#AMD #EPYC #4565P & #4585PX #Benchmarks Against #Xeon #6369P
For "conventional" #server workloads like web serving and databases, the EPYC 4005 series dominates.
With up to 16C/32TH, #AVX512, DDR5-5600 memory and other advantages, the EPYC 4005 series is the very easy answer for those that may be looking for affordable #HPC
The AMD #EPYC4005 series #CPU deliver excellent generational uplift over the EPYC 4004 series and outright obliterating the #Xeon6300 series
https://www.phoronix.com/review/amd-epyc-4585px-4565p-benchmarks

You know how sometimes a little hobby side-project can get a bit out of hand? An unexpected performance regression on speed.python.org that only showed up on GCC 5 (and 7) led me to set up more rigorous tracking of Python performance when using different compilers. I'm still backfilling data but I think it's pretty awesome to see how much, and how consistently, free-threaded Python performance has improved since 3.13:

https://github.com/Yhg1s/python-benchmarking-public

New Essay

"The Intelligent AI Coin: A Thought Experiment"

Open Access here: https://seanfobbe.com/posts/2025-02-21_intelligent-ai-coin-thought-experiment/

Recent years have seen a concerning trend towards normalizing decisionmaking by Large Language Models (LLM), including in the adoption of legislation, the writing of judicial opinions and the routine administration of the rule of law. AI agents acting on behalf of human principals are supposed to lead us into a new age of productivity and convenience. The eloquence of AI-generated text and the narrative of super-human intelligence invite us to trust these systems more than we have trusted any human or algorithm ever before.

It is difficult to know whether a machine is actually intelligent because of problems with construct validity, plagiarism, reproducibility and transferability in AI benchmarks. Most people will either have to personally evaluate the usefulness of AI tools against the benchmark of their own lived experience or be forced to trust an expert.

To explain this conundrum I propose the Intelligent AI Coin Thought Experiment and discuss four objections: the restriction of agents to low-value decisions, making AI decisionmakers open source, adding a human-in-the-loop and the general limits of trust in human agents.

@histodons @politicalscience

Guess what? Linux faster than Windows increasingly | '25 Gaming benchmarks

Enjoy the show!

https://youtu.be/D45AknAsIPw

#NVIDIA #GeForce #RTX5090 #Linux #GPU Compute Performance #Benchmarks
When taking geo mean across 60+ benchmarks of #CUDA / #OptiX / #OpenCL / #Vulkan Compute, the GeForce RTX 5090 was delivering 1.42x the performance of GeForce #RTX4090. On performance-per-Watt GeForce RTX 5090 tended to deliver similar power efficiency to the RTX 4080/4090 graphics cards.
GeForce RTX 5090 Founders Edition was running cooler than many of the other Founders Edition graphics cards tested.
https://www.phoronix.com/review/nvidia-geforce-rtx5090-linux

#RedHat #EnterpriseLinux 10 Beta Performance Looks Great - Initial #RHEL 9 vs. #RHEL10 #Benchmarks
#RHEL10Beta is on Linux 6.11 (though #CentOSStream 10 has moved to# Linux 6.12 LTS), GCC 14.2 is default compiler, XFS the default file-system, and there is a wealth of other package updates like moving to Python 3.12 by default and OpenJDK Java 21 as its default version. RHEL 10 also goes Wayland-only with GNOME Shell.
10 Beta was on average 10% faster than current RHEL 9.5
https://www.phoronix.com/review/rhel-10-beta-benchmarks

Has anyone created technical #MachineLearnjng #benchmarks for generated code #accessibility?

#AMD #EPYC9655 #Benchmarks Show The Terrific Generational Gains With 5th Gen #EPYC Review
Consistent ~1.4x uplift from #EPYC9654 to EPYC 9655 was all more impressive when finding that average #CPU power use only increased by about ~5% on average and only an 11% increase to the default/peak TDP. 5th Gen AMD EPYC processors continue to impress not only for raw performance over prior generation EPYC processors but also for impactful power efficiency improvements out of #Zen5
https://www.phoronix.com/review/amd-epyc-9655

Nice benchmark for AI models, Simple Bench.
https://simple-bench.com/
#AI #benchmarks #simplebench #llm

