No one buys desktop cpus based on geekbench scores.
It's actually Passmark scores, I never liked geekbench.
Is there anywhere some third party has independently verified this ?
They're my workloads. Since I'm needed at home (both parents are above 80 and distrustful), I do various freelance stuff, so my "work" varies quite a bit. One of the two CPU-intensive tasks is batch processing/optimizing images for blogs, and the other is keeping a few thousand bots online.
Image processing produces JPGs with Google's Guetzli encoder, WEBPs, and AVIFs. A single image would need 10 variants across those three file types in x1, x2 and x3 resolutions. This takes a lot of time on my M1 mini and a 5900X was slower per image but only marginally faster overall. See the CPU usage here on the M1, since you're familiar with Linux you might understand the numbers better than I do:
https://techenclave.com/attachments/screen-shot-2024-10-11-at-3-45-24-am-png.210862/
Sample HTML for those blogs:
HTML:
<a href="https://webhostdotcom-01.jpeg">
<picture>
<source srcset="https://webhostdotcom-01.avif 1x, https://webhostdotcom-01-x2.avif 2x, https://webhostdotcom-01-x3.avif 3x" type="image/avif">
<source srcset="https://webhostdotcom-01.webp 1x, https://webhostdotcom-01-x2.webp 2x, https://webhostdotcom-01-x3.webp 3x" type="image/webp">
<img src="https://webhostdotcom-01.jpg" alt="01" width="770" height="513" loading="eager" srcset="https://webhostdotcom-01-x2.jpg 2x, https://webhostdotcom-01-x3.jpg 3x">
</picture>
</a>
I try and keep the Google PageSpeed Insights Performance score above 90 so these variants are one of the ways I maintain that with an image-heavy page. For comparison, a similar page on Techspot scores around 60. I developed this mindset when I was young and impressionable and spent an unnatural amount of time on alistapart.com
The virtualization workloads are all on separate machines, so I don't pay much attention to them but the bots that are on Ryzen 1000 and 2000 systems are far less responsive than the ones on Ryzen 5000 systems, even with they all having 128GB of memory and KC3000 SSDs. They're web-scraping bots, and the market has been decimated with the IT downturn this year so I'm not even close to considering upgrading them. Those systems are barely making $100 in profit after electricity and internet costs —
in total. It used to be $500 this time last year.
But I would like faster processing for the image stuff, and I really want at least a 25% improvement for an upgrade to be worth it but the M4 is about 20% and so is the 9600X. The 7600X wouldn't even be 12%. Hard to justify 50k for a compute box when it's only 12% faster than what I already have.
So Its still cheaper and with more disk space and much more flexibility.
This is true, and I'm more inclined to a 9950X compute box with the cheapest possible motherboard/ram/ssd and a gold-rated power supply since these are just intermittent workloads.
The M4 mini that I
wish I could buy comes up to 240K: 10G networking, 14-core processor and 64GB of memory. But that's not going to happen. Unless a client decides to pay me with Apple Gift Cards (if you're a freelancer only accept gift cards at half value).
And upgrade costs seem quite extortionate. I bought 2tb + 2tb SSDs for roughly 20k
When I got my M1 mini a year after launch, I built a 10G TrueNAS machine with 3x 1TB NVME drives for half the price that Apple wanted for a 2TB upgrade. I understand most of the cost goes to the manufacturing overhead of low-volume SKUs, recouping R&D investments and there's the improved security factor but I don't need the security and offsetting Apple's R&D/manufacturing costs is not part of my personal life goals.
.)
