#petaflop

[pulsaris]

NVIDIA - CES 2025

A NVIDIA esteve presente em força na CES e anunciou várias novidades no campo da IA e do processamento gráfico.

O destaque vai para as novas placas-gráficas RTX da série 50 com especial atenção para a placa GeForce RTX 5090 Founders Edition.

Outro destaque vai para o supercomputador em formato desktop com a designação de “Project DIGITS” e que tem um petaflop de processamento de inteligência artificial.

Saiba tudo em: https://blogs.nvidia.com/blog/ces-2025-jensen-huang/

_______ Direitos de imagem: © NVIDIA Corporation (via https://blogs.nvidia.com/)

[darkmaga-returns]

What took a space larger than an 18-wheel truck a few years ago to play with even a thousand cubits, Microsoft discovered a way to fit 1,000,000 or more cubits on a palm-sized chip. This breakthrough came thanks to the discovery of an “entirely new state of matter.” By comparison, the largest traditional supercomputer in the world that performs 1,742 petaFLOPS is easily outpaced by a 50-qubit quantum computer, but each qubit added exponentially increases performance; thus, a palm-sized chip could outrun 20,000 supercomputers.

Essentially, Microsoft’s approach will easily blow the lid off every current limitation. If millions of these chips reach the market, imagine what would come of AI? It will control the world, not the other way around. ⁃ Patrick Wood, Editor.

[argumate]

an Nvidia H100 can deliver a petaflop or more depending on the workload and the number of cores, and DeepSeek apparently has 50,000 (!) of these H100s, so that's approaching 5 x 10^19 flops compared with my old 486 DX/2 that was capable of 5 x 10^7 flops, so that's a trillion times more computation per second; guess it really would have been challenging to implement this thirty years ago.

[therealgchu]

WIP Wednesday - To the Shore

hey, check it out, actually doing a main fic thing! mainly cos my brane is still stuck in the middle of chapter 4 of Seven Days, and i'm getting real tired of its shit. so tired of it that, apparently, i'm writing back on my main fic.

tagging the coemancer crew, and anyone else if you want to share a WIP.

anyway, something fuzzy and warm for today's WIP wednesday. i've skipped a bit ahead, but not too far. still trying to plot out some major plot points and how that's all going to happen.

if you want to read To the Shore from the beginning, check it out here.

there's also the backstories, Anamnesis here. i'll be putting some some spicy backstories soon, but it'll be in a different fic since i don't want to muddy the tags for the OG backstories.

and, Seven Days, if my brane ever gets chapter 4 finished.

on with the sneak peek!

“I want a bunk bed,” Cora said.

Sam smiled, “I think we can do that, right Hwa?” he asked as he turned to her.

Hwa smiled, “Of course. Why a bunk bed?”

“So, I can put my plushies on top!” Cora answered.

“Makes sense,” Hwa nodded, “yeah, no problem.”

“And, I want a computer, of course. Have you seen the new processors that came out recently? Some of them can get up to a couple petaflops!” Cora gushed. “I’ve got the specs written down,” she announced as she rummaged through her backpack

Sam looked in askance at his daughter, “That might be pretty expensive, gumdrop,” he said, attempting to temper her enthusiasm.

Hwa grinned at the girl, the first real smile she’d had in weeks. “Don’t worry about computers. I’ve already got some specc’d out for the both of us.”

“Hey, don’t I get to have a say?” Sam interrupted. 

Hwa moved over to stand by Cora’s side and the two looked at each other, then looked back to Sam. “Umm…Dad, would you even know what we’re talking about?” Cora asked.

He grunted and shook his head. “Fine, you two do whatever. No illegal stuff, though, got it?”

“Yessss,” both answered sarcastically.

“Good,” he said as Cora raced down to her berth in the lower deck. Sam crossed the way and put his arm around Hwa’s waist. “It’s good to see you smiling again, darlin,” he said as he kissed her cheek.

She leaned against his chest, “I’m sorry. I know it’s been hard on the both of you,” she said.

“There’s nothing to apologize for,” he said as he gave her waist a squeeze. “I’m glad we’re going back to Ternion. I kinda missed working on the homestead.”

The term made her smile, “Homestead, huh? Your Akila is showing,” she ribbed him.

“Well now, miss, I reckon that as an Akila man, born and raised, we should rightly call our home a homestead. It only sounds right and proper,” Sam said, laying it on thick.

