ARM based Opterons incoming!

[Vesku]

I wouldn't use Calxeda as a bellwhether for ARM servers. Imo, they just entered into the field too early with 32 bit ARM and a mediocre memory system. Part of Calxeda's PR blitz was to drum up investment so they could survive to produce a 64 bit system with better memory density.

I fully expect at least Nvidia and Qualcomm to offer some server focused 64 bit ARM products.

 

[Cerb]

Is that the only choice you can think?

No. I think it is the most obvious choice Intel could make that would make all these potential competitors irrelevant; but it would lead to Intel becoming like Sony, or Samsung, over time; and I don't know that another company would be able or willing to try to be such an innovative R&D monster. Intel and nVidia very much stand out, as companies consistently able to keep volumes and margins high enough to perform tons and tons of R&D, to make better products that allow them to command price premiums. if Intel went away from that, would any other company be willing to step up and take the huge risks they do? With AMD in its current stae, I'm not so sure.

Drop margins should NEVER be your first choice. There's a lot of things that should come first, like Intel going for a far denser manufacturing node or using synthesis to make chip design cheaper. Intel is pursuing exactly that route.

Except that once it gets small enough, it won't be any cheaper. The cost of developing and using the smaller node, and packaging those teeny tiny chips, will come to eclipse the savings, if the product prices were to have to come down.

The better thing to do, IMO, would be for Intel to continue to make better products, that they can command high enough prices for to keep R&D up, to keep making them in the future. But, by doing so, in any market but Windows-based PCs, the door stays open for cheaper competition, offering value products. In doing so, however, those smaller companies are not necessarily doomed, and, as bonus for us, Intel will have continued need to improve what they sell.

 

[Arachnotronic]

Ashraf, you're smarter than that: don't use SPEC to compare Intel (or any company using their own compilers) vs others. Icc is overtuned for SPEC. It typically is about 10-20% over gcc on SPEC, while on most real non vectorizable code it's hardly faster if not slower, give icc a try it's free on Linux and compare against gcc. Just don't be afraid by the number of internal errors icc will spit at you

IMHO the ISA is not an issue as this kind of chip will probably be used for servers running open source code.

But I certainly agree the TDP is very disappointing and the chip doesn't look good enough to switch away from x86. Just like Silvermont isn't good enough to switch away from ARM

Silvermont is a good CPU core for mobiles, I daresay rivaled only by Apple's Cyclone. The problem is that the rest of the SoC isn't quite there.

4 EU Gen 7 was lousy for the tablet market, and Merrifield will lack the integration needed to really win in phones. But I think Silvermont, around a better uncore, would have been a very potent threat in the mobile market.

C'est la vie

 

[USER8000]

Ashraf, you're smarter than that: don't use SPEC to compare Intel (or any company using their own compilers) vs others. Icc is overtuned for SPEC. It typically is about 10-20% over gcc on SPEC, while on most real non vectorizable code it's hardly faster if not slower, give icc a try it's free on Linux and compare against gcc. Just don't be afraid by the number of internal errors icc will spit at you

IMHO the ISA is not an issue as this kind of chip will probably be used for servers running open source code.

But I certainly agree the TDP is very disappointing and the chip doesn't look good enough to switch away from x86. Just like Silvermont isn't good enough to switch away from ARM

Wait he is the bloke who writes some articles for certain sites? Okaaaaaaaaaaaaay.

 

[beginner99]

I don't think it is going to make AMD any money.

Anything made at GF will at least help AMD with their WSA.

 

[Nothingness]

Silvermont is a good CPU core for mobiles, I daresay rivaled only by Apple's Cyclone. The problem is that the rest of the SoC isn't quite there.

4 EU Gen 7 was lousy for the tablet market, and Merrifield will lack the integration needed to really win in phones. But I think Silvermont, around a better uncore, would have been a very potent threat in the mobile market.

C'est la vie

Yes, that was my point: the core is not the whole story, and IMHO this applies to these micro server chips.

@USER8000: Ashraf is able to change his mind based on facts, he's not blinded by the admiration he has for Intel; if Intel underperforms he just says so. How many of us are able to turn down what we like a lot?

 

[NTMBK]

Ashraf, you're smarter than that[...]

