Return to

Intel ChipShortages To Continue Through 2020 - Are intel just up to shenanigans at this stage?

Just reading through yet another intel chip shortage article and I am kind of wondering if they are actually under that much pressure or are they forcing the market to absorb their back log.

Based on demand, we are expecting supply will remain constrained through 2020. Server platforms which use these processors are affected. In order to minimise customer impact as a result of these supply constraints with Cascade Lake processors, HPE urges customers to consider alternative processors, which are still available. We are in constant dialogue with our partners at Intel and have a strong relationship with them, and we know they are working on the issue.

Emphasis mine.

HPE is pushing people to older Skylake Xeons, Cascade Lake is an evolution of Skylake, so I am wondering if this so deliberate to clear the back log of still high margin 2017 Skylake Xeons. After all they will probably sit if there is a better chip and apparently higher demand for it.

I also wonder if the servers could take a drop in CPU upgrade? Sell them the racks of old CPUs now and then sell then another CPU in a few months time when they are available. And maybe another jump to the 10nm if that is ready.

But I also wonder at what pace enterprise harware takes to spin a new platform and if HPE, Lenovo and specifically Dell will put up with it. I say Dell because they have had $1bn wiped off their estimates for the year in part due to this shortage. So would it be more worth their time to spin a new AMD platform and offer that in place of Intel just so they can continue to actually sell hardware at a reasonable rate.

This has gone in since summer '18 and looks like it will be 2021 before they are back to anything near reasonable pace to keep up. And that is just for 14nm parts, not the 10nm because those effectively don’t exist right now. It is a comedy of errors and confidence beyond reality.

We continue to focus on improving supply for our customers. We have invested record levels of CapEx the last two years to expand our capacity and increase our supply, but customer demand has outpaced these improvements. We continue to prioritise the production of our highest performance processors as we also continue ramping volume on 10nm and working closely with customers to align our available supply to their demand.

Emphasis mine.

That sounds like a bad idea at this point in time. Spending record levels on bringing back 14nm when world +dog has been shown 7nm parts handily beating out now 3 year old 14nm parts. Then there is the no doubt collosal amount spent on 10nm that has yet to be seen to do much, yes there are some laptops but their money is made in servers, to the point that they are holding back laptop and desktop chips to help get server stuff out the door.

So who is snapping up all these Intel server chips? The hyperscalers. Intel cannot make enough processors fast enough for Amazon, Microsoft, Google, Facebook, and others of similar size that are constantly expanding their data center operations.

Those have historically need happy to use whatever is available fast and efficient all things that AMD and ohers are doing well at while also being available now.

Hungry for more, the hyperscalers are also looking to other CPU architectures. Amazon, Google, Microsoft, and others, continue to make noises about using AMD, Arm, Power, and RISC-V designs in their operations.

Like that.

All this is just a layman reading news badly, so I am sure there will be corrections and other views.


Semi Accurate reads intel the same:


echo “you mean limiting the supply of ${diamonds} will increase the profits we can make?”

1 Like

And that is sort of what I was thinking but they have competition now, so sure they can try but it looks really stupid now, like everyone can see it is a bad idea but them.

There are legit problems but they also seem to be playing some game that does not yet make sense.


Their 14nm fabs are running full steam, and it’s well documented that the reason for that is because of smeltdown massively drove up demand for replacement hardware.

Everyone wants cascade lake, but Intel is unable to deliver that to everyone in a timely manner, so the other option is to go skylake.


I’d say intel are under real pressure.

Basically they were meant to be on 10nm years ago, and i believe (read somewhere) that they never expected to be doing more than 4c/8t for consumer on 10nm (and thats why we got 4 cores for mainstream for so long until AMD forced their hand - 10nm wasn’t ready and they didn’t want to go up in count until then). Probably never expected to go beyond 8c/16t on 14nm for server either.

But, because 10nm is so broken, they are having to make much, much bigger core count chips on 14nm than they expected.

