Intel Linux Graphics Driver To Do A Better Job Of Keeping Track Of Its Engine Busyness

[cutterjohn]

Ew GuC

Not directly related, but I'm still confused about firmware load on pre-Arc. dmesg reports clearly GuC/HuC firmware loading isn't supported and worded to sound like it fails with guc=3, but the sys option shows it definitely load (logs). Apparently there's something about the media driver and better encode or decode with HuC enabled (never tested it) but I'm not actively aware of any differences with it default/0 or forced/3 on UHD 630. I'm also not sure if it interfaces with Management Engine (I have it HAP bit disabled no HECI but GuC/HuC still loads; I vaguely recall one of those 3 being related to i915 PAVP too).

Is there something obvious GuC and HuC firmware do vs not having either where it's optional?

Intel opted NOT to refactor in their terms 'hacked' video encode support to the Xe driver from the i915 driver for the Alchemist series of card(and presumably any similar design iGPU) as it was too teh hard. So you get no video encode if you use the Xe driver w/alchemist, or have other broken graphics elements if you require hardware video encoding

Presumably meaning that Battlemage forward have some 'easier' to program for hw video encoder? Or perhaps it's the memory arch, or a combo of the two(or more things)?

So for Alchemist you get a choicing of rebooting per needs/desires switching between i915(still default AFAICT) or dual booting w/windows... Ofc most people I know still have a dedicated windows machine at least for gaming, if not for some other niche proprietary software, usually related to video and graphics production... there are also a number of scientific and engineering applications that really only run under Windows as well...

...and yes I know there are also people who use macs for similar video/graphic production workflows as it also support much of that niche software pretty well... I just have little recent experience w/OSX so can't really comment, and additionally I do no video editing, audio editing, or graphics work... so YMMV...

[geerge]

I have a hard time believing they had any big stock of those things. No retailer knew how well those cards would be received, not even Intel I think, so no reason to keep a good stock of them. And who can blame them?

Now Intel is running against the clock before Nvidia and AMD announce their new lineup in a month or so, before a good chunk of potential buyers change their minds and decide to go with the usual suspects.

Shame the latest intel CPU's are on a different node, would have been nice if they could re-allocate the silicon meant for the 285k (which must be selling like absolute garbage) to something that would actually sell.

[the-burrito-triangle]

Yeah, I was thinking the same thing. I doubt it's actually at a loss but there's no way they're making money on these cards.

Intel cannot afford to loose money on these cards. They aren't Sony with the PS5, where they will make more money off of games sold for the platform. I'm not saying Intel is raking in the profits, but they certainly are not selling at a loss.

[the-burrito-triangle]

Ew GuC

Not directly related, but I'm still confused about firmware load on pre-Arc. dmesg reports clearly GuC/HuC firmware loading isn't supported and worded to sound like it fails with guc=3, but the sys option shows it definitely load (logs). Apparently there's something about the media driver and better encode or decode with HuC enabled (never tested it) but I'm not actively aware of any differences with it default/0 or forced/3 on UHD 630. I'm also not sure if it interfaces with Management Engine (I have it HAP bit disabled no HECI but GuC/HuC still loads; I vaguely recall one of those 3 being related to i915 PAVP too).

Is there something obvious GuC and HuC firmware do vs not having either where it's optional?

Enabling both Huc and GuC works for my Tigerlake and Rocket Lake first gen Xe iGPUs. Intel decided to only enable these features by default with Alder Lake and newer. That doesn't mean they can't work on older GPUs. I think there are benefits to doing so as far back as Skylake mobile iGPUs. But the kernel will say its "tainted" when enabled since Intel doesn't want to be held responsible if enabling the GuC/HuC breaks something (like suspend/resume or something else).

https://wiki.archlinux.org/title/Intel_graphics#Loading

Anyways, the GuC is supposed to offload portions of the kernel driver to make the GPU more performant and HuC enables low power video decoding and encoding.

If it works for your GPU, I would enable it. But if you are one of those anti-blob people, don't and have a worse performing GPU.

[smitty3268]

Na that's bullshit, there is hardly any volume produced of those chips.
...
Intel probably sells them at a loss or somewhere at break even.

Yeah, I was thinking the same thing. I doubt it's actually at a loss but there's no way they're making money on these cards.

[Espionage724]

Ew GuC

Not directly related, but I'm still confused about firmware load on pre-Arc. dmesg reports clearly GuC/HuC firmware loading isn't supported and worded to sound like it fails with guc=3, but the sys option shows it definitely load (logs). Apparently there's something about the media driver and better encode or decode with HuC enabled (never tested it) but I'm not actively aware of any differences with it default/0 or forced/3 on UHD 630. I'm also not sure if it interfaces with Management Engine (I have it HAP bit disabled no HECI but GuC/HuC still loads; I vaguely recall one of those 3 being related to i915 PAVP too).

Is there something obvious GuC and HuC firmware do vs not having either where it's optional?

[Quackdoc]

I suppose this is nice, I do check it every now and then to see how far I can push these cards

[M@GOid]

I have a hard time believing they had any big stock of those things. No retailer knew how well those cards would be received, not even Intel I think, so no reason to keep a good stock of them. And who can blame them?

Now Intel is running against the clock before Nvidia and AMD announce their new lineup in a month or so, before a good chunk of potential buyers change their minds and decide to go with the usual suspects.

[Anux]

Demand ... is high ... Intel spokesperson ...

Na that's bullshit, there is hardly any volume produced of those chips. That's why they are instantly "sold out".
If you look at amazon, any other graphics card (even the long out of production Radeon rx 560) is sold more often this month.

Intel probably sells them at a loss or somewhere at break even. Have a look at die sizes:

GeForce RTX 4060    5 nm TSMC 160 mm²
Radeon RX 7600 XT   6 nm TSMC 200 mm²
Arc B580            5 nm TSMC 270 mm²​

And pair that with the news about them loosing their discount at TSMC, it is in Intels best interest to produce as few as possible of those chips or don't sell them to end users but data centers for more money. (not sure if data centers would deal with that "efficiency" class)
It also shows how the other 2 are exploiting their duopoly.

[skeevy420]

Heh, with jobs that only see movement as productivity and gain you have to learn to artificially inflate your busyness metric to show you have a good business metric. In other words, if you have time to lean you have time to clean. Fuck me for actually being efficient.

It's nice seeing Arc improvements and Intel trying to take on the Graphics Duopoly of AMD and NVIDIA.