GFLOPs are out.
How many fruit flies per second can it get
How many fruit flies per second can it get
September 26, 2026 at 6:20 AM
GFLOPs are out.
How many fruit flies per second can it get
How many fruit flies per second can it get
M4 efficiency is stunning. First machine I've tested getting over 5 Gflops/W with HPL, and in fact it goes beyond 6 Gflops/W... with fan only spinning up to a quiet 2500 rpm.
Maybe this is why Qualcomm abandoned their marquee Dev Kit—it is thoroughly trounced by M4.
Maybe this is why Qualcomm abandoned their marquee Dev Kit—it is thoroughly trounced by M4.
November 12, 2024 at 8:11 PM
M4 efficiency is stunning. First machine I've tested getting over 5 Gflops/W with HPL, and in fact it goes beyond 6 Gflops/W... with fan only spinning up to a quiet 2500 rpm.
Maybe this is why Qualcomm abandoned their marquee Dev Kit—it is thoroughly trounced by M4.
Maybe this is why Qualcomm abandoned their marquee Dev Kit—it is thoroughly trounced by M4.
The Cray Y-MP supercomputer they’re sitting on could do about 2.7 Gigaflops. By comparison, a cheapo Raspberry Pi they give to kids for fun throwaway projects that can fit in your hand benchmarks at 25 GFLOPS for a reasonable $50.
July 10, 2026 at 2:34 AM
The Cray Y-MP supercomputer they’re sitting on could do about 2.7 Gigaflops. By comparison, a cheapo Raspberry Pi they give to kids for fun throwaway projects that can fit in your hand benchmarks at 25 GFLOPS for a reasonable $50.
El superordinador més ràpid ha passat de 124 GFLOPS el 1993 a 1,8 mil milions de GFLOPS el 2025, 14,6 milions de vegades més. Ara es resolen en hores problemes que abans haurien ocupat màquines durant anys.
#dataviz
#dataviz
September 25, 2026 at 5:57 AM
El superordinador més ràpid ha passat de 124 GFLOPS el 1993 a 1,8 mil milions de GFLOPS el 2025, 14,6 milions de vegades més. Ara es resolen en hores problemes que abans haurien ocupat màquines durant anys.
#dataviz
#dataviz
Wild M4 Max SME stats:
~850 GFLOPS DGEMM
3400+ GFLOPS SGEMM
Waiting for power data, but looks like ~12W system
~850 GFLOPS DGEMM
3400+ GFLOPS SGEMM
Waiting for power data, but looks like ~12W system
November 11, 2024 at 8:50 PM
Wild M4 Max SME stats:
~850 GFLOPS DGEMM
3400+ GFLOPS SGEMM
Waiting for power data, but looks like ~12W system
~850 GFLOPS DGEMM
3400+ GFLOPS SGEMM
Waiting for power data, but looks like ~12W system
New performance leadership for the cheapest way to get ~1 TFLOP of DGEMM!
Intel's new B580 reaches over 900 GFLOPS DGEMM at 250 USD!
www.phoronix.com/review/intel...
Intel's new B580 reaches over 900 GFLOPS DGEMM at 250 USD!
www.phoronix.com/review/intel...
Intel Arc B580 Delivers Promising Linux GPU Compute Potential For Battlemage Review
.
www.phoronix.com
December 12, 2024 at 2:33 PM
New performance leadership for the cheapest way to get ~1 TFLOP of DGEMM!
Intel's new B580 reaches over 900 GFLOPS DGEMM at 250 USD!
www.phoronix.com/review/intel...
Intel's new B580 reaches over 900 GFLOPS DGEMM at 250 USD!
www.phoronix.com/review/intel...
#Green500: Europe takes the top 3 spots
1. JEDI at FZJ Germany retains the crown at 72.73 GFlops/Watt
2. ROMEO-2025 - new system in France with identical architecture to JEDI but twice as large
3. Adastra 2 - new Cray/AMD system at GENCI, France
top500.org/lists/green5...