Wait, Ashraf? As in Ashraf Eassa from Seeking Alpha? I always thought that guy was pro-Intel, but wow...

 

[meloz]

This could have been a game changer, but due to poor execution AMD are delivering yet another abortion.

Look at the utterly non-existent OS support for ARM's Cortex A57. The Android situation should tell you how ARM vendors expect users to buy new hardware everytime they need to upgrade. Server customers do not have time for that kind of nonsense. Major Linux OS like Debian et al do not have compiled Cortex A57 images yet, and I have seen nothing in pipeline to suggest this will change anytime soon. RHEL have been cooking something forever, still nothing concrete.

So which Linux distro do they expect server customers to install, and how? Gentoo from source? Only possible customers for this 'solution' might be big organizations like Facebook and Google, people who have enough in-house capability to compile and maintain their own Linux kernel based OS.

The reason x86 -and AMD64- have been so succesful in the server space has a lot to do with how good the compiler and kernel support has been. You can install any Debian/CentOS/Ubuntu Server image tagged 'AMD64' on any Xeon and be guranteed to get a stable working OS. ARM vendors poor record in this area will hurt them more than they realize. Their shenanigans with Android have not gone unnoticed.

I am also doubtful about the "1/10th of Xeon" solution cost they are promising. Sounds very cagey. In any case Intel can always make affected Xeons competitive by cutting prices. These type of server solutions sell on performance per watt-$, so Intel only have to change the selling price of their Xeons to become competitive again. And Intel have the proft margin to cut prices whereas AMD does not.

 

[Vesku]

I am also doubtful about the "1/10th of Xeon" solution cost they are promising. Sounds very cagey. In any case Intel can always make affected Xeons competitive by cutting prices. These type of server solutions sell on performance per watt-$, so Intel only have to change the selling price of their Xeons to become competitive again. And Intel have the proft margin to cut prices whereas AMD does not.

I'd expect this ARM Opteron to cost ~$50-60. Doubt the full system price difference will be anywhere close to 1:10.

 

[NTMBK]

Look at the utterly non-existent OS support for ARM's Cortex A57. The Android situation should tell you how ARM vendors expect users to buy new hardware everytime they need to upgrade. Server customers do not have time for that kind of nonsense. Major Linux OS like Debian et al do not have compiled Cortex A57 images yet, and I have seen nothing in pipeline to suggest this will change anytime soon. RHEL have been cooking something forever, still nothing concrete.

I guess you just missed the story on the front page about the new ARM server platform standard? http://www.anandtech.com/show/7721/arm-and-partners-deliver-first-arm-server-platform-standard

The new specification is supported by a very broad range of companies ranging from software companies such as Canonical, Citrix, Linaro, Microsoft, Red Hat and SUSE, OEMs (Dell and HP) and the most important component vendors active in this field such as AMD, Cavium, Applied Micro and Texas Instruments. In fact, the Opteron A1100 that was just announced adheres to this new spec.

OS makers will be able to provide a single image that works with this platform, and the ARM Opteron will run it just fine.

 

[Abwx]

I'd expect this ARM Opteron to cost ~$50-60. Doubt the full system price difference will be anywhere close to 1:10.

Methink you are off by a 2 ratio..100-120$ is more likely given
the eventual competition prices, the chip should be in the 120mm2.

 

[Vesku]

Methink you are off by a 2 ratio..100-120$ is more likely given
the eventual competition prices, the chip should be in the 120mm2.

That would make getting near a 1:10 price advantage even less likely. AMD should still be able to compete on price but their initial PR is a bit too pie in the sky, imo.

 

[Nothingness]

Look at the utterly non-existent OS support for ARM's Cortex A57. The Android situation should tell you how ARM vendors expect users to buy new hardware everytime they need to upgrade. Server customers do not have time for that kind of nonsense. Major Linux OS like Debian et al do not have compiled Cortex A57 images yet, and I have seen nothing in pipeline to suggest this will change anytime soon. RHEL have been cooking something forever, still nothing concrete.

You mean you really missed SUSE, Fedora, Ubuntu etc? Debian is also working on Aarch64. Linux has already been demoed on Aarch64 hardware more than 6 months ago. And as NTMBK already mentioned, there is standardization going on to ensure the mess that happened in the mobile SoC market doesn't repeat itself.

