Intel seems to be bleeding out ...

[madires]

New twist in the story. Rumours claim that Intel did know of the microcode issue much ealier and delayed things on purpose to provide the microcode fixes after the release of AMD's Zen 5. It's assumed that the microcode fixes will decrease the CPU performance and Intel is afraid that Zen 5 might take the lead when media outlets publish first benchmarks. Users would compare Zen 5  to unpatched Intel CPUs before the microcode fixes will slow them down. Now to the latest from AMD:
-  AMD delays its Ryzen 9000 launch due to unspecified quality issue (https://www.tomshardware.com/pc-components/cpus/amd-delays-its-ryzen-9000-launch-due-to-unspecified-quality-issue-new-launch-in-august-chipmaker-pulls-back-all-units-shipped-globally-for-quality-checks)

Is there really a quaility issue or are they trying to counter Intel's move? 

[SiliconWizard]

Yes, that's looking bad for Intel. First time they are going through something this problematic in a rather long time. Problem is that it's likely to damage trust beyond these affected series. We'll see.

It is not that long:

1994 - Pentium FDIV
2017 - Atom Clock Signal Issue
2018 - Meltdown and Spectre

1994 was 30 years ago. And while it didn't look good, it wasn't that bad in practice. It was easy to work around, and FP division was not really a performance bottleneck for any of the common uses for this chip.
I never heard of the Atom clock thing, can't tell how bad it was, but I don't remember it making that much noise compared to the current situation.
As to exploits, almost all vendors of modern CPUs have, or will face similar issues. New ones appear on a regular basis. It's very hard not to with speculative execution. And while again this didn't look good, there's a workaround that works fine without degrading performance that much.

So IMO, the current situation is much worse, because it doesn't have to do with a "silicon bug" that is easy (if costly) to fix and/or can be worked around easily in software (and I'm not sure their "microcode" mitigation is gonna make the chips as reliable as the previous generations). This is potentially going to erode trust in a much more severe way, in a market that has changed as well and in which competition is much more severe.

[iMo]

New twist in the story. Rumours claim that Intel did know of the microcode issue much ealier and delayed things on purpose to provide the microcode fixes after the release of AMD's Zen 5. It's assumed that the microcode fixes will decrease the CPU performance and Intel is afraid that Zen 5 might take the lead when media outlets publish first benchmarks. Users would compare Zen 5  to unpatched Intel CPUs before the microcode fixes will slow them down. Now to the latest from AMD:
-  AMD delays its Ryzen 9000 launch due to unspecified quality issue (https://www.tomshardware.com/pc-components/cpus/amd-delays-its-ryzen-9000-launch-due-to-unspecified-quality-issue-new-launch-in-august-chipmaker-pulls-back-all-units-shipped-globally-for-quality-checks)

Is there really a quaility issue or are they trying to counter Intel's move? 

This is what I asked above - how do they measure the actual (on the fly - while running the cores) I, V, P and Temperature in their CPUs??  Both AMD and Intel push their silicon to the absolute limit without knowing the actual A/V/W precisely, imho.

My bet their measurements are +/- 20% easily (again - you have to measure 1-350A, 0.03-1.6V, 0-300W, on chip 25-100+C temperature with perhaps 8-10bit on chip ADCs pretty fast). Therefore AMD is rather double checking the voltage regulation stuff before the launch (also messing with their microcode only as the new silicon will not be ready by mid August "this year" if any fix needed).

[wraper]

^AMD says it's packaging, not silicon issue.

[iMo]

^AMD says it's packaging, not silicon issue.

Packaging will not be the matter for elaborating the "quality problem" with mobo vendors, I would say. Also it is perhaps not the silicon issue, rather the similar annoyance as the Intel is facing..

While topping the clock speeds the elevated temperature may degrade the chip's metalisation/bonding, heatsink-die attachment material, etc. Now, the big question is how trustworthy the Intel and AMD chips show the die temperature and how they measure the I/V/P.

For example you may set the microcode such their CPUs report lower values, while in reality it is all over the edge and the chips will die after a couple of months. None blogger is able to measure the real values, btw., they just comment what they see in HWinfo or AIDA, etc..
Something similar to the Dieselgate..

It could be both Intel and AMD decide to start showing real numbers and both will go down with the benchmark figures then. And the chips will happily live forever

[wraper]

^AMD says it's packaging, not silicon issue.

