The answer is NO. IBM Roadrunner was tested with en.wikipedia.org/wiki/LINPACK... which using double-precision floating-point format aka FP64
Nvidia claims "1 petaflop of AI performance at FP4 precision" which is nowhere close
Nvidia claims "1 petaflop of AI performance at FP4 precision" which is nowhere close
LINPACK benchmarks - Wikipedia
en.wikipedia.org
January 9, 2025 at 12:53 AM
The answer is NO. IBM Roadrunner was tested with en.wikipedia.org/wiki/LINPACK... which using double-precision floating-point format aka FP64
Nvidia claims "1 petaflop of AI performance at FP4 precision" which is nowhere close
Nvidia claims "1 petaflop of AI performance at FP4 precision" which is nowhere close
Yes, it's a QR decomposition, a bespoke one modified by Ross from LINPACK fortran code.
The key difference from LINPACK (in src/appl/dqrdc2.f) is that it pivots only when really needed, so that computing sequential "this beta adjusted for the previous ones" is more straightforward.
The key difference from LINPACK (in src/appl/dqrdc2.f) is that it pivots only when really needed, so that computing sequential "this beta adjusted for the previous ones" is more straightforward.
June 20, 2025 at 3:45 AM
Yes, it's a QR decomposition, a bespoke one modified by Ross from LINPACK fortran code.
The key difference from LINPACK (in src/appl/dqrdc2.f) is that it pivots only when really needed, so that computing sequential "this beta adjusted for the previous ones" is more straightforward.
The key difference from LINPACK (in src/appl/dqrdc2.f) is that it pivots only when really needed, so that computing sequential "this beta adjusted for the previous ones" is more straightforward.
This is my attitude to reading LINPACK in the original Fortran
And this is why you should never *read* Shakespeare plays. You need a cast that really understands the play to signal all the jokes, and a bit of modernisation doesn't hurt either
Most "classics" are boring as fuck to a modern reader in no small part because the context of the story is not understood by them and so many of the allusions and "inside jokes" known to then-contemporary readers have been lost in time and in translation
October 13, 2025 at 11:33 PM
This is my attitude to reading LINPACK in the original Fortran
Thanks Microsoft for restarting my PC to again install Teams without my permission. Not like I had a pair of SSH sessions running HP-Linpack benchmark overnight... Killed the sessions before completing, so now I have to re-run the 3+ hour test on both servers.
November 13, 2024 at 3:47 PM
Thanks Microsoft for restarting my PC to again install Teams without my permission. Not like I had a pair of SSH sessions running HP-Linpack benchmark overnight... Killed the sessions before completing, so now I have to re-run the 3+ hour test on both servers.
Why #Supercomputer #Benchmarking Is So Important
by @agamsh.bsky.social in HPCWire
I expect healthy disagreement on some quotes in this article
“LINPACK measures the time-to-science and value extracted from hardware” is misleading
www.hpcwire.com/2024/11/25/w...
#HPC #SC24
by @agamsh.bsky.social in HPCWire
I expect healthy disagreement on some quotes in this article
“LINPACK measures the time-to-science and value extracted from hardware” is misleading
www.hpcwire.com/2024/11/25/w...
#HPC #SC24
Why Supercomputer Benchmarking Is So Important
SC2024 had a few sessions talking about the importance of benchmarking systems. The relevance of LINPACK — which is used to rank the Top500 — has been a hot topic […]
www.hpcwire.com
November 26, 2024 at 11:39 AM
Why #Supercomputer #Benchmarking Is So Important
by @agamsh.bsky.social in HPCWire
I expect healthy disagreement on some quotes in this article
“LINPACK measures the time-to-science and value extracted from hardware” is misleading
www.hpcwire.com/2024/11/25/w...
#HPC #SC24
by @agamsh.bsky.social in HPCWire
I expect healthy disagreement on some quotes in this article
“LINPACK measures the time-to-science and value extracted from hardware” is misleading
www.hpcwire.com/2024/11/25/w...
#HPC #SC24
Roadrunner was an unmitigated disaster for LANL because nobody could program the Cells. They ran linpack on it and then everyone ran their simulations on the AMD head nodes until it was decommissioned.
January 9, 2025 at 1:26 AM
Roadrunner was an unmitigated disaster for LANL because nobody could program the Cells. They ran linpack on it and then everyone ran their simulations on the AMD head nodes until it was decommissioned.
Memtest86+ to test RAM stability.
Furmark to stress test GPU and CPU.
