News AMD vs. Intel CPU Market Share Q4 2019: EPYC and Ryzen Growth Decelerate, Mobile Ryzen Up

[Admin]

Mercury Research released its quarterly CPU market share report that reveals a deceleration of AMD's Ryzen and EPYC growth, but faster growth in the mobile segment.

AMD vs. Intel CPU Market Share Q4 2019: EPYC and Ryzen Growth Decelerate, Mobile Ryzen Up : Read more

 

[digitalgriffin]

If you think about it, desktop saturation is almost a given. I think the rough estimates say that AMD's production capabilities are only 1/5th that of Intels. And 1/5th of Intels is ~20%...Guess where they are sitting? Yes, they are pretty darn close to that.

AMD doesn't get huge revenue growth (50%) by sitting on unsold product. They are selling out most of what they have.

For AMD's long term growth however, laptop & server markets become crucial. I'm happy to see the laptop market take off because business laptop lease renewals for new laptops is a steady income.

However, I'm worried about the lack luster server growth. This is where the big money is. I know validation takes a long time, but EPYC Rome should be selling like hotcakes given the low entry point / core and TCO.

 

[twotwotwo]

So AMD's opportunities this year seem like 1) better mobile parts to take advantage of the shortages there (and maybe steal a little market share), 2) maybe muscling into the mid-to-high-end GPU market a bit more, and 3) semi-custom, though that's lower-margin.

The most plausible extreme possibilities seem to mostly depend on Intel. If Intel can't make 10nm server parts in large quantities or at high core counts, for example, maybe AMD sees good server growth. If things start clicking for Intel product-wise, and Intel chooses to cut prices a ton rather than continue to bet on their (amazing) inertia, that's harsher.

Longer term it seems like AMD CPUs stay competitive for at least the next couple gens (Zen 4 this year, 5nm shrink after that); to be seen if that actually wears away Intel's ecosystem advantage much.

Also: software seems like a real advantage for their competitors (CUDA, MKL, ICC). Dunno what the software analogue is of hiring Jim Keller for a few years, but I'd love to see it. Given AMD's position they could benefit from making their GPUs' compute units, and the cores and vector capabilities on their CPUs, usable for mere mortal programmers, as huge an effort as it is.

 

[JayNor]

Intel didn't make any 10nm server parts in q4, yet their server revenues were up 19%.

Some leaks about Cascade Lake refresh coming. Cooper Lake is also coming in 1H. Both are 14nm. Cooper Lake will have Barlow Pass Optane support, more memory channels, faster memory, bfloat16 support in avx512.

Intel also talking about Snow Ridge 5G infrastructure chips coming in 1H. Those are 10nm. Also, the 10nm Agilex FPGAs are shipping. 10nm Tiger Lake laptop chips are shipping to OEMs in the summer.

I don't think they have fab capacity to start any kind of huge 10nm Ice Lake Server shipments, but Intel says those chips are coming in q4. Maybe that's when the capacity expansion kicks in.

 

[donner]

Intel didn't make any 10nm server parts in q4, yet their server revenues were up 19%.

When the patches for Intel security flaws reduce performance and it takes many months to qualify new servers. The only option a data center has is to buy more of the same Intel servers to get back up to full compute capacity.

 

[GoatGuy]

If you think about it, desktop saturation is almost a given. I think the rough estimates say that AMD's production capabilities are only ⅕th that of Intels. And ⅕th of Intels is about 20%…Guess where they are sitting? Yes, they are pretty darn close to that.

Hmmm… if AMD's output is ⅕ Intel's, then Intel's is ⁵⁄₅. Which means AMD's potential market share is ⅕ / ( ⅕ ⊕ ⁵⁄₅ ) → (⅕ ÷ ⁶⁄₅) → ⅙ → ≈ 17%

Even closer to actual figures!

⋅-⋅-⋅ Just saying, ⋅-⋅-⋅<br>⋅-=≡ <b>Goat</b>Guy ✓ ≡=-⋅<br>

 

[800Volts]

However, I'm worried about the lack luster server growth. This is where the big money is. I know validation takes a long time, but EPYC Rome should be selling like hotcakes given the low entry point / core and TCO.

The problem is, AMD combines server with customer silicon (read Xbox / Play Station).

The consoles crashed harder than anyone imagined, and EPYC saved the day is my hypothesis. Hopefully, they seperate out the business lines for reporting.. until then....

 

[800Volts]

Also: software seems like a real advantage for their competitors (CUDA, MKL, ICC). Dunno what the software analogue is of hiring Jim Keller for a few years, but I'd love to see it. Given AMD's position they could benefit from making their GPUs' compute units, and the cores and vector capabilities on their CPUs, usable for mere mortal programmers, as huge an effort as it is.

Chicken or Egg, my guess is Intel has more software engineers than AMD employees. It's what market leadership margins buy you, same with NVIDIA.

 

[Rdslw]

However, I'm worried about the lack luster server growth. This is where the big money is. I know validation takes a long time, but EPYC Rome should be selling like hotcakes given the low entry point / core and TCO.

I know that servers ARE very very high inertia systems. If it works, no touching. And unless it solves other problem it will stay as it is. only netflix did instantly catch the wind of the gains because they had constant problems of bandwith and epic did magically almost tripled it for them.
but for now half of the companies have EPIC in backlog for "review", few execute tests if their crapware wont explode.... it takes a while.

 

[Olle P]

AMD doesn't get huge revenue growth (50%) by sitting on unsold product. They are selling out most of what they have. ...
... However, I'm worried about the lack luster server growth.

Epyc CPUs are essentially "flying off the shelves" which is the reason Threadripper is so hard to find in stock.
AMD sell all Epyc they can produce, and the demand is going up.

The (server market) situation is actually perfect for AMD, with huge deliveries going to a few (new) facilities at for example Amazon and Google while AMD build up the broader support organisation required to handle a larger amount of customers a few years from now.

... half of the companies have EPIC in backlog for "review", ... it takes a while.

Yup. The transition process has begun and demand for CPUs will come.

 

[cryoburner]

Chicken or Egg, my guess is Intel has more software engineers than AMD employees. It's what market leadership margins buy you, same with NVIDIA.

Nvidia and AMD have a roughly similar number of employees, though with AMD covering a wider range of products, they undoubtedly have fewer working in their graphics division. Intel, of course, has somewhere around 10 times the employees of either company.

 

[digitalgriffin]

Nvidia and AMD have a roughly similar number of employees, though with AMD covering a wider range of products, they undoubtedly have fewer working in their graphics division. Intel, of course, has somewhere around 10 times the employees of either company.

While this is true, Intel as has like 20x's the product portfolio. From Vision sensors, AI products, graphics, networking, modems, storage solutions, memory, FPGA, and custom embedded. (Not just x86) So intel has a lot of pokers in various fires. They aren't singularly focused.

It's both a strength and weakness. When a certain division comes under attack, the heads at the top might not be able to properly pay attention to that division because they have a lot under their umbrella to cover. But they have a lot of resources. It's like asking the titanic to change course. With big companies, it takes time.