#SC24 #HPC
1. JEDI at FZJ Germany retains the crown at 72.73 GFlops/Watt
2. ROMEO-2025 - new system in France with identical architecture to JEDI but twice as large
3. Adastra 2 - new Cray/AMD system at GENCI, France
top500.org/lists/green5...
#SC24 #HPC
November 18, 2024 at 7:20 PM
#Green500: Europe takes the top 3 spots
1. JEDI at FZJ Germany retains the crown at 72.73 GFlops/Watt
2. ROMEO-2025 - new system in France with identical architecture to JEDI but twice as large
3. Adastra 2 - new Cray/AMD system at GENCI, France
top500.org/lists/green5...
#SC24 #HPC
1. JEDI at FZJ Germany retains the crown at 72.73 GFlops/Watt
2. ROMEO-2025 - new system in France with identical architecture to JEDI but twice as large
3. Adastra 2 - new Cray/AMD system at GENCI, France
top500.org/lists/green5...
#SC24 #HPC
Cray 2 vector supercomputer from 1985. Fastest computer of that time, at 1.9 GFLOPS peak performance.
For comparison, a Qualcomm Snapdragon 888 in new cell phone is at about 1720 GFLOPS performance.
For comparison, a Qualcomm Snapdragon 888 in new cell phone is at about 1720 GFLOPS performance.
May 9, 2025 at 4:47 AM
Cray 2 vector supercomputer from 1985. Fastest computer of that time, at 1.9 GFLOPS peak performance.
For comparison, a Qualcomm Snapdragon 888 in new cell phone is at about 1720 GFLOPS performance.
For comparison, a Qualcomm Snapdragon 888 in new cell phone is at about 1720 GFLOPS performance.
IBM's Deep Blue supercomputer defeated a chess world champion in 1997 using 32 processors and 11.38 GFLOPS of processing power.
January 8, 2025 at 1:46 PM
IBM's Deep Blue supercomputer defeated a chess world champion in 1997 using 32 processors and 11.38 GFLOPS of processing power.
M4 맥미니의 효율성
- HPL 벤치마크에서 6.74 Gflops/W를 달성 (M1 Max 맥 스튜디오의 4.00 Gflops/W)
- 대기 전력 소비량 또한 라즈베리 파이와 비슷한 3-4W 수준
M4 Mac mini's efficiency is incredible | Jeff Geerling
www.jeffgeerling.com/blog/2024/m4...
- HPL 벤치마크에서 6.74 Gflops/W를 달성 (M1 Max 맥 스튜디오의 4.00 Gflops/W)
- 대기 전력 소비량 또한 라즈베리 파이와 비슷한 3-4W 수준
M4 Mac mini's efficiency is incredible | Jeff Geerling
www.jeffgeerling.com/blog/2024/m4...
M4 Mac mini's efficiency is incredible | Jeff Geerling
www.jeffgeerling.com
November 14, 2024 at 2:19 PM
M4 맥미니의 효율성
- HPL 벤치마크에서 6.74 Gflops/W를 달성 (M1 Max 맥 스튜디오의 4.00 Gflops/W)
- 대기 전력 소비량 또한 라즈베리 파이와 비슷한 3-4W 수준
M4 Mac mini's efficiency is incredible | Jeff Geerling
www.jeffgeerling.com/blog/2024/m4...
- HPL 벤치마크에서 6.74 Gflops/W를 달성 (M1 Max 맥 스튜디오의 4.00 Gflops/W)
- 대기 전력 소비량 또한 라즈베리 파이와 비슷한 3-4W 수준
M4 Mac mini's efficiency is incredible | Jeff Geerling
www.jeffgeerling.com/blog/2024/m4...
AraXL was implemented in a 22nm technology node, achieving a performance peak of 146 GFLOPs on computation-intensive HPC/ML kernels (>99% FPU utilization) and energy efficiency of 40.1 GFLOPs/W (1.15 GHz, TT, 0.8V), with only 3.8x the area of a 16-lane instance.