Packaging will not be the matter for elaborating the "quality problem" with mobo vendors, I would say. Also it is perhaps not the silicon issue, rather the similar annoyance as the Intel is facing..

Do you understand what chip packaging means? It's not about cardboard box.

[iMo]

^AMD says it's packaging, not silicon issue.

Packaging will not be the matter for elaborating the "quality problem" with mobo vendors, I would say. Also it is perhaps not the silicon issue, rather the similar annoyance as the Intel is facing..

Do you understand what chip packaging means? It's not about cardboard box.

I think I do understand the meaning of the chip packaging..
As the bloggers report the "AMD is cooperating with mobo (aka motherboard) vendors" to resolve the issue.. So it is not about the chip packaging but the microcode..

PS: Moreover, you cannot solve a "chip packaging issue" 25 days prior the release, no way..

AMD is simply setting up the new more conservative values in their microcode engines for the temperature vs core(s) frequency vs V/I/P settings, as they know the Intel will do the same (Intel wrote that) till the mid of August this year (the as of today by AMD indicated new release date of their new Ryzen 9000 series).

[coppice]

Yes, that's looking bad for Intel. First time they are going through something this problematic in a rather long time. Problem is that it's likely to damage trust beyond these affected series. We'll see.

It is not that long:

1994 - Pentium FDIV
2017 - Atom Clock Signal Issue
2018 - Meltdown and Spectre

1994 was 30 years ago. And while it didn't look good, it wasn't that bad in practice. It was easy to work around, and FP division was not really a performance bottleneck for any of the common uses for this chip.

That's only a partial list, though. There were two generations of chipset in the 1990s that ran Windows 95 just fine, but crashed very quickly running Wiindows NT or SCO Unix, because their more complex access patterns showed up bugs that Windows 95 didn't. There was the SATA3 ports failing after a new months problem about a decade ago. If I spend a few minutes I can probably think of some more.

[wraper]

^AMD says it's packaging, not silicon issue.

Packaging will not be the matter for elaborating the "quality problem" with mobo vendors, I would say. Also it is perhaps not the silicon issue, rather the similar annoyance as the Intel is facing..

Do you understand what chip packaging means? It's not about cardboard box.

I think I do understand the meaning of the chip packaging..
As the bloggers report the "AMD is cooperating with mobo (aka motherboard) vendors" to resolve the issue.. So it is not about the chip packaging but the microcode..

PS: Moreover, you cannot solve a "chip packaging issue" 25 days prior the release, no way..

AMD is simply setting up the new more conservative values in their microcode engines for the temperature vs core(s) frequency vs V/I/P settings, as they know the Intel will do the same (Intel wrote that) till the mid of August this year (the as of today by AMD indicated new release date of their new Ryzen 9000 series).

AMD pulls back already supplied chips and ships chips with later production date instead (what mobo vendors have to do with replacing the chips?  ). Does not look like they're fixing the design, rather looks like it could be a bad batch. Nor did AMD say anything about motherboard vendors at all. In other words you spread FUD.

We appreciate the excitement around Ryzen 9000 series processors. During final checks, we found the initial production units that were shipped to our channel partners did not meet our full quality expectations. Out of an abundance of caution and to maintain the highest quality experiences for every Ryzen user, we are working with our channel partners to replace the initial production units with fresh units. As a result, there will be a short delay in retail availability. The Ryzen 7 9700X and Ryzen 5 9600X processors will now go on sale on August 8th, and the Ryzen 9 9950X and Ryzen 9 9900X processors will go on-sale on August 15th. Apologies for the delay. We pride ourselves in providing a high quality experience for every Ryzen user, and we look forward to our fans having a great experience with the new Ryzen 9000 series.
-AMD SVP and GM of Computing and Graphics, Jack Huynh

[amyk]

I suspect it's electromigration, which is mainly a voltage-dependent effect.

Electromigration depends on current density and temperature.

There are lots of things which could be causing it, including gate oxide failure like with the Atom failures, which would be voltage dependent.

The odd part is that so many parts are failing in the same but very limited way.  For instance the insufficient VRAM errors came about because an old and very well tested texture decompression routine was returning bad but consistent data.

