JAX 是一个强大的机器学习库，结合了类似 NumPy 的 API、自动微分功能以及 Google 的 XLA 编译器。它在模型并行化方面提供了优秀的 API，因此非常适合像 LLaMA 3.1 405B 这样的超大模型训练。

OpenCV也为GPU加速提供了支持，通过引入OpenCL、CUDA等加速后端，极大地提升图像视频处理、线性代数计算等复杂任务的执行速度，满足更大规模数据 ...

机器之心报道机器之心编辑部随着 AI 模型的参数量越来越大，对算力的需求也水涨船高。比如最近，Llama-3.1 登上了最强开源大模型的宝座，但超大杯 405B 版本的内存就高达 900 多 ...

NVIDIA's next-generation ultra-enthusiast flagship GeForce RTX 5090 spotted in factory trial video, being tested in Indonesia ...

英伟达几乎垄断了数据中心AI芯片市场，占有超过90%的份额。作为市场的第二号选手，AMD一直想要增加自己的市场份额。这次的Instinct MI325X发布，是AMD对英伟达领导地位的一次重大挑战。

A leaked footage from a validation factory in Indonesia gives us our first look at NVIDIA's flagship RTX 5090.

Nvidia has a set of unique attributes such as Domain Experience that sets it apart from most companies. See why I think NVDA ...

实际上，葛卫东在AI领域的投资布局中始终对海光信息保持高度关注，前者自公司上市之初便持有其原始股份。然而今年以来，混沌投资频繁实施减持，从2024年初至2月22日，该投资公司减持海光信息约799万股，而后在3月31日前又减持约409万股。截至9月30 ...

We don’t always JTAG, but when we do, we use a Black Magic Probe. It’s a completely open ARM-chip debugging powerhouse. If you program the small ARM chips and you don’t have a BMP ...

As a debugger, GDB is a veritable Swiss Army knife. And just like exploring all of the non-obvious uses of a those knives, your initial response to the scope of GDB’s feature set is likely to be ...

But that’s ok. Mistakes happen when you’re writing code. What's important is that you know how to debug your code. Debugging is going back over the algorithms and code you have written ...