Path: ...!eternal-september.org!feeder3.eternal-september.org!news.eternal-september.org!.POSTED!not-for-mail
From: Martin Brown <'''newspam'''@nonad.co.uk>
Newsgroups: sci.electronics.design
Subject: Re: Raspberry Pi5 versus other cheap Intel based boxes for general
 computing
Date: Fri, 5 Apr 2024 09:26:52 +0100
Organization: A noiseless patient Spider
Lines: 49
Message-ID: <uuocke$18vkc$2@dont-email.me>
References: <uuip0m$2sb2$1@solani.org> <uuj77i$3r4ca$1@dont-email.me>
 <bo4r0jd46vh4pklpi5ct5i205jotqet0ub@4ax.com> <uulqah$i64f$1@dont-email.me>
 <6egt0j9op9q1bldleihqg0161q76va6chi@4ax.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Injection-Date: Fri, 05 Apr 2024 08:26:55 +0200 (CEST)
Injection-Info: dont-email.me; posting-host="98df5e3fdae107ba0e29d1ff508967ca";
	logging-data="1343116"; mail-complaints-to="abuse@eternal-september.org";	posting-account="U2FsdGVkX19+q3VBpzCha5c9z3wxJ4OFlmSW7l4tLZjHIkb0Ds8HnQ=="
User-Agent: Mozilla Thunderbird
Cancel-Lock: sha1:CAfKDVEDVvpZ5a7jRwhqo/tH+aI=
In-Reply-To: <6egt0j9op9q1bldleihqg0161q76va6chi@4ax.com>
Content-Language: en-GB
Bytes: 3477

On 04/04/2024 16:11, John Larkin wrote:
> On Thu, 4 Apr 2024 10:02:06 +0100, Martin Brown
> <'''newspam'''@nonad.co.uk> wrote:

>> We have sort of hit a point where CPU improvements especially for single
>> threaded code have hit an insurmountable bottleneck. There is a sweet
>> spot for the amount of ram and fast disk.  If you put it onto a UPS and
>> enable all go faster options for the SSD cache write through (risking
>> potential data loss if power is ever lost) you may get some improvement.
>> You would have to decide if the speed gain is worth it to you.
> 
> Most of us have way more compute power than we need. A pokey old
> laptop will show a movie just fine. About the only compute-limited
> things left are games and Spice.

Playing back an MPEG video is essentially trivial load on any modern 
machine. Video editing, rendering video content and editing large images 
is still compute intensive.

Normal office work barely taxes even the most puny hardware.

> Well, Spice is sort of a game too.

In the sense that it is a simulation of reality.

>> If you check LT Spice on various CPUs I think you will find it
>> correlates closely with ram speed and single thread performance on CPU
>> related benchmarks (obviously with a bias towards floating point code).
> 
> Enabling more cores doesn't help much. What does seriously help -
> sometimes 20:1 - is relaxing some of the sim parameters. I do that
> until something obviously breaks, then back off a little.

There might be an advantage in going to 2 or 3 cores for some problems 
depending on how smart its use of additional cores actually is but after 
that you run into memory and/or disk bandwidth problems pretty quickly.

Obviously relaxing the error bounds will speed things up but you have to 
be careful. LT Spice is deliberately conservative in its choice of 
parameters so that the solver mostly stays inside a numerically stable 
simulation (unless you provoke it with a nasty network singularity).

It is even worse for chess problems where the extra cores can easily end 
up doing work deep down the tree that will *never* be needed because it 
will be pruned by a higher level algorithm at some stage.

-- 
Martin Brown