Prome95 and Linpack to error check CPU. (Intel Burn Test, if I recall correctly.)
Furmark to stress test GPU and CPU.
Prome95 and Linpack to error check CPU. (Intel Burn Test, if I recall correctly.)
December 19, 2024 at 8:00 AM
Memtest86+ to test RAM stability.
Furmark to stress test GPU and CPU.
Prome95 and Linpack to error check CPU. (Intel Burn Test, if I recall correctly.)
Furmark to stress test GPU and CPU.
Prome95 and Linpack to error check CPU. (Intel Burn Test, if I recall correctly.)
I think most of the remaining Fortran is older than that. Eg the QR decomposition inside lm() dates back to LINPACK (70s & 80s) with light modifications to only pivot on singularities
August 5, 2025 at 7:56 PM
I think most of the remaining Fortran is older than that. Eg the QR decomposition inside lm() dates back to LINPACK (70s & 80s) with light modifications to only pivot on singularities
Prozessor-Stresstest: Linpack Xtreme 1.1.6 setzt auf HWMonitor 1.55 #LinpackXtreme
Prozessor-Stresstest: Linpack Xtreme 1.1.6 setzt auf HWMonitor 1.55
Der Stabilitäts- und Stresstest für CPUs Linpack Xtreme ist in aktualisierter Version 1.1.6 mit integriertem HWMonitor 1.55 erschienen.
www.computerbase.de
November 18, 2024 at 6:30 AM
Prozessor-Stresstest: Linpack Xtreme 1.1.6 setzt auf HWMonitor 1.55 #LinpackXtreme
A new Linpack/HPL benchmark rewritten in Python. One readable file, numpy as the only dep — and at the end it tells you when your laptop would have been a supercomputer. Mine was world #1 in November 1995 🤯
github.com/dzaragoza/hpl
github.com/dzaragoza/hpl
GitHub - dzaragoza/hpl: The Linpack/HPL benchmark in readable Python — LAPACK through NumPy, with a TOP500 "when was my laptop a supercomputer?" calibration.
The Linpack/HPL benchmark in readable Python — LAPACK through NumPy, with a TOP500 "when was my laptop a supercomputer?" calibration. - dzaragoza/hpl
github.com
September 20, 2026 at 10:58 PM
A new Linpack/HPL benchmark rewritten in Python. One readable file, numpy as the only dep — and at the end it tells you when your laptop would have been a supercomputer. Mine was world #1 in November 1995 🤯
github.com/dzaragoza/hpl
github.com/dzaragoza/hpl
- chemistry: you couldn't devise worse notation if you had an LLM and half of OpenAI's compute
- "AI": imagine a LINPACK fanboy who can't write Fortran, and then half of them are fascists
- PLT: it wouldn't kill you guys to get some sun. and please, talk to the compiler guys
- "AI": imagine a LINPACK fanboy who can't write Fortran, and then half of them are fascists
- PLT: it wouldn't kill you guys to get some sun. and please, talk to the compiler guys
May 10, 2024 at 8:22 PM
- chemistry: you couldn't devise worse notation if you had an LLM and half of OpenAI's compute
- "AI": imagine a LINPACK fanboy who can't write Fortran, and then half of them are fascists
- PLT: it wouldn't kill you guys to get some sun. and please, talk to the compiler guys
- "AI": imagine a LINPACK fanboy who can't write Fortran, and then half of them are fascists
- PLT: it wouldn't kill you guys to get some sun. and please, talk to the compiler guys
In loving memory of Cleve B. #Moler [°Aug. 17, 1939 – May 20, 2026]🕯️😌🌹
#computerscience #numericalanalysis #2014IEEEJohnvonNeumannMedal #MATLAB #MathWorks #2012ComputerPioneerAward #LINPACK #EISPACK #softwarelibrary #programminglanguage #Fortran #JackLittle #GeneHGolub #CleveBarryMoler #CleveMoler
#computerscience #numericalanalysis #2014IEEEJohnvonNeumannMedal #MATLAB #MathWorks #2012ComputerPioneerAward #LINPACK #EISPACK #softwarelibrary #programminglanguage #Fortran #JackLittle #GeneHGolub #CleveBarryMoler #CleveMoler
June 28, 2026 at 8:57 AM
In loving memory of Cleve B. #Moler [°Aug. 17, 1939 – May 20, 2026]🕯️😌🌹
#computerscience #numericalanalysis #2014IEEEJohnvonNeumannMedal #MATLAB #MathWorks #2012ComputerPioneerAward #LINPACK #EISPACK #softwarelibrary #programminglanguage #Fortran #JackLittle #GeneHGolub #CleveBarryMoler #CleveMoler
#computerscience #numericalanalysis #2014IEEEJohnvonNeumannMedal #MATLAB #MathWorks #2012ComputerPioneerAward #LINPACK #EISPACK #softwarelibrary #programminglanguage #Fortran #JackLittle #GeneHGolub #CleveBarryMoler #CleveMoler
So which cloud provider carved out a big enough GPU partition long enough to get a Top 10 ranking Linpack score this time?
