Intel's answer to PCIe 6 heat concerns is a driver that throttles bus speeds

zohaibahd

Posts: 91   +1
Staff
A hot potato: As PCIe versions continue pushing for higher bandwidth with faster transfer rates, the increased speeds are indirectly leading to excess heat generation – a challenge Intel is tackling in a whole new way with an updated Linux driver. The open-source "PCIe bandwidth controller" aims to automatically throttle link speeds when thermals reach a certain threshold.

The core issue is that PCIe's ever-increasing speeds demand better signal integrity and lower signal loss. Achieving that often requires higher clocks, more power, and encoding optimizations – all of which generate extra heat. Take PCIe 5.0 SSDs as an example. With 32GT/s transfer rates (double that of PCIe 4.0), they frequently need dedicated cooling solutions to dissipate heat under sustained loads.

Intel's driver, which has been in development for over a year, lets Linux selectively downshift the PCIe link speeds when thermals get too high. It registers a "cooling device" for each PCIe port that allows tweaking the link speed. Slowing down that PCIe link should help alleviate thermal pressure from hot-running GPUs, SSDs, and other peripherals.

"This series only adds support for controlling PCIe Link Speed," wrote Intel engineer Ilpo Järvinen in the latest patch notes. "Controlling PCIe Link Width might also be useful but AFAIK, there is no mechanism for that until PCIe 6.0 (L0p) so Link Width throttling is not added by this series."

As you can see, the inability to adjust link widths is a current limitation, but Intel is eyeing that capability too.

For now, the new Linux driver focuses solely on link speed throttling to mitigate thermal issues. It's a straightforward software solution to the challenging heat problems posed by ever-faster PCIe speeds. By letting the OS automatically reduce those link rates on a per-port basis, it provides a way to dynamically optimize thermals under load.

While the driver seems primarily targeted at servers initially, the tech could prove highly useful for desktop and mobile systems as well. For instance, an x16 GPU link could temporarily drop to x8 speeds if things run too hot. Of course, that would mean reduced bandwidth which would likely have an impact on performance.

Last year, the PCI-SIG consortium released the first 0.3 review draft specification for PCIe 7.0 to members. Details are still scarce, but the announcement confirmed PCIe's evolution is proceeding on schedule, with 7.0 set to double 6.0's bandwidth for a staggering 512GB/s throughput via x16 links. It's not hard to believe that upcoming standards will require more aggressive cooling.

The new PCIe bandwidth controller driver has been through five revisions so far as Intel engineers continue refining the thermal throttling implementation. It's not finalized yet, but adoption is likely coming in the next year or so.

Permalink to story:

 
Is there any part of an intel rig that doesn't gets sold with an unsustainable volt problem? What's next downthrottling the NIC and Wifi chips too?

At least their laptops chips can be pretty efficient I'll give them that...You can't really push them otherwise you'll immediately throttle those too but people usually want battery life instead of high performance on laptops so it's ok.
 
If this power requirements for all PC parts continues this way we going to need dedicated PC rooms with AC and watercooling like servers do. But at least the fire hazard is controlled with software, for now. Good job Intel.
 
First we had liquid cooled CPUs, then liquid cooled GPUs, next we might as well just liquid cool the whole motherboard (or the whole system). I wonder if submersion cooling at the consumer level will become a thing in a few years (certainly not widespread, but available from niche providers).
 
First we had liquid cooled CPUs, then liquid cooled GPUs, next we might as well just liquid cool the whole motherboard (or the whole system). I wonder if submersion cooling at the consumer level will become a thing in a few years (certainly not widespread, but available from niche providers).
A board with soldered gpu and cpu would be much simpler (cheaper) to cool with one water cooled block.
If this thing took up, eventually the coolers would become as cheap as 1 section AIOs. There is nothing complex, it is just a reservoir and a very cheap pump.
 
If this power requirements for all PC parts continues this way we going to need dedicated PC rooms with AC and watercooling like servers do. But at least the fire hazard is controlled with software, for now. Good job Intel.

Nah, we will have a dedicated mini AC attach directly to the PC case to make sure that all the components are cooled. In that way, we just need 1 exhaust fans, and we will have plenty of spaces to circle cooling air xD

I think it will be cheaper than buying a water cooling kit + a AC for the entire room. Imagine if you can keep a constant 16-18 degree air flowing directly into your components 🤩🤩

 
Funny how everyone is jumping down Intel’s throat but this isn’t an intel problem in any way…they have to adhere to the spec like everyone else or it would kinda defeat the purpose of having a spec no? Besides, they were talking about lowering speeds to PCIe devices if those devices get hot. So blame SSD manufacturers for making drag racers and not a Toyota prius
 
Brilliant- make something fast, but since it’s not sustainable, build in something to make it slower…

Or… maybe just stick with the “slower” speed that doesn’t need to be slowed down in the first place?


Unless we’re talking about supercomputers or high end server racks… which will be the initial target market for this spec.
 
As always I'm mildly surprised by the "Intel is bad" replies. Sure, mistakes were made but AMD and other manufacturers aren't immune. It's just how it works in hardware development.
Of course as an OLD nerd I'm no stranger to fanboyism but I never really understood it. ALL the people who run these companies want to brush their teeth with Dom Perignon while smoking Montecristo's in their Mercedeses off our money. There are no "good" or "bad" companies in this business. I've always it somewhat unsettling that there are people who lie awake at night worrying about the hardware (or software) people they'll never meet use. Obviously it's not REALLY about hard- or software..:)
 
Back