ARM vs x86 Server CPU Market share, per Mercury Research

[Xebec]

I saw this posted elsewhere. This is volume, not revenue, and doesn't include other "still standing" CPU architectures like IBM Power, but provides some interesting trend data.

 

[MKWVentures]

good info, thanks

what was the official number and QoQ change for AMD and Intel?

Nvidia, Google, Amazon CPUs are counted as ARM? correct?

 

[Yunus]

Source:

https://twitter.com/x/status/2021695106198163569

 

[MKWVentures]

View attachment 4205
Source:

https://twitter.com/x/status/2021695106198163569

I got the report summary from an analyst as well.

AMD gained share in revenue and units. AMD ASP is higher than Intels. Intel is limited by IFS capacity, AMD is not limited. AMD GM is much higher than Intels even though they outsource to TSMC.

Customers are wanting old stuff from Intel, Not the new stuff. This explains how Intel can have flat revenue in this PC/Server Boom while everyone else grows.

 

[siliconbruh999]

I got the report summary from an analyst as well.

AMD gained share in revenue and units. AMD ASP is higher than Intels. Intel is limited by IFS capacity, AMD is not limited. AMD GM is much higher than Intels even though they outsource to TSMC.

They use the same dies for client and server so they can move stuff here and there

Customers are wanting old stuff from Intel, Not the new stuff. This explains how Intel can have flat revenue in this PC/Server Boom while everyone else grows.

Intel hasn't even ramped GNR so it's a entirely Intel issue

 

[tomatoma]

I got the report summary from an analyst as well.

AMD gained share in revenue and units. AMD ASP is higher than Intels. Intel is limited by IFS capacity, AMD is not limited. AMD GM is much higher than Intels even though they outsource to TSMC.

Customers are wanting old stuff from Intel, Not the new stuff. This explains how Intel can have flat revenue in this PC/Server Boom while everyone else grows.

That said, that doesn't mean AMD has no restrictions. Right now, the N4 process is the mainstream, so it's fine, but if they move to N2 or something, the competition for production equipment will become much more intense, right?

 

[IanD]

That said, that doesn't mean AMD has no restrictions. Right now, the N4 process is the mainstream, so it's fine, but if they move to N2 or something, the competition for production equipment will become much more intense, right?

True, but also if you're using core chiplets like AMD there's less advantage to going to the next node because density (gates/mm2) doesn't matter so much as with large monolithic die, all that matters is power/speed (only improving slowly per node nowadays) and cost per die (static or even rising).

Of course even a small improvement in power/speed (e.g. 10% extra speed at same power or 20% lower power at same speed) is always worth having, but if higher density/smaller die size doesn't buy you anything and there's no cost saving (or even a cost increase) there's less drive to go to the bleeding-edge as fast as possible...

 

[Paul2]

What are "units"? If they count in all the embedded server SoCs, it's no wonder at all. And on the other side, they may be penalising AMD, as they have large core counts.

 

[siliconbruh999]

What are "units"? If they count in all the embedded server SoCs, it's no wonder at all. And on the other side, they may be penalising AMD, as they have large core counts.

Number of CPU most likely

 

[MKWVentures]

What are "units"? If they count in all the embedded server SoCs, it's no wonder at all. And on the other side, they may be penalising AMD, as they have large core counts.

essentially it is a model for CPU units sold (number of CPUs) sold my each company. 200 core vs 16 core.... both one CPU.

Mercury research is quite accurate.

 

[MKWVentures]

That said, that doesn't mean AMD has no restrictions. Right now, the N4 process is the mainstream, so it's fine, but if they move to N2 or something, the competition for production equipment will become much more intense, right?

The boom is now and AMD is not as constrained as Intel. N2 is a 2027 problem

The issue at Intel is simple: The plan was to move people to newer nodes. It didnt happen. So the AI boom increased Intel 7 Demand while the Intel 3 and 18A demand is less than planned 1-2 year ago. The answer is to force people to newer nodes.

But 18A needs to control ramp to prevent financial issues, Intel 3 products are growing slowly (Datacenter ramp is always slow and the products are not as gangbuster as hoped). Rumor is that Lunar lake is the way out (force people to Lunar Lake, No DRAM for OEM, free up Intel 7 for Server).