#Intel #Xeon 6980P "#GraniteRapids" #Linux #Benchmarks Review
The Xeon 6980P #CPU does manage to land well in front of the AMD EPYC 9684X flagship Genoa-X SKU overall along with the EPYC 9754 Bergamo processors. Generationally around 1.38x the performance with Granite Rapids compared to the prior Emerald Rapids with twice as many cores. AMD is about to launch the 5th Gen EPYC "Turin" it will be very interesting to see how those compete with #XEON6!
https://www.phoronix.com/review/intel-xeon-6980p-performance

#AWS #Graviton4 vs. #AmpereOne 192-Core #Benchmarks For Leading #AArch64 #Server Performance
Particularly with #HPC-type workloads and other scientific computing scenarios the Graviton4 with its #NeoverseV2 cores were delivering significant leads over the custom AmpereOne cores. Its worth reiterating AmpereOne was engineered as a 2022~2023 product but it's taken now until H2'2024 for its production to really ramp up to point of seeing Oracle Cloud A2 general availability
https://www.phoronix.com/review/ampereone-aws-graviton4

First #AI #Benchmarks Pitting #AMD Against #Nvidia
Results are good in that they show #MI300X is absolutely competitive with H100 #GPU on one set of AI inference benchmarks, and based on our estimates of GPU and total system costs can be competitive with Nvidia’s H100 and #H200. But, tests only done for #Llama2 #LLM model from Meta with 70 billion parameters.
A lot will depend, on how AMD prices #MI325 later this year and how many AMD can get its partners to manufacture.
https://www.nextplatform.com/2024/09/03/the-first-ai-benchmarks-pitting-amd-against-nvidia/

#Benchmarks show even an old #Nvidia #RTX3090 is enough to serve #LLM to thousands
For 100 concurrent users, the card delivered 12.88 tokens per second—just slightly faster than average human reading speed
https://www.theregister.com/2024/08/23/3090_ai_benchmark/ #AI #ML

#Amazon's #Graviton Has Evolved Into A Formidable #CPU Contender: #Graviton1 To #Graviton4 #Benchmarks Review
Going from the Graviton3 to Graviton4 metal instance was 1.55x the performance overall, still greater than the 1.5x core difference and would be an even larger difference if isolating to only the fully multi-core benchmarks.
Going from the original #AWS Graviton1 a1.metal instance to Graviton4 metal was an incredible 10.4x!
https://www.phoronix.com/review/aws-graviton-graviton4

This is a longform article from Tom's Hardware, but if you want to know what AMD is giving us this year it's worth the look.

#AMD deep-dives #Zen5 architecture — #Ryzen9000 and #AI300 #benchmarks, RDNA 3.5 #GPU, #XDNA2, and more https://www.tomshardware.com/pc-components/cpus/amd-deep-dives-zen-5-ryzen-9000-and-strix-point-cpu-rdna-35-gpu-and-xdna-2-architectures

#Nvidia #GraceSuperchip loses to #Intel #SapphireRapids in #HPC performance #benchmarks, promises greater efficiency
Barcelona #Supercomputing and #SUNY published benchmarks showing Grace Superchip, couldn't quite match TWO 48-core Sapphire Rapids #CPU. Despite not having earth-shattering performance, Grace promises to be competitive datacenter and HPC processor thanks to its efficiency. Built on Arm architecture. A single Grace CPU comes with 72c and 480GB of LPDDR5X RAM
https://www.tomshardware.com/pc-components/cpus/nvidia-grace-superchip-loses-to-intel-sapphire-rapids-in-hpc-performance-benchmarks-but-promises-greater-efficiency

#framework13 #laptop #benchmarks on #fedora 39 https://openbenchmarking.org/result/2311027-NE-Y9025498961

Especially impressed at how it performed in several of the encoding benchmarks and reached near server CPU levels of performance on several of the test results including #kernel #compilation as well as beating out it's reference chip. Likely due to the #amd pstate epp guided improvements in the 6.5 kernel in fc39

Qualcomm’s Snapdragon X Elite will come in 23W and 80W variants, will support Windows and Linux PCs #Benchmarks, #Chips, #Linux, #Qualcomm, #SnapdragonXElite

https://liliputing.com/qualcomms-snapdragon-x-elite-will-come-in-23w-and-80w-variants-will-support-windows-and-linux-pcs/