Intel Publishes PCIe Bandwidth Controller Linux Driver To Prevent Thermal Issues

[phoronix]

Phoronix: Intel Publishes PCIe Bandwidth Controller Linux Driver To Prevent Thermal Issues

Intel engineer Ilpo Järvinen posted a set of Linux kernel driver patches to introduce a new "bwctrl" PCI Express Bandwidth Controller driver and associated PCIe cooling driver to allow for limiting the PCIe link speed in the event of any system thermal issues...

https://www.phoronix.com/news/Intel-PCIe-Bandwidth-Thermal

[jeisom]

Is this specific to a current issue or a future issue?

[schmidtbag]

I feel that if there's a need to throttle PCIe bandwidth due to thermal issues, maybe we need to stop pushing later generations of PCIe so aggressively...
I don't recall PCIe ever causing heat problems.

[StefanBruens]

I feel that if there's a need to throttle PCIe bandwidth due to thermal issues, maybe we need to stop pushing later generations of PCIe so aggressively...
I don't recall PCIe ever causing heat problems.

Apparently you forgot motherboard chipsets suddenly requiring active cooling again when the first PCIe 4.0 capable chipsets became available.

And there is more than just desktop computers and laptops, e.g. 5G base stations need massive amounts of computing power and bandwidth, and are often deployed in harsh environments. You don't want these to shut down completely during summer, especially when these also handle emergency services.

[bezirg]

I feel that if there's a need to throttle PCIe bandwidth due to thermal issues, maybe we need to stop pushing later generations of PCIe so aggressively...
I don't recall PCIe ever causing heat problems.

We already see this happening with pcie5 nvme ssds and perhaps some optanes.

But i like the idea of this patch anyway.

[avis]

Is this specific to a current issue or a future issue?

Could be a problem for upcoming PCI-e 6.0.

I feel that if there's a need to throttle PCIe bandwidth due to thermal issues, maybe we need to stop pushing later generations of PCIe so aggressively...
I don't recall PCIe ever causing heat problems.

Physics has a limit.

You might have noticed that lots of PCIe 4.0/5.0 SSDs drives come with a massive heatsink. You're free to use HW from 5-10-15-25 years ago if you don't like how [hot] modern HW works/runs.

[schmidtbag]

Apparently you forgot motherboard chipsets suddenly requiring active cooling again when the first PCIe 4.0 capable chipsets became available.

I didn't forget, but that situation seems to have been somewhat alleviated - there are plenty of PCIe 4.0 chipsets that don't need active cooling or oversized heatsinks.

And there is more than just desktop computers and laptops, e.g. 5G base stations need massive amounts of computing power and bandwidth, and are often deployed in harsh environments. You don't want these to shut down completely during summer, especially when these also handle emergency services.

Fair enough, though I still think they should perhaps rethink how they approach the technology is they're getting that hot.

[avis]

I didn't forget, but that situation seems to have been somewhat alleviated - there are plenty of PCIe 4.0 chipsets that don't need active cooling or oversized heatsinks.

Fair enough, though I still think they should perhaps rethink how they approach the technology is they're getting that hot.

Try to reapproach physics. Good luck with that.

[schmidtbag]

We already see this happening with pcie5 nvme ssds and perhaps some optanes.

You might have noticed that lots of PCIe 4.0/5.0 SSDs drives come with a massive heatsink. You're free to use HW from 5-10-15-25 years ago if you don't like how [hot] modern HW works/runs.

I got the impression the SSD NAND controller is what got too hot, not necessarily the system PCIe controller.

Try to reapproach physics. Good luck with that.

Not necessary, as you should be well aware of. Look how much power a Pentium 4 consumed and how little performance it has compared to modern chips. While there is a limit to what physics can allow, we're definitely not at that limit. Hence early PCIe 4.0 chipsets already running cooler, even without the need to remanufacture them.

[erniv2]

This requires working ASPM 1st or ? So this is like a enhanced powermanagment setting, usually idle pcie devices allways should only run 1.1 speeds.

And i guess only server grade hardware implements temp sensors on anything else than the main x16 slot, most desktop boards dont even have a proper ASPM support.