Hacker News Reader

Comment by FirmwareBurner | Hacker News Reader

There's just so much pointless and wasted stuff with this device it hurts.

Why have a SFF desktop PC with a laptop APU and a niche and expensive OCuLink external GPU connector, when you can just build a Mini-ITX/SFF desktop system with a dedicated desktop APU/CPU, and a dedicated desktop GPU for cheaper and much more performance and might also be quieter?

This system is basically a non-portable laptop built from laptop parts but without a display and battery, which costs nearly as much as a laptop but that one comes with the advantage of portability thanks to the display and battery.

I get the cheap Chinese NUCs, but this device basically has the disadvantages of both a desktop PC and a laptop combined, with almost none of the upsides.

View on HN

Oculink is basically free: it's a connector exposing PCIe ports that are already exposed by the APU. It should be semi-standard for SFF, because why not?

Compare versus USB4, which is quite expensive. Alas neither Intel nor AMD offer on-package USB4 in their new line-up: both require a rather expensive & serious chip for USB4 (ASM4242 for AMD). It's a great capability but it has a real cost. Where-as Oculink requires so little.

As for desktop vs mobile: this is the same Phoenix chip that AMD just released as the 8700G APU. The HS part here is targeting a smaller 45W power budget than the desktop 65W part, but otherwise similarly capable & loaded out. And the case seems designed for either some mild overclocking with it's 75w ought, which if not used means it will remain nice & quiet at stock. I think there's many many people for whom stock will be more than adequate for their uses, and who wouldn't need a bigger hotter version. And they can just tell the chip to drink more power of they want a desktop equivalent, since the cooling is there.

These chips are also boast much higher efficiency (computer/watt) at lower power levels too, which is another boon for users, another reason to look here.

Bigger hotter more powerful has a lot going for it, and it seems like it's what you want. But don't make the mistake of assuming it's what everyone wants. The NUC form factor is pretty amazing, is really a delight with how flexibly it can fit into your environment. I ran a NUC velcroes to my monitor for 3 years; it'd be impossible or cumbersome even with a svelte mini-itx build. Let's just try not to have such a harsh negative reaction so strongly out of the game, yeah?

>Oculink is basically free:

Oculink GPUS are not. Far more expensive than PCIe ones.

Ok, let's say the GPU is hella expensive. Is that a reason not to put a $0.50 connector on this system? Is a lack of expandability better? I think that's a farcical idea. It makes all the sense in the world to include Oculink!

As for the price of using Oculink? Here's a riser board and frame for $40. There are even cheaper options about that folks report as working just fine! https://www.aliexpress.us/item/3256806254754959.html

Having PCIe ports is good for more than GPUs too. There's Oculink to nvme adapters aplenty: go right ahead and plug in 2 (or 4 at reduced link width) ssds! Plug in a 25Gbit network card! Ideally some day PCIe switches make a comeback at non-exorbinant prices & we can have whole socks of this stuff; the inherent complexity is lower than pcie-transporting USB4 but we just don't have ecosystem options alas.

Sure, for most users where they want one peripheral all the time getting a much bigger mini-itx system makes a lot of sense. But it keeps seeming to me like this is a near zero cost way to create a small computer that still has really good options, that can adapt incredibly well. If you have a portable powerful compact computer, that you take with you, ideally you can just plug it in to the GPUs, network cards, storage, or other peripherals in the world around you; we're not there in most cases but I cannot emphasize enough that it costs so little to add the flexibility/option.

I've used a thunderbolt eGPU with a laptop before and it was good enough to put me off building a gaming rig for a few years. Most of my dev work these days has moved to a home server or codespaces, so I could see myself getting something like these. the GPU is typically the only thing I upgrade as tech evolves faster than my upgrade cycle (eg, I went straight from ddr3 to ddr5).

I could also see it being a good option for business deployments, similar to the absurdly popular Lenovo SFF PCs, especially where some but not all users might be pushing a GPU pretty hard.

IT can swap out an eGPU in no time, but lose the silicon lottery with an iGPU or a laptop GPU and the entire unit needs to go back. Likewise, there's a reason no IT dept builds their own PCs. It's not that they can't, it's that a factory in China can do it faster, better, cheaper.

>I've used a thunderbolt eGPU with a laptop before and it was good enough to put me off building a gaming rig for a few years

Yeah, but this is not a laptop, it's a desktop PC. This make sense for a laptop since you take it places but don't need the eGPU everywhere, but this is a desktop PC, it's not going anywhere.

So why do you need an expensive eGPU on a desktop PC that doesn't go anywhere when you can just build a PC with a discrete GPU in it, saving you money and not loosing performance from the Thunderbolt bottleneck?

It makes no sense.

>IT can swap out an eGPU in no time

Have you ever swapped the GPU in a profesional HP/Dell/Lenovo tower? It can also be done in no time. Even the water cooled ones. They're designed to be serviced quickly because time is money. They're not like your average gaming tower PC.

I’ve seen some argue on the point of energy efficiency, which might make some sense. Mobile CPUs have lower power consumption caps and tend to consume less at idle and laptop PSUs tend to be more efficient at low power draw than desktop-style PSUs.

One might ask why not just use a laptop in that situation, but depending on how the device is used (e.g. always docked), the screen and battery could be seen as a waste.

That said, I think I’d opt for a custom build too, probably in something like the Streacom DB4[0] which allows low power systems to operate fanlessly by doubling as a heatsink.

[0]: https://streacom.com/products/db4-fanless-mini-itx-case/

Yes sure, this is true-ush. But if you're so constrained on power draw that three or four extra watt idle is critical, why not just go for something really power-efficient like a Pi or similar? Also, you could undervolt one of the 65W TDP Ryzen CPUs and probably get similar results. Or if you're lucky you can get one of the "OEM Only" 45W Ryzens.

As for PSUs, an "80 platinum" rated one will have 90% efficiency at 10% power draw.

The downside of a Pi is dramatically reduced muscle that’s well below that of even many smartphones and tablets these days, which can be prohibitive depending on usage.

Undervolting is a good option I agree, but there’s evidently value in machines that run on a lower power profile out of the box.

Is Pi really more efficient though? In TOPS/watt*h?

That is actually a good question. Now that I think about it, the Cortex CPU in the Pi might not be get more TOPS/Wh than the newest AMD CPUs. The Pi's single-board design and the lack of 12V/24V circuits should make it somewhat more power efficient than a computer that uses all 4 voltages. In conclusion I have no hard data and was influenced by the Pi's reputation.

I agree. I've never gotten the Intel NUCs either from your logic, it was an overpriced desktop that is only maybe suitable for Kowloon walled city or Singapore apartments.

Reply on news.ycombinator.com