Could be lack of timing margin in one area due to process variations combined with  design that wasn't fully verified at the limits, sort of like the 32-bit multiply bug on the early revisions of the 386:

https://groups.google.com/g/comp.arch/c/qm7FVR_YV4A/m/VKR47yLoUdEJ

[Monkeh]

-  AMD delays its Ryzen 9000 launch due to unspecified quality issue

Performance marginality again?

[Bicurico]

My current (company paid) Computer has an Intel Core i9 13900K.

During working hours it is used for CAD/CAM/CAE application.

While the computer is really fast (best benchmark scores in our company), I did have an issue with one simulation software, which aborted on certain tasks. This could be reproduced and did not happen with other computers I have.

To make a long story short: after much research I found that it had to do with this issue. After setting BIOS settings according to instructions, no more crashes of that software. Note that it was just the software that crashed (got terminated), not the OS.

Meanwhile Asus launched a new BIOS, which I flashed. It includes the lates Intel fixes and settings.

I do worry about longevity of my CPU, but other than that, the computer never crashed again. I run it 24/7, by the way.

I always preferred Intel CPU's over AMD ones, after realising many years ago, that some CAD/CAM/CAE software is as not specified for AMD and would crash more often with them.

Hopefully Intel gets it's act together...

[iMo]

Is this guy reading this thread??  

I wonder how the Voltage Regulation is actually made, and, especially how the A, V, P is actually measured (in detail) with those CPUs..
The bloggers shoot tons of information on those CPU's power consumption in Ampers (like 1-350A range), voltages (like 0.03 to 1.6V range), and powers (like 1-300W range).

Imagine, say, a 50-100mVpp noise ripple, an 8bit ADCs perhaps - how they know what is the actual voltage, current and power the CPU is getting?? Especially when the CPUs changes their power consumption in microseconds??
Would be great to know more on how the stuff works there..

..
For example you may set the microcode such their CPUs report lower values, while in reality it is all over the edge and the chips will die after a couple of months. None blogger is able to measure the real values, btw., they just comment what they see in HWinfo or AIDA, etc..
..

[madires]

The latest:
There is no fix for Intel’s crashing 13th and 14th Gen CPUs — any damage is permanent (https://www.theverge.com/2024/7/26/24206529/intel-13th-14th-gen-crashing-instability-cpu-voltage-q-a)

[radiolistener]

limited lifetime by design allows to get more money, because these who bought that processors will need to buy a new one very soon... Just a business..

[brucehoult]

limited lifetime by design allows to get more money, because these who bought that processors will need to buy a new one very soon... Just a business..

Damn, Apple is soooo stupid.

Every single Macintosh computer I have from the 80s, 90s, or 00s that I didn't sell when I upgraded (or that I bought used cheap later when they were obsolete) still works.

I expect they'll be going out of business soon, with bad planning like that.

[DiTBho]

Damn, Apple is soooo stupid.
...
I expect they'll be going out of business soon, with bad planning like that.

Have you ever been to The Steve Jobs Theater, located at Apple Park in Cupertino, California?

That auditorium ***can seat 1000 people*** and its architecture is an awesome outstanding of "we can do this, because we have the money", including a glass elevator that is 13m high and rotates 171 degrees.

As much as I hate their strategy as a user because Apple forces me to change computers every 4 years only to software planned obsolescence, I have to admit that from a marketing point of view it is very effective, in fact Apple is one of the companies that since 1984 has been making a lot of money ...

... the facts speak for themselves, like that auditorium, and I'll tell you more, Apple can afford to throw millions of euros into the (secret) project of a phantom "Apple Car", and then say "well, nothing's coming, project deleted"

2000 engineers can go home, and they will be reinstated in other roles in the company ...

if Apple was less strong economically, they would have told them "you are all fired, go home, and find another desk" ...
Like SONY did ...

[DiTBho]

Every single Macintosh computer I have from the 80s, 90s, or 00s that I didn't sell when I upgraded (or that I bought used cheap later when they were obsolete) still works.

my PowerBookG3 ...

... back to the early 2000s, Apple had established that you could install MacOS from v10.1 to 10.3 but not 10.4!
And this was not a limitation of the hardware or anything.

I asked a friend to lend me the PowerBook G4, with which I installed MacOS 10.4 on a 2.5" pATA disk connected to the USB port.
Then I mounted that disk inside the PowerBookG3 and ... to my great surprise, MacOS 10.4 worked very well on a laptop for which the installer had blocked saying "your hardware is too old, I cannot continue with the installation"

what? 