The fact that AMD hit it out of the park with Ryzen, AND TSMC are full speed ahead on 7nm have caused double damage to their manufacturing.

  • More cores = bigger dies = less dies per wafer.
  • Bigger dies ALSO = lower yields (due to larger chance of each die having a defect due to increased area) = even less dies per wafer

10nm, if it worked would double their fab capacity for a given transistor count, and that’s where they expected to be, some years ago. Certainly expected to be shipping 10nm by 2019 (originally 2015-2016).

Meltdown and spectre, etc. also happened which increased requirements from their enterprise customers, particularly on the high end parts. Because high end parts are (unlike AMD’s chiplets) a totally different design to their consumer parts… intel have a choice to make - they either build more consumer parts (where they are already squeezed) or try and build more high end server/workstation parts to try and protect their profitable segment of the market.

They’re choosing the latter. But these parts have a low yield rate already… trying to make them even bigger to compete with EPYC/Threadripper has gotta hurt…

This is a perfect storm for intel. It’s a good thing they have a shit load of cash in the bank, because they’re going to need it. They’re only going to be able to compete in the low-mid range performance segment of the market against AMD until at least 2022, but the parts are going to cost them way more than AMD to make for the same core count. If they compete. Lose money on parts when they can barely supply anyway? Doesn’t sound like a win to me?

This is why they’ve even started putting out some old 22nm parts again for the low end, and moved some chipset production to other foundries - to free up 14nm+++(+?) for datacentre/high end.

In datacentre there’s a lot more inertia, so there, they maybe have a chance to try and maintain margins somewhat. But not for too long… so they’ll be pulling every proprietary lever (and benchmark fraud, OEM bribery, etc.) in the book to try and provide some non-cross-platform hook to keep people on intel there (see: thunderbolt, optane, new AI instructions, etc.).

I think intel are in real trouble here, and they really, really need a win. They need to get a working node shrink because right now they’re screwed without it both on price and ability to supply their customer base. they did have a record year in 2019 but the full force of EPYC hadn’t properly hit yet.

And in 2020, AMDs core counts just doubled again (plus, they’re now roughly equal on throughput per core, on enough stuff that matters, too). Intel simply don’t have the manufacturing capacity to compete with that on 14nm. It would mean they need to halve (or less due to increased failure rate) their supply. Or double their fab capacity. But that’s a multi-year lead time…


Oh. I forgot.

That’s if intel can even cool the parts they’d have to try and make to compete while staying on 14nm and double core count.

Intel are already talking liquid cooling in datacentre for 48 core parts, and power draw above 300 watts. AMD are well under that with 64 cores, never mind what they bring with Zen3 based EPYC parts later this year/early next.

Double core count again? Even same core count with another 10% IPC to find to stay where they are vs. Zen3? Tough break

1 Like

Yeah hard times indeed. I just keep getting the feeling there is something going on here though, they keep fucking up and getting rewarded for it?

People are still happy to buy power hungry, hot as balls, old as fuck and with so many holes in it that it makes swiss cheese blush, CPUs?

I can see the cracks growing but if this were anyone else I feel they would be long ago filing for bankruptcy.

There are a few things holding enterprise back from going to AMD. I’ll list a few that directly impact either myself or scenarios I can easily see:

  • AMD are currently supply constrained as Apple and the other mobile vendors are consuming a lot of TSMC’s 7nm capacity. That changes when TSMC ramp up 5nm and the high yielding on early process mobile chips go to 5nm. Basically AMD are getting all of Apple’s current 7nm volume when that happens i believe. This would be why the HPEs, Dells, and the like haven’t ordered a bunch of AMD yet. Because even though intel have supply issues, so will AMD, right at this moment.
  • If you have a datacentre with a bunch of intel hardware in it, AMD and intel virtualisation instructions are not compatible. On intel, even with a new arch you can use e.g., VMware EVC for vmotion to live migrate between old intel CPU based hosts and new ones without downtime. If you’re planning an upgrade (or rather expanding capacity within a cluster) then changing CPU vendor is going to be a pain…
  • Intel do have some advantages in niche workloads and major server vendors are still not yet fully on board with EPYC (see the first point)
  • EPYC maybe still needs to be certified by some software workloads.
  • IT management (i.e., many of those making the purchasing calls) are generally older, and remember various AMD/non-intel chipset vs. software issues from previous generations back in the 90s/00s/10s. who’s fault it was (intel shenanigans or not) doesn’t matter business wise - what matters is actual reliability - and there is some resistance there to make the jump because of this. I’ve been suggesting EPYCs at work for a while now, for next hardware refresh (due this year or next at a stretch). Initially i was rebuffed with the above sort of comments. However even my boss is now seeing a lot of intel problems in the news and records being broken by EPYC at much lower price. So that will change.

But none of these factors are going to be the case forever. Things are only going to get worse for intel in the short term.


That is where my understanding falls apart. I don’t know the big business side of it. Makes sense, I just think on ideal.worlds for a lot of stuff beyond me and the ideal world would be scrambling to ditch Intel so fast considering the potential pitfalls of their current problems.

Thanks for that.

1 Like

Just think of it this way:
You know intel does the job it currently does, you’ve been able to get them when needed for decades. You know that everybody who isn’t an IT nerd “knows” that intel are the biggest and best (whether that is true or not).

EPYC is a newcomer.

Do you put YOUR job on the line, on a 6-7 plus digit PO to buy something that hasn’t been extensively proven in the market?

Its the classic “nobody got fired for buying intel” scenario.

Worst case if you buy intel, and over pay for hardware: Your excuse as a manager is that you were being risk averse and will buy AMD next time because now it is proven. “Understood mate, it was a tough call without the benefit of 20:20 hindsight, i would have done the same!”

Worst case if you buy AMD: you have some sort of issue with migration or unexpected performance - in non-IT or upper management’s eyes (i.e., your boss - the CIO or his boss) you’re the dickhead who took the risk, bought the lesser known part and caused the problem…

Like i said, this situation won’t last forever as even CEOs end up reading about other CEOs like Lisa Su given how well AMD are doing… in fact if AMD stock keeps going the way it is, i think THAT is likely to be as much of a catalyst as anything else for EPYC to really take off.

in smaller, largely tech savvy IT related companies this is less likely to be a problem. but in larger corps who’s core business isn’t IT… good luck!

1 Like

So news this morning.

Intel was going to put PCIe4 into Comet lake, but pulled it quite late.

So you’re going to have expensive boards built for PCIe4 signalling, without CPU/chipset capable of PCIe4. And you can’t just put them in an older board because the socket is different. Derp.


if AMD were dicks, now would be the time to add some proprietary, faster than PCIe4 interface to Zen3. Like intel tried to do with Nvidia. And intel did to AMD with thunderbolt.


Oh wow. So the boards are capable but the chips are not? So this another technical problem with the CPU design. Yikes, and as they point out the boards already likely have the parts so the expense is already there even though it cant be used, and likely will go completely unused if intel do the usual and not let you use the same board for the next line of chips. Also those power requirements? Up to 250W out of the box for the boost speeds, and then leaving it up to vendors if they want to limit to boost time so they can use cheaper parts, which then muddies the waters as to whether your chip will work as advertised or not. WOW!

The hits keep on comin’.



“We are as fast as AMD on single thread! With our CPU rated at 125w TDP!” **

** when you force 350-400 watts through the core. Select motherboards only. YMMV, etc.


All on top of being late and on the wrong node.

Just as a reminder:

1 Like

Anything from Charlie Demerjian regarding Intel can’t be taken seriously.

How about Ian Cutress then?

I would say those two articles are on different things.

But for the record, comet lake is absolutely better than ice-lake in it’s current state.


Has he said something that is factually incorrect?

Intel’s compiler shenanigans are well documented. If you don’t believe charlie…

1 Like