Hwa laughed as Sam cocked his hat to her. They both heard Cora moan from the lower deck, “Daaddd! That’s terrible!” she shouted. He grinned even wider.

“Really, though, we should come up with a name,” Hwa said.

A herd of small horses was heard on the lower deck as Cora made her way back and up the ladder. “I want to name it!” she shouted.

Hwa and Sam looked at each other, both seeing agreement on the other’s face. “Sure, gumdrop. What do you want to name the homestead?”

“Pemberley!” Cora said excitedly.

Sam looked quizzically while Hwa chuckled, “It’s Mr. Darcy’s estate,” Hwa explained.

“I remember that now, thought it sounded familiar,” Sam said, scratching his head.

“You don’t like it,” Cora said, crestfallen.

“It’s not that I don’t like it, it’s that, well, I don’t think we’re a ‘Pemberley Estate’ sort of family,” Sam explained.

“We could be if we wanted to be,” Cora wheedled, pouting a bit.

“Cora, I don’t think I could be an ‘Estate’ person if I tried,” he said, trying not to be too harsh.

“How about a compromise,” Hwa interjected, “Pemberley Homestead,” she suggested.

[machine-saint]

Strangers keep asking me how many petaFLOPS my robot gf has. The answer is ONE. I'M the only flop she pets. ME.

.

[yakdaimeedcsgo]

5 Supercomputer ที่แรงที่สุดในโลก

ปัจจุบันคอมพิวเตอร์ได้กลายมาเป็นส่วนสำคัญของชีวิตประจำวันของพวกเรา และ Supercomputer เป็นหนึ่งในอุปกรณ์ที่ทรงพลังและมีประสิทธิภาพ

Supercomputer เป็นอุปกรณ์ที่มีอัตราคำนวณได้เร็วและแม่นยำกว่าคอมพิวเตอร์ทั่วไป โดยมักถูกใช้ในงานวิจัยทางวิทยาศาสตร์ การพยากรอากาศ การวิเคราะห์ และในบทความนี้จะกล่าวถึง 5 อันดับ Supercomputer ที่ทรงพลังที่สุด

โดยก่อนจะเริ่มเราต้องมาทำความรู้จักกับ FLOPS ก่อน FLOPS หรือ Floating-Point Operation Per Second เป็นหน่วยวัดสมรรถนะการทำงานคอมพิวเตอร์ FLOPS จะนับจำนวนชุดคำสั่งในการประมวลผลเลขทศนิยม

1.Summit (Tennessee, USA.)

พัฒนาโดย IBM และ NVDIA มีพลังคำนวณ 200 petaflops ใช้ในงาน ฟิสิกส์และสุขภาพ

2.Sierra (California, USA.)

พัฒนาโดย IBM และ NVDIA มี���ลังคำนวณ 125 petaflops ใช้ในงาน อาวุธนิวเคลียร์และการเปลี่ย��แปลงของสภาพอากาศ

3.Sunway Taihulight (Wuxi, China.)

พัฒนาโดย National Research Center of Parallel Computer Engineering and Technology มีพลังคำนวณ 93 petaflops ใช้ในงาน วิจัยวิทยาศาสตร์โลกและการศึกษาการเปลี่ยนแปลงของสภาพอากาศ

4.Tuanhe-2A (Guangzhou, China.)

พัฒนาโดย National University of Defense Technology มีพลังคำนวณ 61 petaflops ใช้ในงานวิจัยวัสดุศาสตร์และวิศวกรรม

5.Frontera (Texas, USA.)

พัฒนาโดย Dell มีพลังคำนวณ 23.5 petaflops ใช้ในงานวิจัย พาติเคิลฟิสิฟส์และวิทยาศาสตร์โลก

ที่มา : Top 10 Most Powerful Supercomputers in the World You Should Know (analyticsinsight.net)

[mynameisanacronymofitself]

A Microplastic LARPing petaflop.

I'm stealing this from Twitter

Here's the link

I am a high-definition gateway drug body double!

