How to Make the Framework Desktop Run Even Quieter

Isn't the problem with Noctua (and similar "silent" fans) that they don't offer the same airflow throughput as their noise making bretheren ?

So sometimes its not as easy as "just use Noctua" ?

[0] https://news.ycombinator.com/item?id=32923226

https://smallformfactor.net/forum/threads/monochrome-2-my-cu...

His trick is to use a number of heat pipes (that transfer the heat through vaporization), and a really really big heat sink (a 5kg copper plate).

> In other words, we would only recommend upgrading to the NF-A12x25 G2 if you seek to lower noise levels as much as possible and if you are willing to sacrifice the maximum performance headroom in worst-case scenarios that the G1 HS-PWM fan provides.

Appears that it doesn't pass safety guidelines, so this is one way to get around that.

not really, like if that would be seen as bad in general then the only solution would be to keep it a secret that they "collaborated" on it

like giving you employees a bit of time to work on product related passion projects, which might even cross company boundaries and can be used in a PR context is one thing. And a good example for it is software companies reserving some time of employees to work on OSS (in a context where OSS contributions go beyond what the company needs). It boost employee moral, is positive PR, let people learn/train their skills etc.

but going from there to a product they sell is a HUGE step, like far larger then it seems

like the cost between the tinkerer project from the article and turning it into a product is more like a x-times multipler then some two digit % increase.

a good example is a previous collaboration where for a 3d printed casing for the framework 13 motherboard, where due to high demand they then decided to produce it

but for production lines you now need to meet higher quality standard and 3d printing often isn't an option, so no it needs 2 molds and in addition the screw now need to have proper stable/metal thingies you screw them into in-layed into the mold. And you need to have QA, production line inspections etc. Idk. if they made money or a loss or neutral on it but at least from a QA perspective they got burned as many of the casings had quality issues where you needed to fix them with a sharp knife or they wouldn't close properly

now companies have 3 choices

- not allow such passion projects, which sucks for everyone

- allows them, but keep them secret, which still sucks for most

- rebrand it as "some vision prototype", "experimental change to the form factor" or similar (which is what many other tech companies do), now most people are happy except the tinkerers which could just print it, let it be printed with a 3d printing service

- allow them, show them, and make most people happy except the small amount of people which really want it, aren't fine with any tinkering and blame the company for showing something nice which might not make sense to sell as a product

and in that context I really prefer it the way Framework and Noctua tend to do it

For one, the memory is soldered on because it's integral to the GPU the same way it's integral to the Apple's M3, and can be used the same up to 96gb.

At the form factor, what else are you expecting to upgrade over it's lifetime?

This statement makes it sound like there's a lot to upgrade when it's mostly just memory people seem confused about.

If you’re buying this you’re probably maxing out the memory to start with. 128GB is borderline barely adequate for local LLMs.

    “So we did actually ask AMD about this the first time they told us about Strix Halo. It was literally our first question: ‘How do we get modular memory? We are Framework after all.’ And they didn’t say no actually, they did assign one of their technical architects to really really go deep on this—they ran simulations, they ran studies, and they just determined it’s not possible with Strix Halo to do LPCAMM. The signal integrity doesn’t work out because of how that memory’s fanning out over the 256-bit bus.”

I also use noise canceling mode on my airpods pro quite often.

The advantage is that it works even when I’m in an environment I can’t control, like an airport or waiting room.

What bothers me is that we only regulate the loudest offenders (cars, motorcycles, construction), while the residual of everything else is just accepted as background. But that background adds up.

I wish we had stricter regulations not only for peak noise but also for the residual noise emitted by everyday objects. If reducing a few decibels here and there became a design goal across the board, the cumulative effect on quality of life in cities would be enormous.

I’ve relegated the Synology to the basement. I can still hear it grinding at night.

It's a bit more pronounced in CPU cooling towers, they are solid blocks where durability doesn't play a role. Noctua is good but the top performer and market leader is Thermalright which costs a third of Noctua.

All I had to do was send them the receipt for the cooler.

I don't know how you benchmark a Rolex vs a Seiko for 500 bucks. That is luxury to me. And yet it is perfectly fine to argue a Casio for 10-20 bucks performs just as well.

The GMK X2 doesn't have the Framework motherboard, but it's the same AMD APU, so the characteristics should be similar.

This really ought to be a consumer-grade LLM powerhouse, but the results were underwhelming. Due to the APU design, bandwidth between the APU and the discrete GPU is limited. From the Reddit poster "In this case adding the 7900xtx is effectively like just having another 24GB added to the 128GB."

Running an AMD GPU with the AMD APU should save the headache of dealing with both Nvidia and AMD drivers on the same host. It's possible that an Nvidia GPU might give better performance here, or there might be other optimizations to be made.

you can buy it without the case: https://frame.work/marketplace/mainboards?compatibility%5B%5...

P.S. the igpu in this really is comparable to a 'real' gpu in performance, with the bonus of being able to allocate arbitrary amount of vram

There exist solutions for that with active cables or optical cables, but that quickly gets expensive and complicated.

Can confirm. I was unaware of the large difference in different fans that weren't priced far apart previous to a Noctua recommendation. After hearing them in situ beside some other brands it is incredible the difference in sound. I strictly use them on my homelab and the spinning hard-drives make more noise than the fans.