Intel: Alder Lake permanently takes client lead

https://wccftech.com/intel-ceo-on-alder-lake-cpus-amd-is-in-…

The CEO made the remarks in a video wishing everyone a ‘Happy 2022’ and talked about some of the achievements that his company had made ever since he joined Chipzilla. Pat joined Intel back in January 2021 and it has been a year over at Intel. During this period, the blue team launched several key product lineups with the most prominent being Alder Lake CPUs.

“Alder Lake. All of a sudden…Boom! We are back in the game,” exclaims the impish tech CEO. “AMD in the rearview mirror in clients [consumer market],” he adds, “and never again will they be in the windshield; we are just leading the market.”

Intel CEO, Pat Gelsinger on LinkedIn via Tomshardware

As per Pat Gelsinger, Alder Lake CPUs put Intel right back in the game and he further states that after the launch of 12th Gen CPUs, AMD is in the rearview mirror in clients, talking about the consumer segment here. He further goes on to state that Intel will never again let AMD be in the windshield and moving forward, you can expect the blue team to lead the market.

1 Like

I dunno… I see the client market as pretty much a tie, or so close it doesn’t matter. OTOH, AMD clearly leads in the segments of the server market where they have products.
Alan

3 Likes

I will have to see more than Intel’s claims to believe they have passed AMD on the client side. Competitive on some metrics, yes, but not better. Since AMD is supply constrained, they really have no reason to push the high volume low margin client market hard. The key thing for AMD now is to manage its wafer contracts, and they extended the Global Foundries contract not that long ago. AMD could start making its controller chips at TSMC and gain some performance, so IMO the only reason to continue using GF (aside from stock considerations) is that GF’s wafers allow AMD to produce more CPUs. When Intel passes AMD in servers, then I think AMD could switch some capacity around to retake clients. It is better for AMD not to see people waiting months to get their new client CPU so reducing demand until they get more (a lot more) wafers seems wise. I suspect if GF has capacity AMD could be pushing into low margin GF produced CPUs for tablets, Chromebooks, phone.

BTW, I needed one more test to get onto the active kidney transplant list. The echocardiogram machine they used made the last echo I had 20 years ago seem like comparing talkies+color movies to the silent movie era. As a gamer, I could see all the modern graphics they were using to create significantly better testing. The pixel count was way higher, the colors and speed of displays to model blood flow reminded me of a WOW raid. and they split the screen into four at various times. That specialty market is going to want chips, and AMD has been in those markets for a long time. Fortunately for me, the test showed normal heart function. So I expect to get a new kidney in 2024-25 in Phoenix or Tampa Bay (I have the wrong blood time to get it at UCLA) and then, Covid willing, we can add travel back into our retirement.

4 Likes

Congrats on the positive health news for the new year!

I will have to see more than Intel’s claims to believe they have passed AMD on the client side.
You can look at pretty much any of the reviews for the desktop alder lake and see that in most benchmarks it outperforms the top of the line Ryzen.
Alan

Since AMD is supply constrained, they really have no reason to push the high volume low margin client market hard. The key thing for AMD now is to manage its wafer contracts, and they extended the Global Foundries contract not that long ago. AMD could start making its controller chips at TSMC and gain some performance, so IMO the only reason to continue using GF (aside from stock considerations) is that GF’s wafers allow AMD to produce more CPUs. When Intel passes AMD in servers, then I think AMD could switch some capacity around to retake clients.

  1. Totally agree on “no reason to push clients right now.” Let Intel have it, as long as AMD can carve out and hold a healthy chunk of server revenue, have pricing power etc…
  2. The GloFo contract extension could be useful for any number of things… assuming Xilinx lands, maybe even some of that work goes over there. I’m assuming GloFo can continue to make plenty of “useful stuff” even if none of it is CPUs.
  3. When Intel passes AMD in servers… it’s probably time to go. The client business is not really worth fighting for, in my opinion.
  4. Will Intel manage to render AMD irrelevant in GPU/HPC/AI/ML etc.? NVidia is still the “boss battle” in those areas.
1 Like

You can look at pretty much any of the reviews for the desktop alder lake and see that in most benchmarks it outperforms the top of the line Ryzen.

But didn’t you say just a post ago, that it was pretty much a tie?