[groovy-computers]

Huawei's shaking up the AI landscape! 🌟 Discover how their Ascend 910C chips are narrowing the technology gap with NVIDIA and boosting China's self-reliance. **Innovation Unleashed** Huawei has launched the Ascend 910C, a next-gen AI chip that powers the CloudMatrix 384 AI cluster. This setup delivers 300 PetaFLOPS of computation power, significantly outperforming NVIDIA's GB200 NVL72. However, it comes with high energy consumption—a challenge China seems ready to meet. **Shifting Tides in AI** This leap in technology represents more than just high performance; it signifies a strategic move for China. By minimizing reliance on Western tech, Huawei boosts domestic capabilities and market independence. **The Power Play: Competition Heats Up** With these advancements, Huawei positions itself as a formidable competitor in the AI market. How will this affect global tech dynamics? **Curious to Know More?** Do you think Huawei's strides in AI will redefine global tech leadership? #Huawei #Ascend910C #AIRevolution #TechInnovation #NVIDIA #CloudComputing #AICluster #GlobalTech #Innovation🌐 #TechnologyUpdates

[ericvanderburg]

Dell delivers monster 20-petaFLOPS desktop built on Nvidia's GB300 Superchip

http://securitytc.com/TJc4pW

[hexa-blogs]

NVIDIA Project DIGITS: Revolutionizing Personal AI Supercomputing

As of January 2025, NVIDIA has introduced Project DIGITS, a groundbreaking personal AI supercomputer designed to democratize access to advanced AI computing. This compact system, comparable in size to a Mac Mini, empowers AI researchers, data scientists, students, and hobbyists to develop, prototype, and fine-tune AI models directly from their desktops without relying on extensive cloud infrastructure.

Key Features of Project DIGITS

Powerful Hardware: At the core of Project DIGITS is the NVIDIA GB10 Grace Blackwell Superchip, which integrates a 20-core Grace CPU and a Blackwell GPU. This architecture delivers up to 1 petaflop of AI computing performance at FP4 precision, allowing users to handle large-scale AI models with up to 200 billion parameters. When two units are linked together, they can support models with up to 405 billion parameters.

Unified Memory and Storage: Each unit comes equipped with 128 GB of unified memory and up to 4 TB of NVMe flash storage. This configuration ensures smooth handling of complex computations and large datasets, making it suitable for demanding tasks such as real-time data analysis and high-resolution video processing.

Seamless Integration with Software Ecosystem: Project DIGITS runs on a Linux-based DGX OS and includes the full NVIDIA AI Enterprise software stack. Users have access to libraries, frameworks, and orchestration tools that facilitate integration with cloud or data center infrastructures. The system supports popular frameworks like PyTorch and Jupyter notebooks, making it accessible for developers familiar with existing tools.

Accessibility and Affordability: Starting at $3,000, Project DIGITS aims to place powerful AI computing capabilities within reach of a broader audience. This pricing strategy is designed to empower individual developers and researchers who may have previously faced barriers due to hardware limitations or high costs.

Applications and Use Cases

Project DIGITS is particularly well-suited for various applications:

AI Model Prototyping and Fine-Tuning: Users can develop and refine large language models (LLMs) locally before deploying them in production environments.

Data Analysis and Simulations: The system can efficiently handle real-time data analysis, making it ideal for industries that rely on big data.

AI Content Generation: It is capable of generating AI-driven content for various media applications.

Impact on the AI Landscape

NVIDIA's Project DIGITS represents a significant step towards making high-performance AI computing accessible to a wider audience. By enabling local model development without the need for extensive cloud resources, it empowers users to engage more deeply with AI technologies. Jensen Huang, NVIDIA's CEO, emphasized that placing an AI supercomputer on the desks of developers will allow them to shape the future of AI across industries.

Conclusion

As we move further into 2025, NVIDIA's Project DIGITS stands out as a transformative tool in the realm of personal computing and artificial intelligence. By providing robust performance in a compact form factor at an accessible price point, it opens new avenues for innovation in AI development. Whether for academic research or independent projects, Project DIGITS is set to empower the next generation of AI practitioners.For more information about NVIDIA's latest innovations and how they can enhance your work in AI development, visit NVIDIA's official page or explore related articles on platforms like Hacker Noon.

[theenterprisemac]

1 Petaflop of FP4 Performance

Is it just me or does it feel a bit cringe to say 1 petaflop of FP4 performance? Something about only having 16 possible numbers and then advertising that you can compute those 16 possible numbers super fast feels a bit less impressive than saying you are delivering 1 petaflop of FP32 or FP64 performance.

[shivamthakrejr]

AI Data Center Builder Nscale Secures $155M Investment

Nscale Ltd., a startup based in London that creates data centers designed for artificial intelligence tasks, has raised $155 million to expand its infrastructure.

