EXTERNAL LINK

These chips don't exactly "repair" themselves. They just switch the dead cores off. Definitely not useful for a personal computer.

Good idea :)

You're probably right, I also assume that this technology will not for the mass market due to different economic implications. I think those are specific sectors that could be concerned, probably those whose purpose or priority is not money.

Now it is common to sell failed chips as less capable. And this is normal and was normal and fine.
Producer made 4 core processor, but found one failed and sell as 3 core one.
But if then more would fail - you are able to request warranty replacement.

When process technology is constantly shrinking parts are getting constantly more unreliable in time.
Ultra-very-small transistors are prone to wear in short time.
That's why that "technology" got invented.
Sell customer say 8 core CPU, within some short time half will fail but 99% of customer will not notice it and will not want replacement.

I think that this rationalization will be implemented mainly where the security is required. It is probable that other niches could apply but for other reasons.

In professional applications, loosing one CPU is probably better than loosing an entire blade, all provided the Os can deal with this situation.

Greetings,
Thomas

Now, vendors start selling such racks without all CPU slots equipped with hardware. However, to be able to deliver them with full memory, *something* then has to replace the CPU in the then otherwise un-occupied slot, only to allow the occupied CPU slot to reach memory behind it. Vendors now instead replace this CPU by a "dummy" ASIC that that has nothing more to do than to implement the bus-access protocol of a CPU to provide access to memory local to the slot. Of course, these dummy ASICs don't do any computation. They only act as bus-bridges.

It's all reality already, you know.

I don't remember the vendor, though.

a) detect the fault - which is what this article described)
b) perform RECOVERY - which is NOT what article described.

Recovery means taking over CPU state to another CPU and continue the task.

This is what IBM z-Architecture mainframes do.

Without this it's nothing more than ability to disable any processor
but every faulting circuit WILL mean crash of whole system.

This is what you can at present without this invention.

But this invention is to HIDE it from customer.
Which is wrong, except for trash-consumers that doesn't even use that processing power, and don't even know how to check how much is available.

To increase stability you have to make redundant circuits to control EVERY step of a CPU, and detect fault and BE able to recovery.

Extreme case are IBM mainframes CPU. Actually for each processor there are two of them and extra logic that controlsthe results.
the extra logic is redundant too.
When single fault is detected - cycle is restarted. This fixed
failures caused by noise, natural radiation  etc.

When multiple faults in short time are detected - it means failure of circuit, and processors are halted and FULL STATE of CPU is extracted and pushed to spare CPU and restarted.

In user point of view nothing happened, except information about failure.
if no spare CPUs are left you have to replace boards.

Not everyone not sharing your crude opinions is "stupid" or "trash". They may just have different opinions or motivations beyond the ones you're understanding.

Hiding=user is signalled of a problem.

I am 99.9% sure this invention is only to be able to hide things from user.

The article doesn't support your claim. It only states that it'll disable damaged cores, it does not say whether or not the user will have access to this information. You're jumping to conclusions you have no basis for.

As for who will use it, consumers are not major purchasers of CPU's with high number of cores as very little consumer software takes advantage of it.

People who want fast servers are. I'm taking delivery of our first 32 core machine today. We could've gone for 48 core but it wasn't really cost effective yet (it'd mean going for 12 core AMD chips, and they're not cheap). It takes up less space in terms of volume than an average consumer level desktop PC.

The larger the number of cores, the higher the cost of keeping stability up. If I could for example get a 48 core version for what I'm paying for the 32 core now rather than the extortionate prices I'd actually have to pay, I'd jump at the opportunity even if a core would fail every 6 months as long as the hardware would detect it and keep running with less capacity.

It's typically cost effective for us to replace hardware every 3 years or so (because of the cost of space and power and operational costs of managing the hardware vs. improving performance of new servers) so we'd end up with more capacity for the same money, even if the total capacity would drop off over time.

EVEN if you're right and this invention won't support automatic recovery (and I don't think so - it explicitly describes allocating the chips workload to available cores, which would seem to at least imply that it's being designed to keep executing if a chip fails), it'd *still* be far better for us to take a crash and a reboot on a core failure than having to yank a server for repairs (and no, z-Architecture is not an option - far too expensive for what we do)