Question x86 and ARM architectures comparison thread.

Page 19 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

poke01

Diamond Member
Mar 8, 2022
4,752
6,092
106
at some point one has to ask "when will intel make fat client cores that sip power in 1T?"

1761392422779.png
 

okoroezenwa

Member
Dec 22, 2020
180
181
116
Intel P cores are bigger physically than Apple M P cores. Why do you think Intel doesn't make "fat" cores?
“fat client cores that sip power in 1T” is a description of one thing, not 2 things. That is, poke is asking for a fat core from Intel that also sips power, not a fat core and a core that sips power.
 

mikegg

Platinum Member
Jan 30, 2010
2,091
633
136
“fat client cores that sip power in 1T” is a description of one thing, not 2 things. That is, poke is asking for a fat core from Intel that also sips power, not a fat core and a core that sips power.
If Intel could make a fat core that sips power, don't you think they would have by now? It's not like they looked at the M1 P core and said, "nah, we don't need to compete".

AMD also can't.

But Qualcomm can.

So obviously there's something inherent in Arm core designs.
 

Thunder 57

Diamond Member
Aug 19, 2007
4,244
7,039
136
AMD CPUs are very server focused though. When was the last time Intel and AMD made a core design just for client?

Probably Conroe for Intel. It wasn't great for server until Nehalem. LL was kind of a one off. As for AMD, maybe K7? Even then they were pushing to get server market share with Athlon MP.

LNC was for client and it sucks absolute donkeyballs for server.

Lion Cove is made by Intel doe.
 

mikegg

Platinum Member
Jan 30, 2010
2,091
633
136
AMD CPUs are very server focused though. When was the last time Intel and AMD made a core design just for client?
Doesn't matter if they are very server focused. At the end of the day, what are consumers actually buying for their money?

Further more, Arm has a bigger server share than AMD now.
 

cherullo

Member
May 19, 2019
59
133
106
AMD CPUs are very server focused though. When was the last time Intel and AMD made a core design just for client?
AMD's Jaguar, circa 2013.
Intel's Atom line of cores, up to at least Tremont, circa 2020.

Edit: For AMD actually it was probably Excavator around 2015. By that time, the dozer line of cores had no chance in servers anymore and so AMD focused it on client applications.
 
Last edited:

soresu

Diamond Member
Dec 19, 2014
4,242
3,744
136
Edit: For AMD actually it was probably Excavator around 2015. By that time, the dozer line of cores had no chance in servers anymore and so AMD focused it on client applications.
Not so sure about that.

IIRC Excavator was also something of a test for high density cell design on a high power CPU.
 

Covfefe

Member
Jul 23, 2025
90
154
66
AMD's Jaguar, circa 2013.
Intel's Atom line of cores, up to at least Tremont, circa 2020.

Edit: For AMD actually it was probably Excavator around 2015. By that time, the dozer line of cores had no chance in servers anymore and so AMD focused it on client applications.

Not so sure about that.

IIRC Excavator was also something of a test for high density cell design on a high power CPU.

Yeah, I think it's more useful to talk about the intended market and target wattage rather than what products actually came out. Jaguar was put in an Opteron SKU, but it's clearly not designed for high performance. AMD had server aspirations for the Bulldozer family, but it was so uncompetitive that they cancelled most of them.

In terms of current archs, I would say Zen and Intel's P-cores are server-first. Whereas all the major ARM cores are mobile-first (with the possible exception of Neoverse). The x86 cores are designed to operate with a higher average power draw.
 
  • Like
Reactions: cherullo

Doug S

Diamond Member
Feb 8, 2020
3,791
6,723
136
Put down the crack pipe

While I don't buy the "ARM is bigger" thing, I don't buy those numbers either. What are they measuring, prebuilt server shipments? Are they estimating server shipments based on server CPU shipments

They would need some way to account for Graviton server builds for example, since Amazon is having those fabbed itself and installing them itself. The analysts can only guess at that, and such guesses could be wildly inaccurate. How will they account for Apple's servers, which would be even harder to guess at than Amazon's.

Even if they were, what's the "value" of an Amazon or Apple server that's built for internal use, versus buying one on the retail market that carries Intel/AMD's profit margin along with Dell's?
 

Doug S

Diamond Member
Feb 8, 2020
3,791
6,723
136
Uh, no?
They're outsourced to Foxconn or Quanta or any other ODM really.

They've outsourced the design and fabrication of Graviton CPUs to Foxconn? Try again.

Maybe they've contracted with Foxconn to have them assemble the Graviton CPUs they've contracted with TSMC to have fabbed into servers. That's probably what Apple has done as well.

How does that translate into numbers that analysts can get hold of? Foxconn isn't going to tell them "we assembled 2 million servers for Amazon and 200K for Apple this year" because that kind of info is going to be covered by NDA.
 

gdansk

Diamond Member
Feb 8, 2011
4,722
7,995
136
I saw that post and thought no way in hell. Thanks for clarifying that.
But like Doug S says it's probably not that as bad as value share for ARM because they have many ways for their sales to not show up well in that metric.

In their previous quarterly reports ARM noted that power consumption is a limit to their ability to compete in the server market. So, hmm, not sure what they mean by that.
 

gdansk

Diamond Member
Feb 8, 2011
4,722
7,995
136
They've outsourced the design and fabrication of Graviton CPUs to Foxconn? Try again.
One of the sources does track the server shipment from ODMs.
I don't have the report so I cannot look up how they estimate CPU value from that. But based even on this chart BofA was "buy" on ARM and unsurprisingly AMD too.
 

mikegg

Platinum Member
Jan 30, 2010
2,091
633
136
  • Haha
Reactions: Thibsie

mikegg

Platinum Member
Jan 30, 2010
2,091
633
136
Put down the crack pipe
View attachment 133296
No need to put the pipe down. The pipe is way up.

Fast forward to today, the adoption of Neoverse has reached new heights: close to 50 percent of the compute shipped to top hyperscalers in 2025 will be Arm-based.

No chance AMD is even close to 50% of CPUs supplied to AWS, Google, Microsoft.

Hyperscalers are vast majority of the CPU server market. Intel still leads AMD in x86 server unit shipped. Arm leads Intel.
 
Last edited: