Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Microsoft Pledges to Use ARM Server Chips, Threatening Intel's Dominance

bloomberg.com

Post reply on HN
Print view View original

151–160 of 239 posts

Re: Microsoft Pledges to Use ARM Server Chips, Threatening Intel's Dominance

#151

Post by trome »

freehunter wrote: post #126
toyg wrote: post #113

Earlier quoted context omitted.

Servers and desktops have very different requirements, just ask the linux guys. I'm not saying it's impossible, but I don't see that happening for another 5 years at least. Apple in particular is already getting a lot of flak for underpowering laptops, switching to Arm now would basically kill the entirety of their remaining high-end offering on desktop.

On the other hand, their A10 chip is roughly on par with the performance of the Intel chip they include in the 12" MacBook. They're not going to replace an i7 with an A10, but maybe replace the M3 or M5 with an A10, replace the i5 with an A11, and shoot for the i7 with an A12 or A13. Why else would they dump so much money into making the fastest ARM chip on the market? The iPhone doesn't need that much power, but the…

Apple does have a massive team working on optimizing their ARM chips, they started out with redesigning the uncore, hired quite a bit of talent from other companies, and have moved to optimizing the CPU itself.

They even forced Global Foundries and Samsung to sync processes at 14nm to build their chips:

http://semiaccurate.com/2014/04/17/semiaccurate-right-apples...

Re: Microsoft Pledges to Use ARM Server Chips, Threatening Intel's Dominance

#152

Post by milcron »

andreiw wrote: post #143
rleigh wrote: post #59

Earlier quoted context omitted.

Potentially capable; their support for non-x86 has always fallen short. If you look at e.g. Linux or BSD distributions, the entire world is rebuilt for every architecture. Running Linux on powerpc, arm, amd64, I get the exact same experience across the board as x86 bar platform-specific bits like openfirmware/efi tools. Microsoft has never done this. The vast majority of their stuff remains x86 only, making arm and e…

Not true. When multiarch mattered (NT 3 and 4), Microsoft was the only vendor that delivered the exact same OS, device support and development environment across x86, Alpha, MIPS and PowerPC, and pushed the industry towards standardization. Microsoft has always taken arch and platform independence seriously, and this was evident even in 2010 when I worked on NT. You could rebuild the whole system for any supported ar…

BTing?

Re: Microsoft Pledges to Use ARM Server Chips, Threatening Intel's Dominance

#153

Post by justin66 »

trome wrote: post #146
justin66 wrote: post #66

Earlier quoted context omitted.

> Running Linux on powerpc, arm, amd64, I get the exact same experience across the board as x86 bar platform-specific bits like openfirmware/efi tools. This is demonstrably not true, there are plenty of ports that end up being Intel-only, and plenty of architectures that involve some sacrifice in terms of software choice if you want to run them. (as a SPARC and Raspberry Pi user I could elaborate, but hopefully you g…

> There are plenty of ports that end up being Intel-only .. as a SPARC and Raspberry Pi user Sure, but you can compile failed builds of software yourself and get that tool on ARM or SPARC. ARM works pretty well on Debian, so long as your not on Raspbian & using their ancient repos with broken software. Part of why I've avoided Raspberry Pis entirely.

A good point, but their repos are so eccentric because Debian refused to support the RPi's older ARM architecture and the RPi community had to roll their own. That speaks to the original point of this subthread, I believe. Not only does Linux not offer the "exact same experience across the board," it cannot even do it for the ARM architecture. For decent enough reasons, really, when you think about it.

Re: Microsoft Pledges to Use ARM Server Chips, Threatening Intel's Dominance

#154

Post by AlphaSite »

deepnotderp wrote: post #125
nomel wrote: post #108

Earlier quoted context omitted.

> actually have the good chips. Has that ever been the case yet, for ARM servers? There's no "magic" in ARM. The reduced power comes from the reduced complexity. Heck, branch prediction is fairly new in ARM. I have trouble seeing a time when an Intel chip will ever fall behind even a bucket full of ARM chips, unless workloads ever fall near zero.

There are a lot of nice things in arm. Ie slightly less register renaming. For example, Jim Keller estimates a fully developed arm chip will be 15-30% more efficient than an equivalent x86. Now, this development hasn't happened yet, but with giants like Qualcomm, it will happen eventually.

If this was the case (and there weren't other issues e.g. insturctuon fusion etc) then wouldn't AMD be pushing K12 far harder? It's the same ceneteral core as Zen but they don't seem to have as much confidence in it.

Re: Microsoft Pledges to Use ARM Server Chips, Threatening Intel's Dominance

#155

Post by trome »

justin66 wrote: post #153
trome wrote: post #146

Earlier quoted context omitted.

> There are plenty of ports that end up being Intel-only .. as a SPARC and Raspberry Pi user Sure, but you can compile failed builds of software yourself and get that tool on ARM or SPARC. ARM works pretty well on Debian, so long as your not on Raspbian & using their ancient repos with broken software. Part of why I've avoided Raspberry Pis entirely.

A good point, but their repos are so eccentric because Debian refused to support the RPi's older ARM architecture and the RPi community had to roll their own. That speaks to the original point of this subthread, I believe. Not only does Linux not offer the "exact same experience across the board," it cannot even do it for the ARM architecture. For decent enough reasons, really, when you think about it.

Only a select few boards were based on ARMv6, the Raspi 2 and above use a modern ARMv7 core. Running Raspbian vs Debian on a Raspi 2 or 3 shows off the massive performance gap between the two, IMO they should have used a single ARMv7 core from the get go.