1 Like

But didn’t you say just a post ago, that it was pretty much a tie?
Yep. I consider +/- 5% or so a tie. It also depends on which benchmarks you look at. There is not enough difference to make a substantial difference, especially considering very few buy the top of the line product. OTOH, when you look at the reviews they declare Intel the winner.

When somebodies top tier product is near the middle of the competitors product stack then you have a problem. Both AMD and Intel have been in that position from time to time.
Alan

Yep. I consider +/- 5% or so a tie. It also depends on which benchmarks you look at. There is not enough difference to make a substantial difference, especially considering very few buy the top of the line product. OTOH, when you look at the reviews they declare Intel the winner.

Ok, got it.

When somebodies top tier product is near the middle of the competitors product stack then you have a problem. Both AMD and Intel have been in that position from time to time.

So, just so I understand, who’s in this unenviable position now, AMD or Intel? I guess Intel?

So, just so I understand, who’s in this unenviable position now, AMD or Intel? I guess Intel?
Intel in the server market until Sapphire Rapids is released.
Alan

Should we be ignoring Threadrippers, they are in the client sphere imho.

Should we be ignoring Threadrippers, they are in the client sphere imho.
Threadripper would win in heavily multithreaded applications. Intel would still win in gaming.
Isn’t threadripper more of a workstation CPU? Intel sells their workstation CPU’s under the Xeon brand for now.
Alan

1 Like

Isn’t threadripper more of a workstation CPU? Intel sells their workstation CPU’s under the Xeon brand for now.
I suppose if Intel doesn’t count workstation as a “client”, then perhaps we shouldn’t count threadrippers either. I was thinking of client as meaning it’s somebody’s primary PC sitting on a desk or under it. Gaming performance is useful for some but not AMD’s main target to date.

1 Like

Intel in the server market until Sapphire Rapids is released.

Except that it is probable that Zen 4 EPYC chips will start shipping first. But there is an elephant in the room, DDR5. Yes, people can buy DDR4 motherboards for Intel’s 12th gen CPUs, but that is sort of instant obsolescence. Will DDR5 catch up to the demand? Probably not this year. I expect Zen 4 for desktop to be held back until early 2023. Zen 4 for servers will act like a vacuum cleaner for DDR5–Sapphire Rapids could do the same, but in either case, figure that server customers will pay what they must to get the newest CPUs, and desktop consumers will have to suck up high prices. AMD might announce a “mid-life kicker” for Zen 3 or they may just decide that the R9 5800X3D is all they need. For that matter they could wake up (Zen 3+) Warhol if DDR5 memory stays tight.*

Hmmm. Two parts to the continuing DDR5 shortage. The actual memory chips are not a problem as such. But two sticks of DDR5 will be 32 Gig of memory. I have 16 Gig of DDR4 3200. I run a mining app in a VM client and trust Process Lasso to keep my foreground processes from being affected. I tend to average about 10 Gig of (DDR4) memory demand. So why would any (normal) user want 32 Gigs of DDR5? To get two DIMMs, but they will also get twice the DRAM. Will some memory makers start making 8 Gig DDR5 DIMMs? I’m sure they can figure out how to do it. Why not now? Because it is a ‘happy time’ for memory makers, but the parts required for DDR5 DIMMs but not DDR4 DIMMs are in short supply. So they are all going into 16 Gig DIMMs.

  • 2Q2022 looks like a good time to buy DDR5 to sell later on E-bay. I won’t do that–too much like work–but it looks like a significant dip before Genoa starts hoovering up the surplus. For all I know, AMD is already shipping Genoa chips to hyperscalers. If not, the issue is SP5 motherboards and BIOSes. Expect ‘sampling’ now, and ramping up hyperscaler shipments until AMD starts holding back some chips for the official Genoa product launch, which I expect in 3Q2022.
2 Likes

Except that it is probable that Zen 4 EPYC chips will start shipping first.
The last word from INTC is Sapphire Rapids ships late Q1, while AMD says H2 for Genoa. OTOH there are rumors that Sapphire Rapids has slipped to Q3. We should hear something from Intel about this at earnings. Since AMD is still talking about Milan-X I don’t see any upside in the Genoa schedule.
Alan

AMD is still talking about Milan-X I don’t see any upside in the Genoa schedule.

