Hacker Dec 11

#Nvidia Says It’s Not Abandoning 64-bit Computing https://www.hpcwire.com/2025/12/09/nvidia-says-its-not-abandoning-64-bit-computing/

#hpc #fp64

Nvidia Says It's Not Abandoning 64-Bit Computing - HPCwire

Nvidia has taken some lumps from some people in the supercomputing community who say it’s neglecting the 64-bit performance that traditional modeling and simulation workloads demand in favor of boosting performance of lower precision computing that’s useful for AI. However, an Nvidia executive tells HPCwire that it’s not abandoning 64-bit computing, that new emulation libraries […]

HPCwire
Benjamin Carr, Ph.D. 👨🏻‍💻🧬Nov 19
#ScientificComputing is about to get a massive injection of #AI
#Nvidia's #IanBuck on the importance of #FP64 to power research, in a world that's hot for inferencing
Enabling the AI-#HPC merger: on Monday, the company unveiled Apollo – a new family of open models designed to accelerate industrial and computation engineering. Nvidia has integrated Apollo into industrial design software suites from Cadence, Synopsys, and Siemens.
https://www.theregister.com/2025/11/18/future_of_scientific_computing/ #SC25
Scientific computing is about to get a massive injection of AI

Interview: Nvidia's Ian Buck on the importance of FP64 to power research, in a world that's hot for inferencing

The Register
HGPU groupOct 5

Performance and Numerical Aspects of Decompositional Factorizations with FP64 Floating-Point Emulation in INT8

#NVIDIA #Int8 #FP64 #Factorization

https://hgpu.org/?p=30278

Performance and Numerical Aspects of Decompositional Factorizations with FP64 Floating-Point Emulation in INT8

Mixing precisions for performance has been an ongoing trend as the modern hardware accelerators started including new, and mostly lower-precision, data formats. The advantage of using them is a gre…

hgpu.org
HGPU groupAug 10

DGEMM without FP64 Arithmetic – using FP64 Emulation and FP8 Tensor Cores with Ozaki Scheme

#CUDA #Performance #DGEMM #FP64 #Package

https://hgpu.org/?p=30081

DGEMM without FP64 Arithmetic – using FP64 Emulation and FP8 Tensor Cores with Ozaki Scheme

Since AI computations require low-precision matrix multiplications, processors with enhanced performance for these operations are increasing along with the growing demand for AI computations. Howev…

hgpu.org
Show threadDr. Moritz LehmannJun 13, 2025

Interesting take from an #Nvidia engineer I met at #ISC25: "Do you need IEEE-754 compliant #FP64, or do you need digits? Digits we can get you through FP64 emulation."

Not sure what to make of that. Things were an absolute mess before the IEEE-754 standard, and I wouldn't want to ever go back to that. No standards means you cannot at all port software between hardware architectures even from within the same vendor. Having to re-architect software for each and every new chip is not gonna happen.

Dr. Moritz LehmannJun 10, 2025

With all the hot discussions about #FP64 right now, what do you say to Nvidia almost entirely axing FP64 on #Blackwell Ultra B300?

B200 180GB FP64 performance: 37.2 TFlops/s
B300 288GB FP64 performance: 1.2 TFlops/s ☠️

They're going from the usual datacenter #GPU FP64:FP32 ratio of 1:2 down to cheap 1:64, like on all their gaming/workstation GPUs.
Personally I think this is fantastic - it opens the doors wide for competitors to step into the #HPC market.

https://www.nvidia.com/en-us/data-center/gb300-nvl72/

NVIDIA GB300 NVL72

Built For The Age of AI Reasoning.

NVIDIA
Andrew Jones (hpcnotes)Jun 10, 2025

Some great debate already in response to my question on #FP64.

But are we just defending FP64 because it is what we have now and what we know?

Would we defend FP32 if that is all we had?

If we had pervasive FP128 would we be arguing that a drop to FP64 is not good enough?

#ISC25 #HPC

Andrew Jones (hpcnotes)Jun 9, 2025

Provocative question to debate at #ISC25 - in talks, meetings, booths, over beers:

What is so special about #FP64 that science holds it as a gold standard?

If #HPC had never got as far as FP64, eg had got stuck at FP32, surely we would still have been able to do science?

Benjamin Carr, Ph.D. 👨🏻‍💻🧬May 21, 2025

#JackDongarra Makes a Stand for Traditional #HPC: "US still doesn’t have a clear, long-term plan for what comes next.... U.S. risks falling behind."

Challenges to high-performance computing threaten #US #innovation

The #AI boom has led chip makers to focus on #FP16 and #FP8, not the #FP64 used by scientific research. If chip companies stop making the parts that #scientists need, then it could become harder to do important research.
https://theconversation.com/challenges-to-high-performance-computing-threaten-us-innovation-255188

Challenges to high-performance computing threaten US innovation

Today’s supercomputers are enormously powerful, but the work they do − running AI and tackling difficult science − is pushing them to their limits. Building bigger supercomputers won’t be easy.

The Conversation
Show threadChristos ArgyropoulosMay 16, 2025

The extreme Haswell processors by Intel packed some nice #fp64 juice. This is post by Puget systems from 2014 about the i5960x (the processor used in the build)
https://www.pugetsystems.com/labs/hpc/linpack-performance-haswell-e-core-i7-5960x-and-5930k-594/?srsltid=AfmBOopy1vVR0e0qroH-WfiuNg9M_yBOBoQf3r8hC7Pw7rbfV05a52Qb
It achieved nearly 320 GFLOPS.

This makes the system shown here a 1.6Tflop system among the Kepler , the Phi and the i7 achieved by a combined power consumption of ~700W)

If you are after lots of TFLOPS you can drop the cash for one of these babies
https://www.pugetsystems.com/labs/hpc/amd-zen4-threadripper-pro-vs-intel-xeon-w9-for-science-and-engineering/
or source cheap lga2011 & gpus from ebay

Linpack performance Haswell E (Core i7 5960X and 5930K)

The new Intel desktop Core i7 processors are out, Haswell E! We look at how the Core i7 5960X and 5930K stack up with some other processors for numerical computing with the Intel optimized MKL Linpack benchmark.

Puget Systems