Oracle? Google? AWS? Azure again?
Oracle? Google? AWS? Azure again?
November 13, 2024 at 11:35 PM
So which cloud provider carved out a big enough GPU partition long enough to get a Top 10 ranking Linpack score this time?
Oracle? Google? AWS? Azure again?
Oracle? Google? AWS? Azure again?
I’m currently working on an article about the Radxa Fogwise Airbox Q900 (with its 8 “big” Qualcomm A78C cores and 36GB of RAM) and it’s an efficient little guy..
The CIX P1 topped the Linpack Gflops/W table for about 2 weeks before this arrived 😅
sbc.compare/linpack
The CIX P1 topped the Linpack Gflops/W table for about 2 weeks before this arrived 😅
sbc.compare/linpack
December 22, 2025 at 1:18 PM
I’m currently working on an article about the Radxa Fogwise Airbox Q900 (with its 8 “big” Qualcomm A78C cores and 36GB of RAM) and it’s an efficient little guy..
The CIX P1 topped the Linpack Gflops/W table for about 2 weeks before this arrived 😅
sbc.compare/linpack
The CIX P1 topped the Linpack Gflops/W table for about 2 weeks before this arrived 😅
sbc.compare/linpack
Doug Bates schooled me in the early Rcpp(Eigen) days. Do NOT write OLS code as (X'X)^-1 X'y in Eigen (or Armadillo) because of the difference between statistical meaning and numerical analysis: R uses LINPACK and pivoting, not LAPACK (as Eigen/Arma do). See his extended fastLm example in RcppEigen.
September 27, 2025 at 2:02 PM
Doug Bates schooled me in the early Rcpp(Eigen) days. Do NOT write OLS code as (X'X)^-1 X'y in Eigen (or Armadillo) because of the difference between statistical meaning and numerical analysis: R uses LINPACK and pivoting, not LAPACK (as Eigen/Arma do). See his extended fastLm example in RcppEigen.
Testes que passaram com no máximo 40 graus na memória :
yCruncher 5h+ ✅
Tm5 PCB Destroyer / usmus v3 = 4h✅
Karhu = 4h✅
Linpack Extreme = 60 lines✅
Só é possível graças a placa mãe, por isso sempre digo e volto a dizer que a Placa mãe é a coisa mais importante do PC!
yCruncher 5h+ ✅
Tm5 PCB Destroyer / usmus v3 = 4h✅
Karhu = 4h✅
Linpack Extreme = 60 lines✅
Só é possível graças a placa mãe, por isso sempre digo e volto a dizer que a Placa mãe é a coisa mais importante do PC!
September 7, 2024 at 3:21 PM
Testes que passaram com no máximo 40 graus na memória :
yCruncher 5h+ ✅
Tm5 PCB Destroyer / usmus v3 = 4h✅
Karhu = 4h✅
Linpack Extreme = 60 lines✅
Só é possível graças a placa mãe, por isso sempre digo e volto a dizer que a Placa mãe é a coisa mais importante do PC!
yCruncher 5h+ ✅
Tm5 PCB Destroyer / usmus v3 = 4h✅
Karhu = 4h✅
Linpack Extreme = 60 lines✅
Só é possível graças a placa mãe, por isso sempre digo e volto a dizer que a Placa mãe é a coisa mais importante do PC!
The machine in question scored about 0.974 teraflops on linpack; the iPhone 16e in my pocket is ballpark 1-2 teraflops so I don't think it's an unreasonable statement insofar as any such comparison could be described as reasonable.
May 27, 2026 at 9:19 AM
The machine in question scored about 0.974 teraflops on linpack; the iPhone 16e in my pocket is ballpark 1-2 teraflops so I don't think it's an unreasonable statement insofar as any such comparison could be described as reasonable.
Linux is actually used on many supercomputers and web servers.