January 20, 2025 at 10:38 AM
AraXL was implemented in a 22nm technology node, achieving a performance peak of 146 GFLOPs on computation-intensive HPC/ML kernels (>99% FPU utilization) and energy efficiency of 40.1 GFLOPs/W (1.15 GHz, TT, 0.8V), with only 3.8x the area of a 16-lane instance.
Steam Deck is a $1000 console with teraflops of gigashits
June 15, 2026 at 11:43 PM
Steam Deck is a $1000 console with teraflops of gigashits
@sjvn.bsky.social @estherschindler.bsky.social @robpegoraro.com
Seems a good time to update this.
Original Cray Y-MP, highest-end configuration, 1988:
32 bit, 2.3 GFLOPS, 512 MB RAM, 4 GB SSD, $5 million
iPhone 16, lowest end configuration, 2024:
64 bit, 1700 GFLOPS, 8 GB RAM, 128 GB SSD, $900
Seems a good time to update this.
Original Cray Y-MP, highest-end configuration, 1988:
32 bit, 2.3 GFLOPS, 512 MB RAM, 4 GB SSD, $5 million
iPhone 16, lowest end configuration, 2024:
64 bit, 1700 GFLOPS, 8 GB RAM, 128 GB SSD, $900
September 12, 2024 at 11:03 PM
@sjvn.bsky.social @estherschindler.bsky.social @robpegoraro.com
Seems a good time to update this.
Original Cray Y-MP, highest-end configuration, 1988:
32 bit, 2.3 GFLOPS, 512 MB RAM, 4 GB SSD, $5 million
iPhone 16, lowest end configuration, 2024:
64 bit, 1700 GFLOPS, 8 GB RAM, 128 GB SSD, $900
Seems a good time to update this.
Original Cray Y-MP, highest-end configuration, 1988:
32 bit, 2.3 GFLOPS, 512 MB RAM, 4 GB SSD, $5 million
iPhone 16, lowest end configuration, 2024:
64 bit, 1700 GFLOPS, 8 GB RAM, 128 GB SSD, $900
Por exemplo, o supercomputador JEDI tem capacidade de processamento de 4.5 PetaFlop/s.
· Desempenho do JEDI em GFLOPS: 4.5 PFlop/s = 4.5 × 1,000,000 GFLOPS = 4,500,000 GFLOPS.
· Número de A23s necessários: 4,500,000 GFLOPS / 85 GFLOPS ≈ 52,941 A23.
Um cluster de 50 mil Galaxy A23
· Desempenho do JEDI em GFLOPS: 4.5 PFlop/s = 4.5 × 1,000,000 GFLOPS = 4,500,000 GFLOPS.
· Número de A23s necessários: 4,500,000 GFLOPS / 85 GFLOPS ≈ 52,941 A23.
Um cluster de 50 mil Galaxy A23
June 9, 2026 at 10:57 AM
Por exemplo, o supercomputador JEDI tem capacidade de processamento de 4.5 PetaFlop/s.
· Desempenho do JEDI em GFLOPS: 4.5 PFlop/s = 4.5 × 1,000,000 GFLOPS = 4,500,000 GFLOPS.
· Número de A23s necessários: 4,500,000 GFLOPS / 85 GFLOPS ≈ 52,941 A23.
Um cluster de 50 mil Galaxy A23
· Desempenho do JEDI em GFLOPS: 4.5 PFlop/s = 4.5 × 1,000,000 GFLOPS = 4,500,000 GFLOPS.
· Número de A23s necessários: 4,500,000 GFLOPS / 85 GFLOPS ≈ 52,941 A23.
Um cluster de 50 mil Galaxy A23
Kolem roku 1990 byl nejvýkonnější superpočítač na světě Cray Y-MP C90, který dosahoval výkonu přibližně 2.4 GFLOPS (neboli miliardy operací s plovoucí desetinnou čárkou za sekundu) a vážil 5500 kg.
Současný iPhone 15 Pro je přibližně 1.000x výkonnější. Váží 30.000x méně.
Současný iPhone 15 Pro je přibližně 1.000x výkonnější. Váží 30.000x méně.
