VFXnow的动态

Boost GPU performance and cut energy costs with QCT’s Qoolrack, designed for NVIDIA GB200 NVL72 systems. Download the white paper today: https://ow.ly/M7nO50Uq8V3

Fine-tune your own Mochi video model to tell your story your way. Here’s what you need to get started: ? 1 NVIDIA GPU ? A few videos ? Infinite creativity ?? Step-by-step guide: https://lnkd.in/eGDWXyrd

Moore's law is still alive! I ran a quick benchmark on #Speos GPU between my former laptop and the replacement one. The new #GPU is of the latest #NVIDIA Ada Generation, direct sequel to the Ampere generation from previous laptop ?? Did you notice in Speos #2024R2, the Live Preview interface has a new Status Bar where you can read active GPU model name, run time and simulation performance? That's an easy way to track your compute power ?? Performance simply DOUBLED while keeping the same weight, form factor and power consumption ?? Impressive, but not really surprising since I observed this doubling trend on Speos GPU across the past 5 generations! ?? Good job NVIDIA for granting Ansys Optics the power to deliver groundbreaking changes in #simulation workflows! ??

Nvidia tiny super computer. https://lnkd.in/gq37GgnM NVIDIA recently launched the Jetson Orin Nano Super Developer Kit, a compact and affordable generative AI supercomputer. Priced at $249, this device offers significant performance improvements over its predecessor, making it accessible to hobbyists, students, and developers. Key Features: High Performance: Powered by NVIDIA's Ampere architecture GPU with tensor cores and a 6-core Arm CPU, the device achieves 67 TOPS (trillion operations per second) with INT8 precision, a 70% increase over previous models. Memory

Accelerate your day with this article:

Major milestone for Polars: you can now run your dataframe processing on GPUs. For vectorized operations, this means massive speedups. Polars was already blazingly fast (owing to its Rust backend and native integration with Apache arrow), but this integration with NVIDIA RAPIDS takes it to another level. Curious? I will be giving a talk about Polars, and will touch upon RAPIDS on October 17 at PyData Valais: https://lnkd.in/eXcy34AV #polars #rapids #cudf #gpgpu #pydatavalais #pydata

Never have I met a customer who wants their workloads to run slower! ?? Today, Polars released a new GPU engine powered by RAPIDS cuDF in open beta. ?? The fastest performance for Polars is now on NVIDIA GPUs, up to 13x speed up over CPUs! ?? Here's a quick video (https://lnkd.in/gT5SXGCK) showing it's immense value for Data Engineers, Data Scientists and ML/AI Practitioners alike. Also, try the Intro Notebook here:?https://lnkd.in/g2k66edc

Nvidia black family gpu is here 5070 with 4090 performance ??

Nice video from Anastasiia about Nvidia Blackwell GPU packaging challenges. Packaging is the term we use to describe the pieces we add around a chip so it can be integrated with the printed circuit board. Over the years, my research team has dabbled in package design and even open sourced some of our designs (https://bjump.org). You can tweak the package or just email a company that we use to manufacture them and they can send you a few thousand or a few million parts. The funny thing about packaging from a EE/CS perspective is that the performance is fairly easily modeled using the EE analyses we have built up over the last few hundred years. Even the thermal properties can be modeled using EE equations. But packaging has a key mechanical question that is out of the EE wheelhouse to analyze. Simply stated, will the different components expand/compress differently under heat cycling and cause connections to be broken? When you look at the Blackwell package, the first question that pops into your head is, wow, that looks like a tricky mechanical problem, must have been hard to solve! So it’s not surprising that this was a problem, but what is surprising is that given that it is an obvious failure mode, that they did not sufficiently derisk it… (Kind of like when you go to the aerospace museum and they show videos of people playing tennis while standing on the wings of a biplane that is flying though the air, and then surprise! you find out some people met an untimely end doing it.) Of course with all this said, Nvidia and TSMC have a world-class engineering team, and I am sure they will overcome this hurdle.

SuperNODE, the world’s only 32-GPU server, flawlessly executed one of the largest CFD simulations ever — the Concorde landing at 40 BILLION cells — in a mere 33 hours. The secret? Massive GPU memory, making the complex fast and easy. See how! https://bit.ly/3BY1RAY