AMD's EPYC 9004 "Genoa" Reference Board Runs The Open-Source OpenBMC

[Developer12]

Can't Google it? There's even a Phoronix article about it. Yes Project X sucks at communicating and would be more successful if they had a blog or something more about what they're doing. They never know who might email them with information that can help them achieve a "new corebooted Ryzen CPU" for example.

"Project X" isn't terribly descriptive. With "ryzen" and "coreboot" it becomes a little better. Thanks.

I recall this project a while back. Pitty that it's all just basically a shim, being as it doesn't start until after the PSP has done all the ram init and other platform startup.

I believe Oxide actually undertook the same work, except they did it in rust.

[coder]

New market to develop a chip for. You'd need to integrate a small CPU to run the BMC OS, plus a display controller and NIC and a software stack to run on it.

Yeah, but I can dream, at least.

Maybe ASpeed will license some Imagination IP, to use in a future BMC, though I doubt it.

[agd5f]

Adding an entire dGPU is a waste of money for a server that mostly runs headless, though. Not only that, it chews up a PCIe slot.

When they're already putting a separate BMC chip on the board, why not go with a pure AMD solution? It's a missed opportunity, I say.

New market to develop a chip for. You'd need to integrate a small CPU to run the BMC OS, plus a display controller and NIC and a software stack to run on it. It also needs to be cheap. Keeping costs low limits the functionality of the integrated display hardware. As the display capabilities increase, you start to need more 3D capabilities in order to provide a better user experience on a large display. That adds additional power and software requirements. There is apparently not a big market for BMC chips that support high resolution displays.

I didn't say that. Does AMD have no other chips on these boards, aside from their CPUs?

On the server side, not really. The platform chipsets are really only a client thing.

[deusexmachina]

What's Project X? The only think I know of like that are the POWER9 machines built by Raptor.

Can't Google it? There's even a Phoronix article about it. Yes Project X sucks at communicating and would be more successful if they had a blog or something more about what they're doing. They never know who might email them with information that can help them achieve a "new corebooted Ryzen CPU" for example.

[Developer12]

Right, so I'm sending my prayers to Project X. A fully open source stack without any management engines that can transparently analyze my ram is a dream.

What's Project X? The only think I know of like that are the POWER9 machines built by Raptor.

[baryluk]

Proprietary BMCs are horrible. Horrible UI, horrible licensing, slow, horrible APIs.

OpenBMC is already better than most BMCs. It has nice REST API, easy LDAP integration, SOL over SSH, REDFISH API support. The only thing that is missing in to make it perfect is native Prometheus metrics for monitoring. (There is OpenBMC exporter that uses REDFISH and exports for Prometheus, but that has issues, needs extra software on other box, and is just not so good. For example data model is different, so you need to hand craft yaml files that translate names and labels for all metrics. And there were no update for 2 years. It was written by Intel appearently).

[deusexmachina]

Platform Security Processor. It's AMD's Management Engine equivalent.

In fact, the current "100% fully open source" init of ryzen platforms in coreboot is only due to the PSP starting up everything and doing ram init all before handing over to coreboot to load the OS. I don't think any of that process can or will ever be replaced by an open source alternative.

Right, so I'm sending my prayers to Project X. A fully open source stack without any management engines that can transparently analyze my ram is a dream.

[coder]

Buy a dGPU if you need more performance?

Adding an entire dGPU is a waste of money for a server that mostly runs headless, though. Not only that, it chews up a PCIe slot.

When they're already putting a separate BMC chip on the board, why not go with a pure AMD solution? It's a missed opportunity, I say.

I don' t think you'd want the BMC integrated into the die of the CPU, if something happens to the CPU,

I didn't say that. Does AMD have no other chips on these boards, aside from their CPUs?

[agd5f]

I was really hoping AMD did something like their desktop Ryzen 7000 and included a little 2 CU iGPU in a southbridge chip on the board.

Why? Oh, why??

ASpeed is so slow. And I know someone is going to pick on me for saying this, but does it even support 4k? I have a HDMI KVM switch that does.

Buy a dGPU if you need more performance? The fact that the BMC has a display connector is a nice bonus, but it provides a lot more functionality than just a display device. I don' t think you'd want the BMC integrated into the die of the CPU, if something happens to the CPU, you'd lose your out of band management interface.

[coder]

@Michael: I saw your comment on a certain other site that reviewed this CPU (and has since corrected their omission). You, sir, are a class act.

Same.