Intel Xeon Server Processor Shipments Fall to a 13-Year Low

[AleksandarK]

Intel's data center business has experienced a lot of decline in recent years. Once the go-to choice for data center buildout, nowadays, Xeon processors have reached a 13-year low. According to SemiAnalysis analyst Sravan Kundojjala on X, the once mighty has fallen to a 13-year low number, less than 50% of its CPU sales in the peak observed in 2021. In a chart that is indexed to 2011 CPU volume, the analysis gathered from server volume and 10K fillings shows the decline that Intel has experienced in recent years. Following the 2021 peak, the volume of shipped CPUs has remained in free fall, reaching less than 50% of its once-dominant position. The main cause for this volume contraction is attributed to Intel's competitors gaining massive traction. AMD, with its EPYC CPUs, has been Intel's primary competitor, pushing the boundaries on CPU core count per socket and performance per watt, all at an attractive price point.

During a recent earnings call, Intel's interim c-CEO leadership admitted that Intel is still behind the competition with regard to performance, even with Granite Rapids and Clearwater Forest, which promised to be their advantage in the data center. "So I think it would not be unfathomable that I would put a data center product outside if that meant that I hit the right product, the right market window as well as the right performance for my customers," said Intel co-CEO Michelle Johnston Holthaus, adding that "Intel Foundry will need to earn my business every day, just as I need to earn the business of my customers." This confirms that the company is now dedicated to restoring its product leadership, even if its internal foundry is not doing okay. It will take some time before Intel CPU volume shipments recover, and with AMD executing well in data center, it is becoming a highly intense battle.



View at TechPowerUp Main Site | Source

 

[Chaitanya]

Its going to get a lot worse for them given they are nowhere near competition at this point.

 

[Tomorrow]

Not to mention server customers stick with what they have for a long time. Meaning once gone it takes a long time to get them back.

 

[rattlehead99]

Not to mention server customers stick with what they have for a long time. Meaning once gone it takes a long time to get them back.

Depends on the demand, sometimes they just add more and more. And the reason intel maintained such high dominance was their sheer production volume against AMD's share of TSMC's production, but now TSMC has a lot more production.

 

[Vayra86]

I'm trying to understand the CEO Lingo here but are they kinda saying that it might not be complete bullshit that they might actually NOT release something every year if the product isn't deemed good enough?

Seeing is believing... I think Intel does need that, a breather. Yes, you won't release something for a year. And then you make something that's competitive again.

 

[ZoneDymo]

Its going to get a lot worse for them given they are nowhere near competition at this point.

^this was a looong time coming and because Intel has nothing to offer in the meantime...yeah rough times ahead.

 

[lexluthermiester]

This is a combination of AMD CPU's and ARM CPU's being utilized in the server market sector. This same thing happened in the mid 2000's with the Opteron's which were very competitive with the Xeon's of the time. Then the Conroe types hit the market and Intel came charging back.

Its going to get a lot worse for them given they are nowhere near competition at this point.

^this was a looong time coming and because Intel has nothing to offer in the meantime...yeah rough times ahead.

Total nonsense. AMD & ARM have eaten into Intel market share, they have not toppled it, nor will they.

 

[Wirko]

Where's that other graph, "number of people fired for buying Intel vs. year"?

 

[Daven]

This is a combination of AMD CPU's and ARM CPU's being utilized in the server market sector. This same thing happened in the mid 2000's with the Opteron's which were very competitive with the Xeon's of the time. Then the Conroe types hit the market and Intel came charging back.



Total nonsense. AMD & ARM have eaten into Intel market share, they have not toppled it, nor will they.

I would add GPUs as well because some tasks that were performed by CPUs in the past have moved to GPUs. Revenue wise, AMD has already toppled Intel. From the looks of it, AMD makes more money from Epycs than Intel does from Xeons. If you combine Epycs and ARM server CPUs, the total shipped is well above the total Xeons shipped. I know this is a hard reality to understand since it's happening almost overnight.

Client CPUs are a different story with Intel shipping way more CPUs into the laptop and desktop space than anyone else.

 

[Lycanwolfen]

Intel Cannot see the Writing on the wall it seems.

They need to basicly go back to engineering again build up some new CPU's from the ground up. They need to think Low Power, Massive ammount of cores, and Solid and good pricing.

Intel has lost the Server market, Lost the Desktop Market. Lost the workstation market to AMD. I mean low in sales do you need to go before you catch on. I have had intels for a long time. Each time I looked at upgrading I wanted a double ammount of cores. Like going from a core 2 duo to a core 2 quad, Then I went into a core i7 2600k to a 3770k. Then I wanted 8 cores so I when with a core i9 9900 KFC still running this CPU today in my gaming box. I been wanting to upgrade but I see nothing in intel that interests me. Not a single pure 16 core CPU for desktop anywhere. Only option is AMD. Kinda leaning towards a Ryzen 9 9950X3D.

