Yeah, it sounds like this Maxio controller is quite decent and it's cheap because they don't have name recognition yet. The Acer FA200 appears to be a non-counterfeit SSD using the same controller. https://www.tomshardware.com/pc-components/ssds/acer-fa200-4...
Russia images all US bases whenever they want. That's what spy satellites are for! It's also a very small leap to assume they are selling intel to Iran. Everyone is aware of this and it's taken into account by planners on all sides.
Then you’d say that Apple doesn’t make their laptops. Foxconn does.
The kind of work ARM would do to “make” a chip themselves goes beyond just design. It’s synthesis, P&R, test, packaging (generally a different company than the fab), yield management, inventory/logistics, etc.
Context: Early in the firmware boot process the memory controller isn't configured yet so the firmware uses the cache as RAM. In this mode cache lines are never evicted since there's no memory to evict them to.
There may be server workloads for which the L3 cache is sufficient, would be interesting if it made sense to create boards for just the CPU and no memory at scale.
I imagine for such a workload you can always solder a small memory chip to avoid having to waste L3 on unused memory and a non-standard booting process so probably not.
Most definitely, I work in finance and optimizing workloads to fit entirely in cache (and not use any memory allocations after initialization) is the de-facto standard of writing high perf / low latency code.
Lots of optimizations happening to make a trading model as small as possible.
I remember the talk about the Wii/WiiU hacking they intentionally kept the early boot code in cache so that the memory couldn’t be sniffed or modified on the ram bus which was external to the CPU and thus glitchable.
Today a local AI box (Strix Halo, DGX Spark, Mac Studio) is $2,500+. Even if it comes down, almost no one will pay upfront when they could pay $20/month instead.
Small models can run on a normal PC but similar or better quality models are free in the cloud.
Not only are third party GPUs not supported on apple silicon, but thunderbolt has significantly more latency and lower bandwidth than 'real' PCIe implementations, even ones with similarly cut down lanes like oculink.
Apple tried before to push everything out into external PCIe enclosures and people hated it. Maybe this'll go differently this time, the Mac Studio is certainly a much more compelling offering than the trashcan Mac Pro. But I think this is still a shitty and painful situation for a lot of specific users.
Oh yeah, the market for these capabilities is tiny, no doubt. But at least historically, the people that wanted these things tended to be very big lucrative customers, and also tended to be very influential word-of-mouth Apple evangalists.
Just as a random personal example, my uncle was an Apple guy since the 80s, and when I was a kid in the early 2000s he always had a Mac Pro, several Macbook pros, and a bunch of other Apple gear. He played a big role in convincing all of his 3 other brothers that Apple was the way to go, and those brothers then raised kids in households with only Apple computers.
This uncle is still a mostly Apple user, but he's increasingly pissed at Apple, and definitely no longer evangelical in recommending it to people. He needs to have a linux server machine for his home NAS, and another one for some specialized work applications and it really frustrates him that Apple abandoned his market segment.
I think there's a real possibility that if Apple had pissed him off like this in the early 2000s instead of the 2020s, that our wider family might have not ended up as being so Apple centric, so Apple may have missed out on a lot more sales than just a couple of Mac Pros and expensive software licenses.
reply