re: Vtech G5 Supercomputer Linpack Performance Revealed
Only 256 processors => 128 machines.
The final cluster has 1100 machines (2200 processors).
arn
Originally posted by Cubeboy
Can be found as entry six (page 54) on this pdf
Pretty impressive, not as outstanding as many of us might have hoped but impressive nonetheless.
Only 256 processors => 128 machines.
The final cluster has 1100 machines (2200 processors).
arn
According to my quick math, that puts it in 9th with a score of around 7055 GFlops, if we extrapolate the score for 256 CPUs (821GFlops x 8.59).
Not in the top 5 like some had hoped, but not bad either.
Not in the top 5 like some had hoped, but not bad either.
Rower
The problem is clusters scale far from linearly (or if working off of your mathematical extrapolation, a step-based linear profile.) That means, unless the current results for the 256 CPUs was far from optimized, there's a fairly good chance the final cluster may never even breach the top 10 (or 20.)
The problem is clusters scale far from linearly (or if working off of your mathematical extrapolation, a step-based linear profile.) That means, unless the current results for the 256 CPUs was far from optimized, there's a fairly good chance the final cluster may never even breach the top 10 (or 20.)
But until we actually see a Linpack score on the whole cluster, all we can do is extrapolate.
Show me a figure you think works better.
It seems that there is alot of confusion (most of which can be attributed to my own ineptitude in providing the wrong page) concerning exactly where the Virginia Tech supercomputer results are posted.
In my recklessness, I had inadvertently provided you with the wrong page number. The Virginia Tech supercomputer result is actually located on page 53. As Abstract pointed out, there is indeed another G5 cluster located on page 55 of the pdf. This discreprency would probably be the source of the confusion.
So, once again, the Virginia Tech supercomputer Linpack HPC result is located on page 53, there is a second much smaller G5 cluster located on page 55. I have also taken the liberty to post Vtech results on this thread as well as provide a definition for all the terms (as requested by Abstract).
Legend
Rmax: sustained performance for the largest problem run
Nmax: size of the largest problem run
Rpeak: theoretical peak performance of supercomputer
I will be only focusing on Rmax since it's really what determines the supercomputer's rank on the top 500.
Virginia Tech supercomputer:
2112 G5s@2.0 GHz used
Rmax: 7417 GFLOP/S
Rank: #6
Linux NetworX supercomputer
2304 Xeons@2.4 GHz used
Rmax: 7634 GFLOP/S
Rank: #5
ASCI Q
8160 Alpha EV68@1.25 GHz used
Rmax: 13880 GFLOP/S
Rank: #2
Earth Simulator
5120 ??? processors
Rmax: 35860 GFLOP/S
Rank: #1
In my recklessness, I had inadvertently provided you with the wrong page number. The Virginia Tech supercomputer result is actually located on page 53. As Abstract pointed out, there is indeed another G5 cluster located on page 55 of the pdf. This discreprency would probably be the source of the confusion.
So, once again, the Virginia Tech supercomputer Linpack HPC result is located on page 53, there is a second much smaller G5 cluster located on page 55. I have also taken the liberty to post Vtech results on this thread as well as provide a definition for all the terms (as requested by Abstract).
Legend
Rmax: sustained performance for the largest problem run
Nmax: size of the largest problem run
Rpeak: theoretical peak performance of supercomputer
I will be only focusing on Rmax since it's really what determines the supercomputer's rank on the top 500.
Virginia Tech supercomputer:
2112 G5s@2.0 GHz used
Rmax: 7417 GFLOP/S
Rank: #6
Linux NetworX supercomputer
2304 Xeons@2.4 GHz used
Rmax: 7634 GFLOP/S
Rank: #5
ASCI Q
8160 Alpha EV68@1.25 GHz used
Rmax: 13880 GFLOP/S
Rank: #2
Earth Simulator
5120 ??? processors
Rmax: 35860 GFLOP/S
Rank: #1
Thanks for the clarification.
Something interesting to note:
# of CPUs = 2112
So not all 1100 G5s are in the cluster. Playing the linear extrapolation game again gives a score of about 7726 Gflops, putting it in 4th, one Gflop behind one of the AlphaServers with 4096 1.25GHz CPUs.
Something interesting to note:
# of CPUs = 2112
So not all 1100 G5s are in the cluster. Playing the linear extrapolation game again gives a score of about 7726 Gflops, putting it in 4th, one Gflop behind one of the AlphaServers with 4096 1.25GHz CPUs.