"Beyond Stupid" Paranoid L1d Cache Flushing Looks Like It Will Try Again For Linux 5.15

[PerformanceExpert]

I wonder why the major ARM server players (Ampere, Cavium, etc) have not bothered with a desktop? If they were to sell their big ARM chip in a socketed ATX motherboard that anyone can integrate, and have it be price competitive with a Xeon E3 system, I bet there's demand for that. I'd buy one. I'd love an ARM system with ECC memory that isn't rackmount. But I guess these days developers all carry Macbooks and do their dev work in the cloud, smh.

Arm based desktops have been available for many years now, see eg. https://store.avantek.co.uk/arm-desktops.html. They are not cheap since these use high-end server chips, but you get a system that is as fast as a Milan server on your desktop.

The M1 is more than fast enough for development, however most Linux development is done in the cloud nowadays on big servers. It makes a huge difference when building large projects, even a LLVM or GCC build is much faster when you have 64-128 cores.

[Vorpal]

I'm willing to bet eventually something similar to thin clients will start to take over there, and they may even be ARM based, but they'll only superficially resemble today's desktop PCs. They'll be set up to do remote desktops centrally managed on either central premises or cloud hosted servers to better manage rapidly evolving security threats.

Depends on what you mean by "eventually" and "take over". There have been attempts at thin clients on and off since the 90s. Hasn't worked yet. There are lots of tasks where you need local processing power.

Some examples:

• Unless you are living close to a data center for cloud gaming, 3D gaming is not usable without a local GPU. Sure, more datacenters will help, but it is too expensive in low density population areas (think most of Australia except for the east coast for example, also large parts of Canada, Russia, etc).
• Video editing is in a similar situation, for much the same reasons.
• A lot of people around the world don't have high bandwidth, low latency, un-metered connections, even in the so called industrialised countries.
• Some of us need to work offline for extended periods of time with nothing but a satellite phone linking us back to civilisation. Sure, starlink might help in the future, especially at lower latitudes (doesn't work at all where I am currently), but the lag is going to be horrendous and there is nothing you can do about that since the big factor is light propagation delay at that point.

So in conclusion, for big corporate offices where people do light tasks (spreadsheets, word processing, email, etc), yes I can see it happening, maybe with a local server serving the thin clients, maybe cloud based. But it is not going to be a replacement everywhere. It is unlikely to be a replacement for home users. After all, phones only keep getting more powerful. Some of the "dock a phone to a laptop shell" kind of products might actually start getting traction though, I could see that happening.

One wildcard in this discussion is what effect (if any) the recent (last several years) increase in cyberattacks will have on this. A large cloud provider is a massive target and would cause huge expenses for a lot of companies if it went down (and issues for supply chains and national security). But it is likely better protected than local deployments. Honestly, I could see that factor weighing either way.

[PerformanceExpert]

The same thing anyone gets from having real hardware in hand to conduct development work on.

"Without a development platform, ARM in the server space is never going to make it." -Linus Torvalds

I still think Linus is wrong here. Arm is already getting significant traction in the server space without a popular development platform. The only actual need for having a system on your desk is when you do hardware bringup of a new board or CPU. Pretty much all serious work is done in the cloud, and there are plenty of cheap (and even free) Arm server offerings.

[stormcrow]

Depends on what you mean by "eventually" and "take over". There have been attempts at thin clients on and off since the 90s. Hasn't worked yet. There are lots of tasks where you need local processing power.

Not any more. That's why I said resembles thin clients. The local community college where I live already offers some of this to its technical students. What I'm talking about is a remote desktop managed by the IT team either at $CLOUDVENDOR or your enterprise IT department. This IS a desktop system, anything from a simple secretary's station, to a high end workstation many core, high RAM, X amount of storage. Only the end at your desk is a simple computer, which may or may not be "thin". It will even work for single applications instead of whole operating systems.

Not every company will take advantage of this, but those that do will be able to shift most security issues to the people that actually know what they're doing, rather than know nothing at all, or just enough to be dangerous.

This basically means the problems associated with thin clients in the past no longer apply.

Obviously this will exacerbate the Internet divide of haves and have nots. The last mile problem will have to be solved very soon, or a great deal of our digital infrastructure will continue to become obsolete.

[hoohoo]

Michael , shorely that was supposed to be a toilet in the background not a vanity basin!!

[foobaz]

There have been attempts at thin clients on and off since the 90s. Hasn't worked yet.

Thin clients already happened. Most people only use a web browser, not local apps. Chromebooks are successful. Data storage and computation happen in the cloud.

[zxy_thf]

Thin clients already happened. Most people only use a web browser, not local apps. Chromebooks are successful. Data storage and computation happen in the cloud.

Moving part of your applications to cloud doesn't mean you can move everything there (and remember cloud is EXPENSIVE)
Any latency sensitive applications won't happen in the cloud, unless some breakthrough discovery about FTL communication were made.
(Apparent this comment shows why Stadia is destined to fail: no gamers in their team, so they don't understand what gamers want)

[torsionbar28]

One of the reasons I'm excited about the idea of ARM server chips in standard form factors is for the homelab self-hosted crowd. For running a bunch of VM's for self-hosted services like Nextcloud, email, media streaming, nfs, etc. you don't need much CPU horsepower. But you do need a bunch of RAM to run the VM's, and preferably ECC since this runs 24/7. Also low power to keep electric and cooling costs low, and fans quiet, since this is in a home. The ideal platform has a CPU TDP around 40 watts, and with 64 GB of ECC memory, and at least two PCIe slots for a SAS controller, and 10Gbe card.

Of currently shipping products, the *only* CPU that meets this requirement is the Xeon W-1290T which has been sold out everywhere for a year now, and if you can even find it, retailers are price gouging asking $1000+ for this sub-$500 part. No thanks.

Previous to this W-1290T, there were no Xeons that fit the bill - the few prior gen low TDP parts only supported 32 GB of memory. You'd have to go way back to 2013 for the Opteron 4365, in order to get a 40w TDP with support for 64 GB ECC - I have several of these running to this day (on Supermicro H8SCM-F boards), because there simply is no newer replacement. IMO this is admittedly a niche market, but one where an ARM server SOC would fit perfectly.

[RussianNeuroMancer]

But you do need a bunch of RAM to run the VM's

Actually if you use containers instead of VMs 4GB RAM is much more than enough for things you mentioned. And in case of NanoPC-T4 or Libre Computer Renegade Elite (this one is also passively cooled) you can have fast swap on NVME SSD.