But the Raspberry Pi is a fundamentally flawed platform, with poor I/O, binary blobs required to make the hardware function, and a community that is toxic towards free software, with its own vaguely supported distro.

An OrangePi Zero ($7) or OrangePi PC Plus ($22) will blow a Raspi out of the water any day, due to each USB port and the ethernet port being directly wired into the SOC, allowing 40MB/s per port. Plus, I can run kernel 4.10 and mainline Debian on it without any blobs, and the only things I'll miss out on are GPU support & WiFi. The VPU has been reverse engineered though, so H.264 & H.265 video works well.

Re: Microsoft Pledges to Use ARM Server Chips, Threatening Intel's Dominance

#156

Post by AlphaSite »

justin66 wrote: post #153
trome wrote: post #146

Earlier quoted context omitted.

> There are plenty of ports that end up being Intel-only .. as a SPARC and Raspberry Pi user Sure, but you can compile failed builds of software yourself and get that tool on ARM or SPARC. ARM works pretty well on Debian, so long as your not on Raspbian & using their ancient repos with broken software. Part of why I've avoided Raspberry Pis entirely.

A good point, but their repos are so eccentric because Debian refused to support the RPi's older ARM architecture and the RPi community had to roll their own. That speaks to the original point of this subthread, I believe. Not only does Linux not offer the "exact same experience across the board," it cannot even do it for the ARM architecture. For decent enough reasons, really, when you think about it.

Isn't that more a case of no support promised, none given. Rather than support promised but falling short.

Re: Microsoft Pledges to Use ARM Server Chips, Threatening Intel's Dominance

#157

Post by deepnotderp »

AlphaSite wrote: post #154
deepnotderp wrote: post #125

Earlier quoted context omitted.

There are a lot of nice things in arm. Ie slightly less register renaming. For example, Jim Keller estimates a fully developed arm chip will be 15-30% more efficient than an equivalent x86. Now, this development hasn't happened yet, but with giants like Qualcomm, it will happen eventually.

If this was the case (and there weren't other issues e.g. insturctuon fusion etc) then wouldn't AMD be pushing K12 far harder? It's the same ceneteral core as Zen but they don't seem to have as much confidence in it.

Because migration is super expensive and a big part of what's preventing the movement to ARM servers. Also, the aforementioned case hasn't necessarily happened, Keller has spent forever making Zen better. They are still working on K12 IIRC, but Ryzen, for good reason, is the focus. Besides, CPUs are more and more becoming simply host systems.

Besides, never doubt the word of jim keller ;)

Re: Microsoft Pledges to Use ARM Server Chips, Threatening Intel's Dominance

#158

Post by trome »

AlphaSite wrote: post #156
justin66 wrote: post #153

Earlier quoted context omitted.

A good point, but their repos are so eccentric because Debian refused to support the RPi's older ARM architecture and the RPi community had to roll their own. That speaks to the original point of this subthread, I believe. Not only does Linux not offer the "exact same experience across the board," it cannot even do it for the ARM architecture. For decent enough reasons, really, when you think about it.

Isn't that more a case of no support promised, none given. Rather than support promised but falling short.

Pretty much, Debian moved forward to recompiling for ARMv7 as it netted significant performance improvements, and if you wanted ARMv6 binaries you could compile them yourself. Raspbian did this, but not very well.

Re: Microsoft Pledges to Use ARM Server Chips, Threatening Intel's Dominance

#159

Post by rleigh »

justin66 wrote: post #153
trome wrote: post #146

Earlier quoted context omitted.

> There are plenty of ports that end up being Intel-only .. as a SPARC and Raspberry Pi user Sure, but you can compile failed builds of software yourself and get that tool on ARM or SPARC. ARM works pretty well on Debian, so long as your not on Raspbian & using their ancient repos with broken software. Part of why I've avoided Raspberry Pis entirely.

A good point, but their repos are so eccentric because Debian refused to support the RPi's older ARM architecture and the RPi community had to roll their own. That speaks to the original point of this subthread, I believe. Not only does Linux not offer the "exact same experience across the board," it cannot even do it for the ARM architecture. For decent enough reasons, really, when you think about it.

Different versions of the ARM architecture with different ABIs. From the Debian POV that's a separate architecture to support. It could have been done by Debian, but was done by a third party, just as other minor platforms are supported.

I don't agree that it has a "different experience" because the tools and infrastructure are there to build the entire distribution from scratch. And this was done. I've done it myself several times. Once done, this architecture variant had the complete package set available for all the official supported architectures, bar any architecture-specific packages being added/omitted. Third parties can and do bootstrap and maintain entire architectures. I can't speak for the raspbian people and their port, but it's not hard to manually bootstrap the kernel and toolchain and then set an autobuilder loose on the entire source archive.

And that's the point I was trying to make about Windows; that's exactly what you can't obtain. Be it the old NT ports or the present day ones, outside the base "Windows" release and some key products, the rest of Microsoft's product portfolio is largely missing.

Re: Microsoft Pledges to Use ARM Server Chips, Threatening Intel's Dominance

#160

Post by tgragnato »

foobiekr wrote: post #112

Has it been long enough for people to forget NT on Alpha, MIPS and x86 (and i860, though not released)? And to forget both PReP ( https://en.wikipedia.org/wiki/PowerPC_Reference_Platform ) and CHRP ( https://en.wikipedia.org/wiki/Common_Hardware_Reference_Plat... ) which were going to break the Intel stranglehold once and for all?

I suddenly feel too young
Post reply on HN