Read more.Quote:
Claims CPU throughput and per-thread performance is a match for x86 processors.
Printable View
Read more.Quote:
Claims CPU throughput and per-thread performance is a match for x86 processors.
Concerned if it only *matches* in power efficiency. I thought the whole point of ARM on servers was to be more power efficient, potentially by sacrificing some performance for a lot of power draw. If it only matches x86 in every metric, what's going to drive uptake? It'll be easier to stay on x86 and know you're not losing out on anything...Quote:
"matches comparable x86 processors in CPU throughput, per-thread performance and power efficiency" ...
Looks like they are touting cost and memory bandwidth but I agree I don't see success for an ARM based CPU that performs like a Xeon.
There have been POWER, SPARC and other instruction set CPUs that could in theory compete with the contemporary Xeons but they haven't grabbed large market share, I don't see why an ARM based CPU will be any different.
ARM based CPUs are successful in mobile devices because the available SoC designs are better optimised for low power or performance in power constrained environments and not because using the ARM instruction set is fairy dust that just makes a CPU better...
While I agree that the power efficiency is a little disappointing for an ARM processor, being cheaper and increased memory bandwidth should go a long way to slicing out a share of the market.
The company has been around since 1950 - they must be doing something right if they are still around!! :p
PRICE? I'm guessing they aren't charging $7000 per chip? :) If you match the other guy in everything and sell at 1/2 or 6/10 the cost, you probably get a few customers who simply can't afford Intel.
"matches comparable x86 processors in CPU throughput, per-thread performance and power efficiency, while offering advantages in memory bandwidth and total cost of ownership."
ARM is successful in mobile because there are a gazillion SOCs out there to choose from, so you can probably find exactly the right chips for your application.
Intel make something like 4 dies, fuse cores and features off to create their entire Xeon and core range. Little chance for Darwinism there. So whilst the AMD64 instruction set probably only adds about 5% to the die size and doesn't rob much from the execution speed these days compared to a proper RISC ISA, all the interesting developments seem to happen in the ARM ecosystem.
Yup, a processor that it matches could definitely said to be comparable.Quote:
matches comparable x86 processors in CPU throughput, per-thread performance and power efficiency
Well yes, that, same point. The success was not down the ARM instruction set but the actual marketed products... could have been MIPS or some other architecture that dominated if the available SoCs had been right.
Some people seem to think that a ARM based CPUs are somehow innately amazing and are going to tear up the order book and dominate the server market, I just don't see it happening - the designs need to offer some real concrete and mainstream benefit. A 32 core CPU isn't going to be cheap, Intel list upwards of $4500 for 22+ core Xeons - it's a niche market and one that AMD look likely to disrupt with an instruction set compatible product later this year...
ARMs server value comes from datacentre running costs. More efficient and lower power = less electricity for servers and less DC cooling needed. Win win.
I was hoping to see by now bigger core ARM chips for laptop use. As far as I know only Apple seems to be big core low core count.
Is it just not possible or extremely not worthwhile (in terms of CPU grunt not market value) to make bigger high IPC cores using ARM?
It is partly instruction set, partly cost reasons.
In the past Intel got inroads into previously expensive areas like servers where SPARC dominated and workstations where MIPS dominated. They used their mass production to make good enough chips that took enough of the market away that it became difficult for the incumbent architectures to continue.
These days ARM chips are made in numbers that make Intel look like the giants of old. Propped up by the phone industry in the same way Intel was propped up by the PC industry, they have a business model that Intel can't compete with. So here is the concrete benefit: If you want an SOC designed, you can go to multiple companies and get competitive quotes. All Intel needed was good enough chips and an overpriced incumbent, and that is exactly what ARM have now.
On the flip side, I don't think Intel could open up their architecture if they wanted to. If they did, then x86 doesn't scale well at the sort of low end that phones use.
The sad part to me is that Intel have the answer, they own the DEC Alpha architecture, one of the best ever. They killed it for politics, but I expect it would make a cracking low power server and they could license it without threatening their AMD64 sales. They won't because they believe everything should be x86, and after a long list of failed embedded processors that isn't too surprising, but I think that attitude will be their undoing.