Intel Touch Host Controller Drivers Nearing The Mainline Linux Kernel

[phoronix]

Phoronix: Intel Touch Host Controller Drivers Nearing The Mainline Linux Kernel

For the past several months Intel Linux software engineers have been working on Intel Touch Host Controller drivers as an IP block on the PCH for handling touchscreen, touchpad, and related touch input devices. On Sunday the fourth iteration of these driver patches were sent out as these new Intel open-source drivers near the mainline Linux kernel...

https://www.phoronix.com/news/Intel-Touch-Host-Controller-v4

[theme118]

And when the MIPI cameras will work? What a disaster.

[M@GOid]

Speaking of Intel, hey Michael, are you aware of the recent uncovered problem with Intel GPUs on Windows? They have a huge driver overhead, and apparently any CPU more that 2 generations older will result in significant losses on performance. It would be a nice article investigating the situation on Linux. And if you can include a Alchemist GPU on the test will be a nice bonus, since YT channels didn't bother to test them yet.

[pWe00Iri3e7Z9lHOX2Qx]

And when the MIPI cameras will work? What a disaster.

IPU6 cameras have been a total shit show. They can work now, but the video quality is terrible even on high resolution webcams because there's no hardware ISP support. Also, there was a bad backport that totally broke them and rendered people's machines unbootable until they blacklisted the driver.

hansdegoede | IPU6 camera support is broken in kernel 6.11.11 / 6.12.2-6.12.4 https://hansdegoede.dreamwidth.org/29039.html

[schmidtbag]

Speaking of Intel, hey Michael, are you aware of the recent uncovered problem with Intel GPUs on Windows? They have a huge driver overhead, and apparently any CPU more that 2 generations older will result in significant losses on performance. It would be a nice article investigating the situation on Linux. And if you can include a Alchemist GPU on the test will be a nice bonus, since YT channels didn't bother to test them yet.

Actually, I think it'd be a cool benchmark to see CPU overhead for other drivers too. Might even be an interesting metric to see performance-per-CPU-usage to help put into perspective which drivers are more optimized.

[Anux]

Speaking of Intel, hey Michael, are you aware of the recent uncovered problem with Intel GPUs on Windows? They have a huge driver overhead, and apparently any CPU more that 2 generations older will result in significant losses on performance.

And none of the big testers noticed it because they all test with the fastest CPU in 720p. Regardless of no one ever would buy this GPU to pair it with a 3000 $ high end machine to play games at 720p.

It would be interesting to see if the Linux drivers have the same problem. Independent of that, the B580 has only PCIe x8 which makes it useless for older / low end machines anyway.

[schmidtbag]

And none of the big testers noticed it because they all test with the fastest CPU in 720p. Regardless of no one ever would buy this GPU to pair it with a 3000 $ high end machine to play games at 720p.

That's not really the point... If Intel doesn't fix the CPU overhead problem, I'm sure it will only get worse as they release the B700 series. They need to fix this ASAP while the B500 series isn't going to be bottlenecked by most modern CPUs.

Independent of that, the B580 has only PCIe x8 which makes it useless for older / low end machines anyway.

I imagine that a B580 on PCIe 3.0 x8 would be more hampered by the lack of ReBAR than the total bandwidth. Once assets are loaded into a GPU, games really don't demand much PCIe bandwidth at all. At least for AMD and Nvidia; I wouldn't be surprised if Intel has excessive PCIe overhead too.

[Anux]

I imagine that a B580 on PCIe 3.0 x8 would be more hampered by the lack of ReBAR than the total bandwidth. Once assets are loaded into a GPU, games really don't demand much PCIe bandwidth at all. At least for AMD and Nvidia; I wouldn't be surprised if Intel has excessive PCIe overhead too.

Well PCIe 3 doesn't exclude ReBar but missing ReBar is of course a second factor with probably the same effects. Both together would be pure hell but even one alone is enough to make this card just bad for old machines and the inefficient driver makes it even bad for modern low end machines.

[pWe00Iri3e7Z9lHOX2Qx]

Well PCIe 3 doesn't exclude ReBar but missing ReBar is of course a second factor with probably the same effects. Both together would be pure hell but even one alone is enough to make this card just bad for old machines and the inefficient driver makes it even bad for modern low end machines.

Every test I've ever seen has shown minimal impact in most games from running x16 cards at x8, even from PCIe 4.0 to 3.0. But these Xe cards essentially don't work in Linux without ReBAR. They can show your desktop just fine, but they can't deliver usable frame rates in any game. I tried an A770 in multiple workstations with oodles of free PCIe 3.0 lanes but no ReBAR. It got returned.

[Anux]

Every test I've ever seen has shown minimal impact in most games from running x16 cards at x8, even from PCIe 4.0 to 3.0. But these Xe cards essentially don't work in Linux without ReBAR. They can show your desktop just fine, but they can't deliver usable frame rates in any game. I tried an A770 in multiple workstations with oodles of free PCIe 3.0 lanes but no ReBAR. It got returned.

https://www.igorslab.de/en/pcie-4-0-and-pcie-3-0-different-between-x8-andx16-with-the-fast-fastest-cards-where-the-bottle-neck-begins/7/

And that is just "older" (slower) cards and only average FPS, the 1%lows are much worse. 5700XT was the slowest card in that test and the only one to see no drops on a 3.0x8 interface.