Processors
Intel® Processors, Tools, and Utilities
16047 Discussions

What is actually wrong with Raptor Lake?

NubMan55
Novice
34,512 Views
I have a 14900k that keeps increasing in instability. I have posted here about the issues before. High-end Raptor Lake chips seem to be breaking down on the daily, even CPU's used in server environments seem to have a failure rate of ~25%. Intel as a company is not giving us, the consumers, any updates into their investigation. I have been offered a replacement CPU/refund by you guys before. I do not believe that is a good deal for me since as of right now it seems if I replace the CPU, I'll be needing a new one in a few months. If I get a refund, I'll be stuck with an expensive motherboard that only works with your CPU's that seem to turn to sand in a matter of months. I'd like to receive some form of an update as I'm getting fairly frustrated.

Please do not copy paste the same "Intel is aware of the problems experienced in certain workloads and is working with blah blah". I require a proper answer. If the root cause is truly unknown, so be it. It seems Intel as a company have weighed the cost of silence vs doing the right thing and have decided to stay silent until next gen release and hope customers forget and suck up the cost.
0 Kudos
21 Replies
Eisbar
New Contributor I
11,789 Views

@NubMan55 


As others have said, including Intel in their roundabout way. The problem with Raptor Lake is voltages and frequencies have been pushed beyond reasonable limits for Intels one-hit-wonder 10nm process.  

 

There's some history here that's important that gives a better picture.

 

In 2012, Intels director of Process Architecture Mark Bohr made a bold claim at Intels Developer Forum that Intel was going to have their 10nm node in production in 2015.

""The 14nm technology is in full development mode now and on track for full production readiness at the end of next year," Bohr said."

"While 10nm processors, code-named Skymont, are on tap for 2015, 7nm and 5nm architectures are also in the pipeline beyond that," Bohr said.

So two important things here. Bohr had stated 14nm would be at full production readiness by Q4 2013, and Intel 10nm would be ready by 2015.

In September of 2013 Intel revealed a notebook that used their new Broadwell CPU, built on 14nm and in this use case demonstrated on a low power device. Intels CEO at the time Brian Krzanich stated that 14nm Broadwell would be shipping by the end of 2013

That did not happen. 

It is now August 2014. Intel announces their Core M line of CPUs, their first product to be build on their 14nm process they say. The first Core M products become available at the end of 2014.

It is 2015, 14nm is only just becoming widely available. Intels promised 10nm from 2012 is nowhere to be found. The 14nm process bring forth Skylake in Q4 2015. This is Intel 6th Gen.

It is 2018, Intel has still been unable to deliver 10nm. They launch and sweep Cannon Lake under the rug. Cannon Lake was due in 2015, Intel announces to their investors they are shipping 10nm now in low volume and expect things to turn around in 2019. 

Cannon Lake is a disaster of a product and it wouldn't surprise me if you never heard of it. The CPUs were only able to made for low power applications, the onboard GPUs were defective and had to be disabled, etc.  Cannon Lake yielded exactly one CPU model, that's all. It was axed by Intel in Q4 2019. Cannon Lake gave us the Palm Cove cores.

 

Intel has been unable to innovate and get their 10nm process off the ground. 

Intel 10th Generation is announced in 2019 and arrives in 2020, this is Comet Lake, built on the Intel 14nm process and is the third revision, or reiteration, or broken record if you will. The 14nm 10900K is released in the Summer of 2020.

Jim Keller resigns from Intel and it basically boils down to Intel refuses to outsource their fab.  This was a big deal.

 

It is now 2021. Intel 11th gen Rocket Lake which was supposed to be built on the 10nm process has been backported to 14nm, Intels 10nm is still unable to perform in anything but a low power capacity. Backporting came with consequences and again, Intel has swept things under the rug. 

Intel 11th generation was catastrophically defective but people seem to forget that.

Screenshot 2024-08-28 083356.png

 

Pugets data is unintentionally biased because of the workloads that their customers will most likely be doing and the problems of Raptor Lake will be less experienced due to the amount of baked in error handling in content creation software and I can elaborate more on this if need be.

Alder Lake has arrived. Success has now been achieved on the Intel 10nm node that they've rebranded as Intel 7 because people have short attention spans and it's best to capitalize on that I guess. 

Alder Lake is the absolute limit of what the 10nm can provide. The cap is around a 12900K, I say that because Intel silently axed the 12900KS only recently when the media began to pick this story up. 

A node that Bohr promised in 2012 to be ready in 2015 is seven years late.

So Intel had a choice. 

Tell the investors that a node that took seven years to deliver will be a one hit wonder or hope that you can kick the can down the road long enough for people to buy the idea that you're on track with yet another bold claim. To admit Alder Lake was the maximum that could be squeezed from such a resource sink that had gone on for so long would have leveled Intel. I have no doubt in my mind about that.

We now know that their node roadmap is again, gone to pot. Lunar Lake is pure TSMC, Arrow Lake is allegedly pure TSMC on mid-to-high power SKUS.  Lip-Bu Tan has resigned from Intel over disagreements with the management of the company. Tan is one of the most respected people in semi from what I know and I think it's saying something to the world that he has decided Intel is not for him.

This is a red flag just as Jim Kellers departure was.

Just circling around, the 12900KS SKU was axed, but not a large portion of Alder Lake SKUs, Intel 13th Gen SKUs were axed as well, but seemingly only the K SKUs. Actions again speak louder than words.

12900K had a max boost clock of 5.2GHZ and its fused VID was on average eh 1.33v for this. 

12900KS had a max boost clock of 5.5GHZ and its fused VID was on average ~1.40-45v

In Intels passthrough QA document for handling customers questions Thomas Hannaford says some important things.

 

Q: Why aren’t we seeing this issue on prior gen unlocked desktop processors?
A: Based on Intel’s analysis to date, Intel Core 12th Gen desktop processors are not at risk due to
lower voltages and turbo frequencies compared to Intel Core 13th and 14th Gen desktop
processors.

Q: Is Intel declaring elevated voltages to be root cause of the instability issue?
A: Incorrect voltages are one aspect of Vmin Shift Instability issues. Intel has delivered a microcode
patch (0x129) as a partial mitigation addressing exposure to elevated voltages which is a key
element of the Vmin Shift Instability issue.
To date, three mitigations have been identified related to this issue:

 

Mitigations are not solutions, they are meant to lessen the impact or severity by definition. That wording is chosen extremely carefully.

The problem is the processors themselves and what Intel sold them as being capable of. They fail in a very specific way and have been doing so since 2022, I've spoken about this before and how specific cores fail in very specific way. IE: P-Core 5 and 6 will most likely fail at the highest rates declining to 0% with Cores 0 and 1 due to their location on the die. This can be demonstrated. 

 

In the end I don't think a recall will solve anything at this point, it should have happened when Intel became aware of these failures from the beginning back in 2022, but it didn't. I think Intels actions have landed it where it is today and my sympathy goes out to the employees who had no say in what took place but are taking the brunt of it, either being let go from their jobs to cut costs due to managements decisions or the customer service people who are probably being worked to death right now and dealing with some dorks I imagine. 

My hope is that Intel does right by their customers, each and every one of them. The customers are the unwitting shoulders that took the burden to keep investors happy for the past two years. 

Pat Gelsinger likes to keep talking about node leadership as a goal, it's good to have goals. I don't think having a more advanced node than any others under the circumstance is anything I would ever attach the word leader to though. I think that branching out to seek a leadership position in ethics and morality would make a fun little side project for Intel management. 

 

 

 

 

 

 

 

 

 

 

 

0 Kudos
Reply