The exact same thing happens with Apple Silicon And even worse.
It happens with programs like FinalCut, which strictly need a version of MacOS ...
The problem is ... with the new Apple { MacMini, MacStudio, ... } you cannot remove the SSD and use the trick I described above.
The mobo is tied to the serial number of an SSD ...
So if you work professionally and need to work with updated versions of sw for MacOS, expect to have to replace your Mac every 4 years.

[brucehoult]

Every single Macintosh computer I have from the 80s, 90s, or 00s that I didn't sell when I upgraded (or that I bought used cheap later when they were obsolete) still works.

my PowerBookG3 ...

Of course I was talking about "physically works, with the original supported software"

... back to the early 2000s, Apple had established that you could install MacOS from v10.1 to 10.3 but not 10.4!
And this was not a limitation of the hardware or anything.

I asked a friend to lend me the PowerBook G4, with which I installed MacOS 10.4 on a 2.5" pATA disk connected to the USB port.
Then I mounted that disk inside the PowerBookG3 and ... to my great surprise, MacOS 10.4 worked very well on a laptop for which the installer had blocked saying "your hardware is too old, I cannot continue with the installation"

These soft restrictions are not because the hardware physically can't run the new OS, but because it's slow and has little RAM and is missing things that the new OS wasn't to take advantage of such as SIMD, Firewire, or better GPUs.  In the case of 10.4 Lion Apple drew the line at 400 and 500 MHz G3 Pismos, though it probably worked much better on a G4.

It's not "it won't work" but "you probably won't have a good experience".

And yes you can install both 10.3 and 10.4 on unsupported G3s and probably even 604e and so forth by using the "install it on a newer machine" or more simply by using XPostFacto to trick the installer.

My G3 PowerBook is a PDQ 266 MHz that originally came with 64 MB RAM. I think it has 192 MB now. It's running OS X 10.2.8 but quite honestly that runs badly enough that I'm tempted to revert it to OS 9.

[iMo]

The latest:
There is no fix for Intel’s crashing 13th and 14th Gen CPUs — any damage is permanent (https://www.theverge.com/2024/7/26/24206529/intel-13th-14th-gen-crashing-instability-cpu-voltage-q-a)

Who will get the Pulitzer for this Intelgate story? 

[DiTBho]

@brucehoult
that dude may have the answer

[DiTBho]

The latest:
There is no fix for Intel’s crashing 13th and 14th Gen CPUs — any damage is permanent

Who will get the Pulitzer for this Intelgate story? 

that's why my boss is very worried ... 

[BrokenYugo]

limited lifetime by design allows to get more money, because these who bought that processors will need to buy a new one very soon... Just a business..

They're not that smart, it's just shortsighted greed and bad management. No conspiracy is required.

[coppice]

limited lifetime by design allows to get more money, because these who bought that processors will need to buy a new one very soon... Just a business..

This is paranoid nonsense. Everyone designs around a lifetime goal, but you will usually find its longer than most people expect from negative comments like this one. Most of the time engineers are struggling to keep the lifetime up to a reasonable figure, while keeping costs under control. When things fail really fast is usually bad engineering, or rushed engineering, or things being used in ways the engineers didn't anticipate.

In the server chip world they still seem to be using more conservative design rules for CPUs than they do for comparable consumer parts. This is a tradeoff. The conservative parts have a longer expected life, running hard 24 hours a day, while consumer parts from the same family usually run up to higher clock rates. This is regardless of the number of cores packing into a single package. The current Intel problem seems to be with consumer parts in hard working servers. There have been some issues with this in the past, although I can't remember specific cases off the top of my head. Have there been any issues with the most recent Xeons?

[radiolistener]

This is paranoid nonsense.

No. All modern electronics lifetime equals to guarantee period. So you can expect that it will be broken just after guarantee period expiration. And it really fails just when guarantee period expires. Old electronics usually has much more lifetime and worked much longer than guarantee period. This is because there was no technology to precisely predict lifetime for a short period. Modern technology allows to use materials with predictable lifetime.

For example, modern mixer uses plastic gears inside which wear out and fails after 1-2 years (guarantee period). Old mixers were made with metal gears and can works for years with no fail. This is how programmable lifetime works...