Intel 13/14 gen instability: Shady practices, terrible response, failure to act

To be fair, Intel's latest solution to the Raptor Lake/Alder Lake problem, which is the extension of the warranty time by two years, is the correct solution, if their pending microcode update will really prevent any further degradation of the CPUs.

This is not a design error that affects all CPUs, which would justify a recall. This appears to be just a wrong estimation of the correspondence relationship between the failure times that can be determined in accelerated tests and the failure times in the conditions normal at a customer site. This is hard to avoid when there is not enough experience with a new manufacturing process.

The CPUs owned by the customers may be affected or not. Those defective should be replaced and the extension of the warranty should cover the case of the partially degraded CPUs, which will develop defects only later.

Unfortunately, there are good chances that instead of streamlining the RMA process, Intel may attempt to make it as tedious as possible, with the hope that many customers may give up to complete it.

When AMD has launched the first Ryzen in 2017, I have preordered the most expensive model, Ryzen 7 1700X and the best ASRock motherboard for it, thinking that it would be good to encourage some competition for Intel.

Unfortunately, the first Ryzen had an exceptionally ugly bug, which did not prevent its use for playing games in Windows, but which made impossible its use for any serious purpose, like compiling some software project in Linux.

For several months AMD tried to deny that a bug exists, then eventually it accepted to replace any Ryzen upon a RMA request.

Nevertheless, in order to replace the CPU, I would have had to ship it from Europe to USA, which was expensive. Moreover, because AMD refused to give any explanation about the nature of the bug (which manifested as some kind of address mistranslation caused by a race condition when all threads were busy, and which resulted in data corruption, because wrong addresses were accessed), I could not be sure that a replacement CPU would be free of bugs.

Therefore I have accepted a total loss of over EUR 800 and I have dumped as garbage both the Ryzen CPU and the now useless ASRock MB, salvaging only the DIMMs, which I used in a replacement system with Intel Kaby Lake.

Despite this, I did not held a grudge against AMD, so later I have bought more and more AMD CPUs, upgrading most of my older Intel CPUs with them, because since 2019 Intel has become unable to launch any product that can compete with AMD for the applications in which I am interested, with only one exception, the Intel CPUs that use only Atom cores a.k.a. E-cores, like Alder Lake N or Amston Lake, and which are cheap enough to be used in very cheap computers, i.e. in computer systems with a total price in the $50 to $300 range. For such cheap computers, the Intel CPUs are still better than any competitors, including those with Arm cores.

This year Intel has shown again a very interesting E-core, Skymont, with great improvements in performance over their older Gracemont/Crestmont cores. On the other hand, their new big Lion Cove core, while decent, is far from impressive. It should have a performance very similar with Zen 5, perhaps marginally better, but at a much greater area and power consumption in an identical manufacturing process (judging from the comparison between the previous Redwood Cove core with Zen 4). The Intel big cores of Lunar Lake and of "Arrow Lake S" will be saved only by being made in a superior manufacturing process by TSMC, because Intel has outbid AMD for it.

For the applications in which I am interested, Lunar Lake will be either too weak (in comparison with AMD CPUs with AVX-512 and non-shared L2 caches) or too expensive (in comparison with Intel Atom CPUs). On the other hand, if Intel had launched a replacement for Alder Lake N/Amston Lake, using the Skymont cores of Lunar Lake, that is something that I would have bought immediately.