The Series A funding round was announced today. Sandton Capital Partners led the investment, with contributions from Kestrel 0x1, Blue Sky Capital Managers, and Florence Capital. The funding announcement comes just a few weeks after one of Nscale’s AI clusters was listed in the Top500 as one of the world’s most powerful supercomputers.

The Svartisen Cluster took the 156th spot with a maximum performance of 12.38 petaflops and 66,528 cores. Nscale built the system using servers that each have six chips from Advanced Micro Devices Inc.: two central processing units and four MI250X machine learning accelerators. The MI250X has two graphics cards made with a six-nanometer process, plus 128 gigabytes of memory to store data for AI models.

The servers are connected through an Ethernet network that Nscale created using chips from Broadcom Inc. The network uses a technology called RoCE, which allows data to move directly between two machines without going through their CPUs, making the process faster. RoCE also automatically handles tasks like finding overloaded network links and sending data to other connections to avoid delays.

On the software side, Nscale’s hardware runs on a custom-built platform that manages the entire infrastructure. It combines Kubernetes with Slurm, a well-known open-source tool for managing data center systems. Both Kubernetes and Slurm automatically decide which tasks should run on which server in a cluster. However, they are different in a few ways. Kubernetes has a self-healing feature that lets it fix certain problems on its own. Slurm, on the other hand, uses a network technology called MPI, which moves data between different parts of an AI task very efficiently.

Nscale built the Svartisen Cluster in Glomfjord, a small village in Norway, which is located inside the Arctic Circle. The data center (shown in the picture) gets its power from a nearby hydroelectric dam and is directly connected to the internet through a fiber-optic cable. The cable has double redundancy, meaning it can keep working even if several key parts fail. 

The company makes its infrastructure available to customers in multiple ways. It offers AI training clusters and an inference service that automatically adjusts hardware resources depending on the workload. There are also bare-metal infrastructure options, which let users customize the software that runs their systems in more detail.

Customers can either download AI models from Nscale's algorithm library or upload their own. The company says it provides a ready-made compiler toolkit that helps convert user workloads into a format that runs smoothly on its servers. For users wanting to create their own custom AI solutions, Nscale provides flexible, high-performance infrastructure that acts as a builder ai platform, helping them optimize and deploy personalized models at scale.

Right now, Nscale is building data centers that together use 300 megawatts of power. That’s 10 times more electricity than the company’s Glomfjord facility uses. Using the Series A funding round announced today, Nscale will grow its pipeline by 1,000 megawatts. “The biggest challenge to scaling the market is the huge amount of continuous electricity needed to power these large GPU superclusters,” said Nscale CEO Joshua Payne. Read this link also : https://sifted.eu/articles/tech-events-2025

“Nscale has a 1.3GW pipeline of sites in our portfolio, which lets us design everything from scratch – the data center, the supercluster, and the cloud environment – all the way through for our customers.” The company will build new data centers in North America and Europe. The company plans to build 120 megawatts of data center capacity next year. The new infrastructure will support Nscale’s upcoming public cloud service, which will focus on training and inference tasks, and is expected to launch in the first quarter of 2025.

[dailyreportonline]

Nvidia Unveils Project Digits Personal AI Supercomputer With GB10 Chipset, Can Run Large AI Models at CES 2025 | Daily Reports Online

Nvidia unveiled a personal supercomputer equipped with its in-house Grace Blackwell Superchip on Monday at the Consumer Electronics Show (CES) 2025. Dubbed Project Digits, the device can develop, run inference, and deploy large artificial intelligence (AI) models. The personal supercomputer is equipped with the NVIDIA GB10 Grace Blackwell Superchip chipset and delivers up to one petaflop (1,000…

[tecnobinaria]

NVIDIA presenta el nuevo superchip GB10, el proyecto DIGITS, la supercomputadora de inteligencia artificial más pequeña del mundo capaz de ejecutar modelos con 200B de parámetros

NVIDIA presenta el nuevo superchip GB10, el proyecto DIGITS, la supercomputadora de inteligencia artificial más pequeña del mundo capaz de ejecutar modelos con 200B de parámetros

NVIDIA presentó NVIDIA ® Project DIGITS, una supercomputadora personal de IA que brinda a investigadores de IA, científicos de datos y estudiantes de todo el mundo acceso al poder de la plataforma NVIDIA Grace Blackwell. El proyecto DIGITS presenta el nuevo superchip NVIDIA GB10 Grace Blackwell, que ofrece un petaflop de rendimiento informático de IA para crear prototipos, ajustar y ejecutar…