Most of them are.

It’s so much better than The Wild Robot!

Amid the DeepSeek surge, smaller firms using AI may turn to RISC-V chips for cost savings. The report notes that even if a 10 million yuan RISC-V setup delivers 30% of NVIDIA or Huawei’s performance, purchasing three could still be more affordable.

Yup. If these guys come up with a half-decent CPUs they won’t have to pay the ARM, Intel or AMD premium, they can significantly decrease the cost of compute in a whole lot of sectors and use cases. And if they open source and / or sell those chips to everyone…

And I think it’s just a matter of time for them to develop those designs.

A swastica on the back of the phones would increase the accuracy by 35%.

I’m also looking forward to working a couple of days a week, training and coaching young developers.

Alright so this is where the next great cores are likely to come from.

As far as I read LPCAMM in its current state does not work for this. The electrical noise is too high. These things aren’t the same. A repairable waterproof phone can be made without glue by making it a bit thicker. In the case of RAM today, we’re hitting fundamental physics limitations with speed of electricity and noise. At this point the physical interconnect itself becomes a problem. Gold contact points become antennas that induce noise into adjacent parts of the system. I’m not trying to excuse Framework here. I’m saying that the difficulty here borders on the impossible. If this RAM was soldered and it had bandwidth no different than SODIMM or LPCAMM modules then I’d say Framework fucked up making it soldered, majorly. As I said, there’s no point buying this if you don’t care about the fast RAM and use cases that need it like LLMs. Regular ITX board with regular AM5 is the way to go.

E: To be clear, if this bandwidth could be achieved with LPCAMM, then Framework fucked up.

You get fast memory as a result. If you don’t care about the fast memory, there’s no good reason to buy this, with their motherboard. There’s a use case this serves which can’t be served by traditional slotted memory and the alternative is to buy 4-5 NVIDIA 3090/4090/5090. If you want that use case, then this is a pretty good deal.

My AM5 system doesn’t post with 128GB of 5600 DDR5 at higher than 4400 at JEDEC timings and voltage. 2 DIMMs are fine. 4 DIMMs… rip. So I’d say the present of DIMMs is already a bit shaky. DIMMs are great for lots of cheap RAM. I paid a lot less than what I’d have to pay for the equivalent size of RAM in a Framework desktop.

No joke these cars were pretty good. Short haul, long haul, as long as you had good parts, it ran great. And I took the drunken workers as a joke. In reality they were made pretty well. People were driving 30+ year old Trabants in the 90s with no issues as long as they could find parts. As genuine parts disappeared, people stopped driving them.

Trabants were fucking amazing. The most reliable car I’ve driven. Fuel pump problems? No fuel pump. Oil pump problems? No oil pump. Radiator/water pump problems? Air cooling. Ignition timing problems? Dedicated per-cyl ignition switch. Rust? Cardboard. Had a fender bender? Replace the panel in half an hour, painting is optional. Getting killed in high speed accident? No high speed.

Don’t badmouth cardboard derivatives. The Trabant didn’t fall apart like this!

I was trying to figure out how graphene would achieve this. By smearing some on the pins? But then why graphene and not just graphite? And how do you keep the charging pins from being shorted?

What you’re experiencing isn’t hyperinflation. Hyperinflation is more like when a loaf of bread is $1 today, $2 a month after and $10-100 by the end of the year. Grown up in country during hyperinflation.

Right so I guess the question of 3 is whether it means 3 backups or 3 copies. If we take it literally - 3 copies, then it does protect from user error only. If 3 backups, it protects against hardware failure too.

E: Seagate calls them copies and explicitly says the implementer can choose how the copies are distributed across the 2 media. The woodchipper scenario would be handled by the 2 media requirement.

Hm I wonder why snapshots wouldn’t satisfy 3. Copies on the same disk like /file, /backup1/file, /backup2/file should satisfy 3. Why wouldn’t snapshots be equivalent if 3 doesn’t guard against filesystem or hardware failure? Just thinking and curious to see opinion.

• Primary ZFS pool with automatic snapshots
  • Provides 3+ copies of the files via snapshots (3)
• Secondary ZFS pool at a different location replicates the primary
  • Provides more copies of the files (3)
  • Provides second media (2)
  • Is off-site (1)

Does this make sense?

The answer for doing this on your PC is Docker with one of the open source containers built for this. For example this.

If you want a dedicated machine, your Pi would do just fine but you’ll have to connect your storage where the downloads are to it. You should still use Docker for that since it’s the easiest setup and it would take care of important corner cases such as blocking seed when the VPN is down.

If Raid is backup, then Unraid is?

Try ZFS send if you have ZFS on the other side. It’s insane. No file IO, just snap and time for the network transfer of the delta.