AMD, and to a lesser extent Intel, have always been willing to ship “risk” chips to customers who understand what exactly the risk is. Recently, on the AMD side anyway, the risk has been with the BIOS, not any serious chip errata.

You have to remember that my experience was in a bleeding-edge Air Force lab where we got ‘EC’ engineering samples up to two years before that chip became a product.

Let me describe the situation back then. There was the SIPRnet. (Stands for Secret Internet Protocol Routed network. All of the SIPRnet nodes were behind encryption devices that used random numbers thousands of digits long to encode the traffic. Many allies have SIPRnet access. If a document or a directory is marked as NOFORN, they just don’t see it.

Eventually, there was a TS (Top Secret) net on top of SIPRnet, but that is sort of irrelevant. However, GCSS running on top of SIPRnet, managed all logistics and personnel movements. Want to send a division to Iraq? Invasion or friendly move? In either case, GCSS would take a large movement and break it down into “frag” orders, which might move one plane, one tank, or even one individual. The important part of the system was that it made sure that there was enough fuel for that tank at that location, or to refuel that airplane, etc.

The servers were all Sun Microsystems machines. But obviously, there was a need for thousands of PCs to serve as clients so those frag orders could get quickly printed and distributed.

I didn’t do much on the PC side–a lab with a dozen or more Sun machines running Solaris kept me busy. But on either side of the lab, we needed to get software and hardware early so we could have a secure version of Windows on that new hardware when the new hardware went out to Air Force bases around the globe. We supported the Air Force, I assume there were similar facilities for the Army and Navy. Marines, Coast Guard, and others? I have no clue–but we supported the SOC.

So now you know why I’m used to having hardware in hand up way before it was announced.

Oops! …hand up way… should be …hand, way…

until Sapphire Rapids is released…

Except that it is probable that Zen 4 EPYC chips will start shipping first.
Intel confirmed during the earnings CC that Sapphire Rapids ships for revenue this quarter (Q1).
Alan

Intel confirmed during the earnings CC that Sapphire Rapids ships for revenue this quarter (Q1).

But Sapphire Rapids is scheduled for full production in 4Q2022. How to untangle this information? The simplest is that Intel is doing the same thing as AMD, shipping “samples” to large volume customers, except that Intel is putting them in the billing system this quarter. If so, shrug. It has always been the case that big customers get these early chips to test their software (end-users) or their hardware (mainly motherboard makers).

My second guess, and a lot more likely, is that Intel is putting Sapphire Rapids chips into Aurora. And while Aurora is not expected to be completed before 4Q2022, the billing might be going from Intel to HPE. HPE/Cray may not have fully working ROMs, but working well enough to update without needing to be pulled off the board. (The PROMs usually contain errata patches for the CPU chip in addition to initializing the rest of the I/O structure.) Intel wouldn’t need to provide a working BIOS/CPU combo until 4Q. Plenty of time.

Been there, done that, have the t-shirt. :wink: Once the BIOS can support a disk drive and an internet connection, the initial status of the other I/O devices may range from good to horrible. And occasionally you are expected to start your debugging with a motherboard that supports a couple of SATA ports, period.

My third guess, and I hope nowhere near as likely, is that Intel is shipping one low-end Sapphire Rapids SKU while continuing to work toward high yield production. I list this mostly because Intel has done it before. I think the current top leadership is not in favor of such hi-jinks, but someone in marketing may not have gotten the message. Intel could even rename a current generation chip as a Sapphire Lake SKU. This used to be a common practice in the GPU market. It is a lot less common now.

Do I think that AMD is doing the same thing with Genoa and Raphael? I don’t care about the accounting, but I am sure that AMD has EC or even beta chips in the hands of the mobo makers. Two new sockets, with a new (DDR5) memory type, the mobo makers need at least six months for debugging and test, then the OEMs and ODMs need motherboards for their own debugging. Do these things tend to overlap? Sure. But I think that the mobo makers had “working” EC chips last quarter.

But Sapphire Rapids is scheduled for full production in 4Q2022
Source?
Sapphire Rapids samples have been in customer hands for quite some time. They mentioned production shipments to several key customers in Q1.

I think the critical path for Aurora is the N5 TSMC ponte vecchio GPU chips.
Alan