Path: ...!eternal-september.org!feeder3.eternal-september.org!news.eternal-september.org!eternal-september.org!.POSTED!not-for-mail
From: Paul <nospam@needed.invalid>
Newsgroups: alt.comp.os.windows-10,comp.os.linux.advocacy
Subject: Re: Gaming Laptops
Date: Mon, 3 Feb 2025 19:24:37 -0500
Organization: A noiseless patient Spider
Lines: 101
Message-ID: <vnrmo5$1hlel$1@dont-email.me>
References: <vm40cl$21e8l$2@dont-email.me>
 <6h1bojt7kdp4d5euq0f78rtuvqpg7edc3e@4ax.com> <vm86er$2u8jo$1@dont-email.me>
 <cqlfoj93e6jvua3is08kbm6f9p32h8cl4a@4ax.com> <vm8o1d$313ov$1@dont-email.me>
 <d63goj9qcpdk1q2o6ah4r1sq5r776dfdb7@4ax.com> <vm976v$33jmh$1@dont-email.me>
 <vm9nn9$36us4$1@dont-email.me> <vmaq5q$3fd0d$1@dont-email.me>
 <vmbeia$3jepl$1@dont-email.me> <fdjiojt0fj9ft4plm2eu47itkdsi21mdcl@4ax.com>
 <m06k2qFj9dpU1@mid.individual.net> <vnmdsp$biin$1@dont-email.me>
 <tiuupjp1p0k2569aacau6e4alabj5h431k@4ax.com>
 <slrnvq28fn.3finp.candycanearter07@candydeb.host.invalid>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
Injection-Date: Tue, 04 Feb 2025 01:24:42 +0100 (CET)
Injection-Info: dont-email.me; posting-host="f9791869256f0ee8bbc66450f3e69058";
	logging-data="1627605"; mail-complaints-to="abuse@eternal-september.org";	posting-account="U2FsdGVkX1/i7Bt/MZD+RdlMmJVLnaNOpcsdGBC8BmU="
User-Agent: Ratcatcher/2.0.0.25 (Windows/20130802)
Cancel-Lock: sha1:j3Ntxrzbgijd0fcUR7UXuzgvpyM=
In-Reply-To: <slrnvq28fn.3finp.candycanearter07@candydeb.host.invalid>
Content-Language: en-US
Bytes: 6325

On Mon, 2/3/2025 3:10 PM, candycanearter07 wrote:
> chrisv <chrisv@nospam.invalid> wrote at 14:05 this Sunday (GMT):
>> Paul wrote:
>>
>>> vallor wrote:
>>>>
>>>> After updating it to Linux Mint 22.1, and while looking at the output
>>>> of lspci(8), discovered this:
>>>>
>>>> 0000:00:08.0 System peripheral: Intel Corporation 12th Gen Core Processor 
>>>> Gaussian & Neural Accelerator (rev 02)
>>>>
>>>> There's an out-of-tree driver Linux driver for it, and apparently Intel
>>>> is working to get it into the Linux kernel.
>>>>
>>>> Having said that, can't imagine why I'd use it.  (I guess perhaps Windows
>>>> Copilot might use it, but I'll defer to others regarding whether or not
>>>> that is the case.)
>>>
>>> Your guess is as good as any.
>>>
>>> https://edc.intel.com/content/www/us/en/design/ipla/software-development-platforms/client/platforms/alder-lake-desktop/12th-generation-intel-core-processors-datasheet-volume-1-of-2/003/intel-gmm-and-neural-network-accelerator/
>>>
>>> Even the person making a URL for the article, was running out of letters.
>>
>> Interesting.  I hadn't even heard of this GNA thingy before.
>>
>>> In a strange twist of fate, it's being used as a Direct Render Manager "thingy".
>>> No resource goes wasted, I would guess.
>>>
>>> https://www.phoronix.com/news/Intel-GNA-To-DRM-Driver
>>
>> The article notes that Linux does not yet have a dedicated AI
>> accelerator subsystem.  I wonder when that's coming, and if we should
>> be afraid.
> 
> 
> Well, I probably won't use it, but that is I /guess/ neat.
> 

The NVidia driver and the CUDA kit, are the most likely to expose
the bottom layer of a model you might want. The rest of a desired model
or activity, could run in userspace.

AMD also has capabilities, but they tend to be packaged
in the most expensive cards (7900XTX). And due to the level of market
penetration, AMD does not currently have a large portion of the
high end video card market. The number of people adding AMD kits
to their PC, that's going to be a smaller number of people.

But once some hardware drivers are in there, for one thing or another,
the models you want to experiment with, will be in userspace.
Maybe you get a copy of LMStudio and their launcher,
and load something on your system. With the understanding that the
datacenter version has a lot more hardware horsepower (but
still gives answers relatively slowly).

There are people who have been working on ONNX/DirectML for
the last four years, but it's hard to say how many of
those efforts are ready for prime time.

I only learned a tiny bit about this, from someone who wrote in
and complained his astronomy program wasn't running right. And
that's an image processor that uses Machine Learning to
adaptively process astronomy pictures. The code was basically
malfunctioning, right at the stage it was running hardware
detection, and the program would not load while it was
figuring out that only the user CPU was available. The
code was tripping up poking hardware that could not
possibly do the job for him. And it turns out there are a
lot of libraries and stuff to load, to do all those
detection processes properly. An "abomination of initialization".
And I could see that taking at least another year, to set right.
That's not a job for the program dev to fix, it's a library
developer issue.

One way to set that one right, would be to have a control panel,
with a "CPU button", click the CPU button and tell the initialization code
to "go away and stop bothering me" :-)

I've had this problem, on a few attempts to run unique things.
I could get the CPU to run the demo. I couldn't get my video
card to run it. The setup just refused, and I'd loaded all
the drivers and the CUDA kit. The correct version of everything
is nicely packaged by Canonical for you to use. That wasn't the problem.
I've had other situations, where I was trying to build a package
that involved CUDA, the compile stopped and it would tell me
"library mismatch". But it would not tell me what version the
two incompatible ends were using, so I could figure out which
2GB thing I needed to download. That's the value the packaging
guys add to your distro, is they make sure a reasonable set of
aligned things are available in the tree.

But with computers, not everything in life is like a trip
to the restaurant. A plate does not come out with your meal
ready to eat. Instead, with computers, it's like the grocery store,
you have a bunch of potential ingredients, but you have to cook them.
And there are lots of ways to screw that up. Maybe your oven
isn't big enough to cook a water buffalo.

   Paul