Epyc

HPE has introduced the first four systems in its next generation of ProLiant servers.

ProLiant Gen 13 starts with the 6th generation AMD EPYC processors and covers a pretty wide range of infrastructure.

At one end is the DL525, a 1U single socket server with as many as 256 CPU cores.

At the other is the DL585A, which supports two EPYC processors and as many as eight double-wide PCIe GPUs for AI inference, model fine-tuning, and other accelerated workflows.

HPE is also introducing two dense, open-compute project-based systems aimed at AI and high-performance computing.

But the generational change isn't only about feeding more AI into the data center.

Like, that's exactly what NVIDIA does.

They spent the last decade building the software, networking libraries, models, et cetera, around the GPU, so the developers end up choosing the entire ecosystem rather than comparing, "Oh, does this benchmark do better than this benchmark in the accelerator department?" So, like, I think that this World Labs really does give AMD a chance to be- begin creating the same kind of feedb- feedback loop where model researchers are building directly on Instinct and EPYC, and they learn where the software and hardware limitations are, and then that information goes straight back into the next generation of silicon that Fei-Fei Li is gonna have her hand in.

I think it's a genius acquisition.

Like you mentioned, eight pr- it's, $8 billion is a big number, but like, come on.

Hardware innovation is also moving at full throttle, with fresh updates aimed at improving the backbone of enterprise and cloud computing.

The Linux kernel has introduced critical improvements for AMD EPYC server systems.

promising better performance and optimized GPU management for intensive applications and large data operations.

These kernel upgrades underline the importance of collaborating closely between hardware and open-source communities to get the most out of today's most powerful systems.

And it's funny, you do the double click on how Muse actually works.

they're actually giving a VM per person, and it happens to be sitting on EPYC, AMD EPYC processors, which I thought was pretty cool.

And Alexander Wang, I keep forgetting his name, he actually came from, he was famous for an open-claw fork, right? And so essentially and I pinged a couple very more technical people than me on this on this thesis says this is essentially open claw in the cloud.

Okay.

And then the caveat that validates his specific worry about the phone being unusually idle.

Those benchmarks ran on an AMD EPYC server CPU pinned to 32 cores.

The paper itself warns that on weaker consumer hardware, the margin shrinks.

Under an illustrative assumption of roughly two times lower throughput...