June 26, 2024 at 9:36 PM
Kolem roku 1990 byl nejvýkonnější superpočítač na světě Cray Y-MP C90, který dosahoval výkonu přibližně 2.4 GFLOPS (neboli miliardy operací s plovoucí desetinnou čárkou za sekundu) a vážil 5500 kg.
Současný iPhone 15 Pro je přibližně 1.000x výkonnější. Váží 30.000x méně.
Současný iPhone 15 Pro je přibližně 1.000x výkonnější. Váží 30.000x méně.
Day 23 of #30DayChartChallenge: Log-Scale.
Interactive version for all the labels: www.datawrapper.de/_/mun1B/
Data: thegamingsetup.com/console-powe...
Interactive version for all the labels: www.datawrapper.de/_/mun1B/
Data: thegamingsetup.com/console-powe...
April 23, 2025 at 6:58 PM
Day 23 of #30DayChartChallenge: Log-Scale.
Interactive version for all the labels: www.datawrapper.de/_/mun1B/
Data: thegamingsetup.com/console-powe...
Interactive version for all the labels: www.datawrapper.de/_/mun1B/
Data: thegamingsetup.com/console-powe...
I got the math wrong. Irs just four orders magnitude. 3 GFLOPS to 30 TFLOPS or 30,000 GFLOPS
July 10, 2026 at 11:41 PM
I got the math wrong. Irs just four orders magnitude. 3 GFLOPS to 30 TFLOPS or 30,000 GFLOPS
Really missed an opportunity not naming it the cray cray
August 7, 2025 at 5:02 PM
Really missed an opportunity not naming it the cray cray
Literally saw a Cray-1 yesterday in the Valladolid Science Museum. An incredible piece of computing history, and my nearly 3 year old S22 out-performs in raw computing power by ~15000x
2500 GFLOPS vs 160 MFLOPS
2500 GFLOPS vs 160 MFLOPS
September 20, 2024 at 3:03 PM
Literally saw a Cray-1 yesterday in the Valladolid Science Museum. An incredible piece of computing history, and my nearly 3 year old S22 out-performs in raw computing power by ~15000x
2500 GFLOPS vs 160 MFLOPS
2500 GFLOPS vs 160 MFLOPS
This page (forums.raspberrypi.com/viewtopic.ph...) suggests
fp16-scalar = 14.47 GFLOPS
fp16-vec4 = 50.99 GFLOPS
fp16-scalar = 14.47 GFLOPS
fp16-vec4 = 50.99 GFLOPS
RPi 5 Vulkan performance - Raspberry Pi ForumsRaspberry Pi
forums.raspberrypi.com
December 3, 2024 at 10:48 PM
This page (forums.raspberrypi.com/viewtopic.ph...) suggests
fp16-scalar = 14.47 GFLOPS
fp16-vec4 = 50.99 GFLOPS
fp16-scalar = 14.47 GFLOPS
fp16-vec4 = 50.99 GFLOPS
The processor had four parallel vector processing units, and when carefully programmed could reach nearly 2 GFLOPS of throughput. This is of course nothing these days, a Raspberry Pi can run circles around it but I'd much rather have a Cray to play around with.
April 9, 2025 at 7:33 AM
The processor had four parallel vector processing units, and when carefully programmed could reach nearly 2 GFLOPS of throughput. This is of course nothing these days, a Raspberry Pi can run circles around it but I'd much rather have a Cray to play around with.
85.3 GFlops: Optimizing FP32 Matrix Multiplication on a Single AMD Zen 3 Core | Discussion
GitHub - houslast3/85.30-GFLOPS-Single-Core-FP32-Matrix-Multiplication-on-AMD-Zen-3
Contribute to houslast3/85.30-GFLOPS-Single-Core-FP32-Matrix-Multiplication-on-AMD-Zen-3 development by creating an account on GitHub.
github.com
July 20, 2026 at 8:40 PM
85.3 GFlops: Optimizing FP32 Matrix Multiplication on a Single AMD Zen 3 Core | Discussion