The world's fastest supercomputer was tested using Linux software.
interestingengineering.com/innovation/u...
The world's fastest supercomputer was tested using Linux software.
interestingengineering.com/innovation/u...
US: World’s fastest supercomputer with 11 million cores powers up
'El Capitan' is the world's fastest supercomputer, achieving 1.742 exaFLOPS in the High-Performance Linpack benchmark.
interestingengineering.com
January 24, 2025 at 11:26 PM
Linux is actually used on many supercomputers and web servers.
The world's fastest supercomputer was tested using Linux software.
interestingengineering.com/innovation/u...
The world's fastest supercomputer was tested using Linux software.
interestingengineering.com/innovation/u...
The Single Board Computer Cluster competition (sbcc.io) for 2025 has concluded with the two teams from UC San Diego taking 1st and 3rd places overall and the team from Texas Tech University in 2nd place overall and 1st in HP Linpack results! single-board-cluster-competition.github.io/sbcc25-compe...
SBCC 2025
SBCC 2025 is a Single Board Cluster Competition made for students by students
sbcc.io
April 13, 2025 at 1:57 AM
The Single Board Computer Cluster competition (sbcc.io) for 2025 has concluded with the two teams from UC San Diego taking 1st and 3rd places overall and the team from Texas Tech University in 2nd place overall and 1st in HP Linpack results! single-board-cluster-competition.github.io/sbcc25-compe...
Prozessor-Stresstest: Linpack Xtreme 1.1.6 setzt auf HWMonitor 1.55 (Update) #LinpackXtreme
Prozessor-Stresstest: Linpack Xtreme 1.1.6 setzt auf HWMonitor 1.55
Der Stabilitäts- und Stresstest für CPUs Linpack Xtreme ist in aktualisierter Version 1.1.6 mit integriertem HWMonitor 1.55 erschienen.
www.computerbase.de
November 19, 2024 at 12:13 AM
Prozessor-Stresstest: Linpack Xtreme 1.1.6 setzt auf HWMonitor 1.55 (Update) #LinpackXtreme
16bitセンセーションというアニメでPC-9821(Ra?)を8台負荷分散させた「スーパーオクタコア」の絵ずらが出てきたので,うちも98を積んで真似てみた.(ただし4台が限界)
ちなみに作中1999年夏の世界最速のスパコンはアメリカの「ASCI Red」で、
"Pentium Pro 333MHz"が"9,632個"並列稼働させていて、Linpackのベンチマーク結果が"2,379GFlop/s".
比較として,うちの"Core i7 9700K"が"369GFlops"なので当時のスパコンの圧勝だったり.
スパコンって速いデスネ.
ちなみに作中1999年夏の世界最速のスパコンはアメリカの「ASCI Red」で、
"Pentium Pro 333MHz"が"9,632個"並列稼働させていて、Linpackのベンチマーク結果が"2,379GFlop/s".
比較として,うちの"Core i7 9700K"が"369GFlops"なので当時のスパコンの圧勝だったり.
スパコンって速いデスネ.
April 21, 2024 at 2:06 AM
16bitセンセーションというアニメでPC-9821(Ra?)を8台負荷分散させた「スーパーオクタコア」の絵ずらが出てきたので,うちも98を積んで真似てみた.(ただし4台が限界)
ちなみに作中1999年夏の世界最速のスパコンはアメリカの「ASCI Red」で、
"Pentium Pro 333MHz"が"9,632個"並列稼働させていて、Linpackのベンチマーク結果が"2,379GFlop/s".
比較として,うちの"Core i7 9700K"が"369GFlops"なので当時のスパコンの圧勝だったり.
スパコンって速いデスネ.
ちなみに作中1999年夏の世界最速のスパコンはアメリカの「ASCI Red」で、
"Pentium Pro 333MHz"が"9,632個"並列稼働させていて、Linpackのベンチマーク結果が"2,379GFlop/s".
比較として,うちの"Core i7 9700K"が"369GFlops"なので当時のスパコンの圧勝だったり.
スパコンって速いデスネ.
this is my very favourite comment from the R code base: github.com/r-devel/r-sv... (this more than 28 years ago - I didn't feel like doing the archaeology to trace the code back through a file reorganization that was done in 1998 ...)
June 20, 2025 at 1:08 PM
this is my very favourite comment from the R code base: github.com/r-devel/r-sv... (this more than 28 years ago - I didn't feel like doing the archaeology to trace the code back through a file reorganization that was done in 1998 ...)