Server makers love Intel Xeons (true) - but not the price tag

Re: Competition

It's price per server "transaction", not raw CPU power or CPU price.

If two ARM servers give similar performance to one Intel but cost less to buy and less to run ...

Re: Competition

AMD has been essentially broke for most of the past 20 years, but that hasn't stopped them competing. Intel always backed off a bit when AMD was under stress because they needed to keep them around so they could claim they have competition.

They probably still need them around since the FTC may consider "server CPUs" to be a relevant market, and the pittance produced by IBM and Oracle (for their own use only) is a drop in the bucket - far less of the total market share than AMD has historically had in desktop CPUs.

If it weren't for server CPUs I think Intel could drop their prices for a year or two and kill AMD without issue, since they can likely argue successfully that desktop, laptop and mobile CPUs constitute a single relevant market. They can point to the obvious cannibalization of PCs by mobile devices in the past five years.

Why Opteron (briefly) succeeded

It was better than Intel's offerings at the time it came out, but AMD only had the advantage over Intel because of two huge Intel mistakes.

One, they pursued the dead end Netburst architecture, believing they could goose it up to 10 GHz by the end of the decade. They probably could have (by exposing the double pumped pipeline this would have been rather easy, in fact) but it turns out almost no one wants PCs with 200 watt CPUs. Their engineers obviously understood the relationship between frequency and power, so clearly they must have thought that the curve of ever increasing power demand for desktop CPUs that had held for 20 years would continue to hold. Perhaps if CPUs weren't "fast enough" for most people it would have continued, but once they were people valued quiet for desktops and battery life for laptops far more than Intel's engineers had figured.

Two, they purposefully held back offering a 64 bit ISA for x86, because they hoped to keep x86 32 bit only and force everyone to their proprietary Itanium CPUs over a decade's time as lower and lower end markets hit the 2GB/4GB limit and needed to go 64 bit. They assumed AMD could never get any support or backing if they developed their own 64 bit x86 ISA, similar to how their attempts at SIMD instructions like 3DNow! went nowhere and AMD was eventually forced to license MMX and SSE. If they hadn't make the concurrent mistake with Netburst, maybe that would have held true, and we'd all be using Itaniums right now.

When Microsoft announced support for AMD64, Intel was caught off guard since they hadn't notified them of this in advance. Itanium never recovered, and Intel was forced to be compatible with AMD's instruction set for once.

Interestingly, Hans De Vries discovered evidence of 64 bit support in Intel x86 CPUs that predated the release of AMD64. Perhaps there was a war in Intel between the x86 people and the Itanium people and the x86 people put it in hoping to get management approval to enable it, but it never was and nothing about it was ever published.

Re: It's a case of ....

"But when pretty much no-body makes off-the-self servers using AMD chips then what choice do you have?"

And that has a lot to do with AMD's chips being cheaper but having significantly higher power consumption compared to the Intel equivalent. TOC ends up being higher and that's a bigger deal in a server farm than CapEx.

Re: Make Servers, Not War!

"but in the end it's still x86 with all the baggage and compatibilities to the old-world it drags or brings."

What matters is what it does and how much power it consumes whilst doing so, not what baggage it brings.

This is why Intel have beaten AMD in the datacenter for some time (The only reason I have AMD compute servers is because at the time it was the only way to get more cores in the box)

If ARM beats Intel on this then it will rapidly take over for non compute-centric applications. The problem is that as ARM has gotten faster its power consumption and purchase price has approached the low end Xeons for the same performance as those low-end Xeons.

Where ARM might win is if a few hundred cores could be put into the box, but that's not happening either (yet).