I'm quite Sure AMD will release a 32 core Desktop CPU in the next couple of years. At that point if intel does not have something in those lines might as well go out of business.

 

[Daven]

Intel Cannot see the Writing on the wall it seems.

Intel went from the very peak of its business in 2020-2021 to being targeted for buyout after collapsing just four years later. Everyone at that company has to be shellshocked right now with the chaos of massive layoffs hurting moral in every department.

 

[Chomiq]

No wonder Papa went out to buy cigarettes.

 

[DaemonForce]

Each time I looked at upgrading I wanted a double ammount of cores. Like going from a core 2 duo to a core 2 quad, Then I went into a core i7 2600k to a 3770k.

I'm not sure if upgrade is the right word but I do a bit of back and forth like a rubberband with cores/threads.
In 2004 I wanted the Pentium 4 because high core clock and 2x threads. It was a workhorse but a high voltage thermal bomb at best.
In 2009 I went Phenom II X4. First 64-bit chip, factory speeds breaching beyond 3GHz and quad core/thread. Great chip limited by 4GB of the worst memory.
I was on that until the VR era before hopping over to FX-8370 and kinda stayed there until curious about Ryzen's SMT and new storage technology like M.2.
While the jump from 8c/8t to 6c/12t is weird and underwhelming, it definitely stirred enough efficiency improvements that I'm satisfied with it.
89W->125W->125W->65W is a wild roadmap when you consider that I've been running them all on the same water cooling technology the entire time.

I've never been able to consider a Xeon chip when new or used. There's just nothing exciting about them other than dirt cheap used prices.
A few years ago I picked up a 1c/1t eMachines dono to do my heavy lifting and it works GREAT. Not flawless but 1c/1t 15W running 2 HDDs 24/7 is fantastic.
Everyone has different needs but some of us push everything into the extremes just because it can be done. We do need more low power chips on the market.

 

[Visible Noise]

Intel will be fine as a department of the US government.

 

[Daven]

Everyone has different needs but some of us push everything into the extremes just because it can be done. We do need more low power chips on the market.

I'm an IPC guy myself. I was all Intel until K8 since the P4 was horrible. I stayed on K8 and its derivatives until Core was too hard to ignore and switched to Haswell since Bulldozer was horrible. I upgraded to Coffee Lake and then Zen was too hard to ignore so I switched to Zen 4. Alder Lake was okay but I wanted just the P-cores for the IPC. Now it looks like IPC stagnation has set in so I might have to switch my upgrades to higher core counts.

Intel will be fine as a department of the US government.

Only about 10% of Intel's business is from the government so you are saying revenues will be in the $1B to $2B range per quarter which means massive debt running mostly empty fabs at a tenth capacity. That would mean massive tax payer bailouts to compensate for the debt every quarter. I'm not sure I know of any examples where a company received such treatment from the government quarter after quarter for more than maybe a year.

 

[MaMoo]

Part of the Intel strategy seems to be their specialized accelerators in their server CPUs, for big data processing and analytics. Problem is, they require software support to enable, some cost extra as far as I know, and do not exist on non-server chips. Once people leave Intel on the client and workstation side, the advantage of familiarity evaporates and special accelerators become more of a hindrance. I have viewed their client and workstation chips as a gateway to their server stuff because of things like AVX-512, AMX and so on.

 

[phanbuey]

No wonder Papa went out to buy cigarettes.

Papa fell out a window russian style.

Intel will be fine as a department of the US government.

Government wants only the fabs though.

 

[TheinsanegamerN]

Part of the Intel strategy seems to be their specialized accelerators in their server CPUs, for big data processing and analytics. Problem is, they require software support to enable, some cost extra as far as I know, and do not exist on non-server chips. Once people leave Intel on the client and workstation side, the advantage of familiarity evaporates and special accelerators become more of a hindrance. I have viewed their client and workstation chips as a gateway to their server stuff because of things like AVX-512, AMX and so on.

Their biggest issue there is that AMD absolutely clobbered Intel in AVX512 implementation, taking all the wind out of their sales.

 

[Dr_b_]

Their SKUs got really weird and difficult to understand, there were like 60 different parts with different core counts, weird niche accelerators that were locked behind a paywall and required code to leverage, all of which were overpriced compared to alternatives that might have more cores, more performance and better power consumption. Then there was the huge issue with intels inability to keep up with TSMC. It was a fatal blow of marketing and manufacturing incompetence and bad leadership.

Expect a headline like this for client segment coming soon. Was in a best buy the other day, and saw 14900Ks on the shelf for $499, there's a price those will sell at, but its not that.

 

[ScaLibBDP]

...They need to think Low Power, Massive ammount of cores, and Solid and good pricing...

Actually Intel did that many years ago with Intel Xeon Phi processors and failed. This is because for the core part of Intel Xeon Phi architecture Intel Atom architecture was selected instead of Intel Ivy Bridge.

I was using an Intel Xeon Phi server for many months to port a medium size C/C++ project. All my feelings I would describe as follows ( some kind of transitions ):

Very Excited -> I have a lot of questions -> Sorry, but that does Not work Too Fast and I need More memory! -> On one morning the Project was cancelled. Oops...

Here is my quick summary:

Low power - Intel Atom based low power cores did Not help. The Intel Xeon Phi servers were using a lot of power.

Number of Cores - Looked good but out of 256 Logical Processors only 64 had to be used in order to achieve top HPC performance ( in FLOPS ).
Note: There was only one physical FPU ( Floating Point Unit ) for 4 Logical Processors in a physical core.

Pricing - All these Intel Xeon Phi servers were very-very expensive.

 

[223 Fan]

14th Gen reliability issues were also first noticed in server farms.

 

[mkppo]

One of Intel's biggest issues is Turin. Right when Intel launched all new products that was an order of magnitude faster than their previous gen and sort of matched AMD's previous flagship (Genoa), Turin comes in and absolutely rips Xeons. Even the ARM competition can't hold a candle to it.

In the server space it wasn't anything like Zen 4 > Zen 5, even factoring in Zen 5's sizeable performance increase in server applications. It's more like they jumped two straight generations.

This downward spiral isn't stopping anytime soon. Right now they're just trying to hold on to the markets they have by giving heavy discounts to SI's.

 

[AnotherReader]

Actually Intel did that many years ago with Intel Xeon Phi processors and failed. This is because for the core part of Intel Xeon Phi architecture Intel Atom architecture was selected instead of Intel Ivy Bridge.

I was using an Intel Xeon Phi server for many months to port a medium size C/C++ project. All my feelings I would describe as follows ( some kind of transitions ):

Very Excited -> I have a lot of questions -> Sorry, but that does Not work Too Fast and I need More memory! -> On one morning the Project was cancelled. Oops...

Here is my quick summary:

Low power - Intel Atom based low power cores did Not help. The Intel Xeon Phi servers were using a lot of power.

Number of Cores - Looked good but out of 256 Logical Processors only 64 had to be used in order to achieve top HPC performance ( in FLOPS ).
Note: There was only one physical FPU ( Floating Point Unit ) for 4 Logical Processors in a physical core.

Pricing - All these Intel Xeon Phi servers were very-very expensive.

SMT4 was a strange choice for a HPC focused product.

 

[unwind-protect]

Let's not forget that they wasted development effort on special computation units for the current Xeons. I don't think anybody is using those.

14th Gen reliability issues were also first noticed in server farms.

Not on datacenter Xeons. Those were socket 1700 units, just in mainboards that support ECC memory (unbuffered).

As far as I have seen the Xeons have at least been problem free.

 

[ScaLibBDP]

SMT4 was a strange choice for a HPC focused product.

Here are specs for Intel Xeon Phi Processor 7210 I was working with:

Intel product specifications

Intel® product specifications, features and compatibility quick reference guide and code name decoder. Compare products including processors, desktop boards, server products and networking products.

ark.intel.com

Intel Xeon Phi Processor 7210 ( 16GB, 1.30 GHz, 64 core )
Processor name: Intel(R) Xeon Phi(TM) 7210
Packages ( sockets ): 1
Cores: 64
Processors ( CPUs ): 256
Cores per package : 64
Threads per core: 4
Peak Processing Power: 2.662 TFLOPs
Note: Calculated as follows: 1.30 * 64 * ( 512-bit / 32-bit ) * 2 = 2662.4 GFLOPs for Single-Precision FPU data type

Another tech-mess was related to AVX-512 because of fragmentation the AVX-512 ISA.

It was hard to imaging that Intel did it!
...
Intel AVX-512 family of fnstructions:
- Intel AVX-512F Foundation instructions.
- Intel AVX-512CD Conflict Detection instructions.
- Intel AVX-512ER Exponential and Reciprocal instructions.
- Intel AVX-512PF Prefetch instructions.
- Intel AVX-512BW Integer operations on 8-bit and 16-bit operands.
- Intel AVX-512DQ Enhanced Integer and Floating-Point operations on 32-bit and 64-bit operands.
- Intel AVX-512VL Vector Length Extensions.

Intel Xeon Phi processor x200 products support:
- Intel AVX-512F Foundation instructions.
- Intel AVX-512CD Conflict Detection instructions.
- Intel AVX-512ER Exponential and Reciprocal instructions.
- Intel AVX-512PF Prefetch instructions.

Intel Xeon processors support:
- Intel AVX-512F Foundation instructions.
- Intel AVX-512CD Conflict Detection instructions.
- Intel AVX-512BW Integer operations on 8-bit and 16-bit operands.
- Intel AVX-512DQ Enhanced Integer and Floating-Point operations on 32-bit and 64-bit operands.
- Intel AVX-512VL Vector Length Extensions.
...