Path: ...!eternal-september.org!feeder2.eternal-september.org!news.eternal-september.org!.POSTED!not-for-mail
From: Paul S Person <psperson@old.netcom.invalid>
Newsgroups: rec.arts.sf.written
Subject: Re: ongoing infrastructure changes with AI in the USA
Date: Mon, 18 Nov 2024 08:35:44 -0800
Organization: A noiseless patient Spider
Lines: 52
Message-ID: <21rmjj5vlg3cb4jcr6dbu3f9m07v46hdru@4ax.com>
References: <vh3dim$2ehti$1@dont-email.me> <RfoZO.6$7ZKc.5@fx34.iad> <vhdf20$pbs3$1@dont-email.me> <mXH_O.18406$OVd1.4832@fx10.iad>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: quoted-printable
Injection-Date: Mon, 18 Nov 2024 17:35:45 +0100 (CET)
Injection-Info: dont-email.me; posting-host="34d39f943cd6e029e175876ccf7bee6b";
	logging-data="1410276"; mail-complaints-to="abuse@eternal-september.org";	posting-account="U2FsdGVkX1+E7/KQouFhiG9bCCA9RomAixPqhWEUzEw="
User-Agent: ForteAgent/8.00.32.1272
Cancel-Lock: sha1:7jJq3lQHvndrzYTmLnFO3B5QiKc=
Bytes: 3317

On Mon, 18 Nov 2024 14:12:34 GMT, scott@slp53.sl.home (Scott Lurndal)
wrote:

>Lynn McGuire <lynnmcguire5@gmail.com> writes:
>>On 11/14/2024 9:00 AM, Scott Lurndal wrote:
>>> Lynn McGuire <lynnmcguire5@gmail.com> writes:
>>>> I am on the periphery of the ongoing blanketing of the USA with AI
>>>> servers.  I have a few facts that might just blow you away.
>>>>
>>>> The expected number of AI servers in the USA alone is presently a
>>>> million (SWAG).  The current cost for a single AI server is $500,000=
 US.
>>>>     1,000,000 x $500,000 =3D $500 billion US of capital.
>>>=20
>>> First, What is your source for this data?  Be specific.
>>>=20
>>> Second, define precisely what an "AI server" is.
>>
>>BTW, I should have mentioned that the AI Servers are not uniform in any=
=20
>>of their aspects.  I just gave the specs for one of the high end=20
>>machines that the manufacturer has a three month waiting list for.
>
>Some of us actually produce ML hardware.  The term AI is an marketing
>gimic, not reality.
>
>ML hardware ranges from custom logic in a desktop CPU to massively =
parallel
>specialized hardware (e.g. plug-in GPUs).   An ML-enabled server can
>range from a simple ML accelerator block in the CPU itself (Apple,
>Google, Amazon, Marvell) to a large, expensive, power-hungry GPU from =
Nvidia.
>
>Yes, there are large racks of ML-enabled servers, particularly for
>training.  No, they're not really anything special other than using
>high-end, high performance CPUs, GPUs and high-speed interconnects
>(Infiniband, 100 and 400Gb ethernet).      Basically no different
>than any supercomputer in scale and power requirements.

1 million of them will rather increase the power requirements,
possibly causing shortages.

I would say "perhaps Congress should pass a law allowing local power
systems to refuse to connect to these things" but, judging from the
last two years, Congress will be doing well to be sufficiently
organized by Jan 6 to do its duty to the country, never mind passing
any laws any time soon. Well, except a massive Tax Cut for 1%-ers, of
course.
--=20
"Here lies the Tuscan poet Aretino,
Who evil spoke of everyone but God,
Giving as his excuse, 'I never knew him.'"