HN Theater @HNTheaterMonth

Jeff Geerling (HN:geerlingguy, who is participating in this comments section) has demonstrated using a NVMe drive with the RPi 4 Compute Module and its IO board, which has a PCIe slot.

https://youtu.be/HUamq0ey8_M?t=429

⬐ throwaway894345

Hah, just came across his blog while trying to debug an issue I'm having: https://www.jeffgeerling.com/blog/2018/fixing-503-service-un...

Jeff Gerrling goes over why they changed... https://www.youtube.com/watch?v=HUamq0ey8_M&ab_channel=JeffG...

⬐ rbanffy

Sadly, they can't be mounted perpendicular to a cluster board like the previous version.

As ARM is regressing in the Cloud market, as you can see with Graviton, we want to invest in the future. I don't think it will be particularly bad for 7 Pi 4s to match with a E5-2670 v2 in anyway. You can also listen to [this guy](https://youtu.be/HUamq0ey8_M?t=797) for a briefing.

From my perspective its 4*7 = 28 weak ARM cores vs 8 strong x86 cores. You can see that ARM actually had more cores, giving it an advantage plus to parallelized workload compared to x86.

Hell, maybe we can mix them in a bunch so that x86 runs powerful applications like GitLab, Prometheus and Postgres while ARM runs massively parallel workload that GPUs can't handle: Function as a Service (in AWS terms, Lambda; in CNCF's term, OpenFaaS), Linkerd handler (service mesh needs some kind of scheduling though), microservice replicas.

In the end CPU are all going to have a designated purposes, despite it should have had been "general purpose".

⬐ geerlingguy

You can boot over USB but not yet via NVMe. The same NVMe drive (Samsung 970 EVO) was 2x faster for random access when mounted as an NVMe drive vs. in a USB 3.0 adapter with a superspeed PCIe USB 3.0 adapter card.

⬐ nsky-world

The Compute Module 4 is basically a Raspberry Pi 4 model B, with all the ports cut off. Instead of the ports, you plug the Compute Module into another board with its special board-to-board connectors. But the Compute module has a few other tricks up its sleeve:

Faster eMMC: It has optional onboard eMMC storage, which is now much faster than any microSD card I've tested PCI Express: It drops the USB 3.0 interface for a PCI Express interface, meaning you can do some pretty cool things in lieu of having a couple USB 3.0 ports. WiFi and U.FL: It has an external antenna connector for it's wireless interface. What's that? Oh yes, there's now a version of the Compute Module with Bluetooth and WiFi! More Options: There are now thirty two different Compute Module flavors to choose from, whether you want onboard WiFi or not, whether you want eMMC storage, or whether you want 1, 2, 4, or even eight gigabytes of RAM!

⬐ 3np

The exposed PCIe lanes looks like the killer here. I guess you can now do 2.5GBe, NVMe drives, beefier GPUs, and other fun things?

⬐ 0-_-0

IIRC the PCIe interface is not compatible with GPUs.

⬐ geerlingguy

It may be, but nobody has proven it yet... the biggest issue is lack of ARM video drivers :(

I might pick up an older PCIe card and see how far I can get using it with the CM4.

⬐ 0-_-0

I think it's proven it doesn't work because the RPi CPU doesn't have enough BAR (Base Address Register) space. More details in this discussion:

https://news.ycombinator.com/item?id=20399652

⬐ the_only_law

I have a project where I was going to build a full x86 machine to serve as a domain specific server because the HW interface I need is only available in PCI amd PCIe cards. Wonder if I could hack the drivers to work on one of these.

⬐ butterisgood

Did you get the content for your post from here? https://www.jeffgeerling.com/blog/2020/raspberry-pi-compute-...