The reason x86 -and AMD64- have been so succesful in the server space has a lot to do with how good the compiler and kernel support has been. You can install any Debian/CentOS/Ubuntu Server image tagged 'AMD64' on any Xeon and be guranteed to get a stable working OS.

Yes and as soon as x86 started attacking the market server, firmwares, compilers and OS were perfect for the target market

 

[meloz]

I guess you just missed the story on the front page about the new ARM server platform standard? http://www.anandtech.com/show/7721/arm-and-partners-deliver-first-arm-server-platform-standard



OS makers will be able to provide a single image that works with this platform, and the ARM Opteron will run it just fine.

Vaporware. 'Standard' does not mean a working product. If you have been involved in using Linux and following this ARM farce you would know very well they have been promising great things for over two years and nothing has changed to date. This is just another PR fluff. We get one such promise every 2-3 months, depending on which party needs media attention.

Also, standards only have meaning if people adhere to them. Most ARM vendors business model depends on breaking the ability to upgrade, thus forcing the users to buy a new SoC. I feel for AMD here, because they might be waging a lost war single-handedly whilst their own "partners" at ARM undermine them. AMD's goal -and goal of everyone else in ARM space, including ARM- are quiet different.

When you have a link where people can download Debian or Ubuntu Server ISO image for ARM Cortex, then we will talk. Then Intel, too, will take notice. Until then 99.9% of the market won't even look at this ARM-AMD solution. I want these type of non-Intel solutions to succeed but the way these people are going about it Intel must be laughing their heads off. I want to tear my hair out in frustration, so much potential going wasted.

 

[ancientarcher]

AMD is many years and billions of dollars behind Apple, Qualcomm, Samsung, and NVIDIA when it comes to development of a fully custom ARM CPU. That is why their slides talk about 2019. Note that it takes about 5 years to develop a fully custom CPU. In the meantime, these other companies with better foresight from some years past will be marching ahead in the ARM space (while Intel will be marching ahead in the x86 space).

WRONG!
You neither need billions of $ nor 5 years to develop a new custom built ARM core.

http://semiaccurate.com/2013/05/15/amds-andrew-feldman-talks-about-arm/
"By Mr. Feldman’s count it takes 400 million dollars and three years to build an x86 server chip. But only 30 Million dollars and 18 months are required to make an ARM chip. As far as AMD is concerned the ARM ecosystem offers massive cost and time-to-market advantages over building traditional x86 chips."

AMD didn't release this ARM chip because they were tired of not getting any market share with their x86 versions. It is a real advantage in time to market, cost and possibly margins.

Apple developed the cyclone core in 2 years, starting when ARM announced the v8 arch, and did the uncore in that time as well. If anything AMD has a much more seasoned CPU design team and the uncore already done. Just because all existing ARM chips are low power/low performance doesn't necessarily mean all ARM chips will confirm to that spec. Expect to see high power/high performance ARM chips from AMD, AMCC etc. The only difference vs Intel - well the cost, of course... and maybe the power consumption. But in this market, cost matters.

 

[AtenRa]

So which Linux distro do they expect server customers to install, and how? Gentoo from source? Only possible customers for this 'solution' might be big organizations like Facebook and Google, people who have enough in-house capability to compile and maintain their own Linux kernel based OS.

Ubuntu, RedHat and SUSE will support ARM-64.
Also, Oracle, Java and GNU C has been ported to ARM-64

Ubuntu Server

What’s new in 13.10

Juju Charms

• Local provider: spin up a full application suite including all supporting services on your machine
• Manual provisioning: you can now use pre-existing machines — no need for a cloud account
• Manage Juju environments from Linux, OS X and Windows
• Automatic provisioning support for Azure

OpenStack

• Havana built for 12.04 LTS, available through Cloud Archive
• Introducing support for KVM Virtualization in ARM A15, and experimental ARM A15 support in Xen 4.3
• Fast new installer — up to five times faster

 

[Abwx]

That would make getting near a 1:10 price advantage even less likely. AMD should still be able to compete on price but their initial PR is a bit too pie in the sky, imo.

They surely took account of whole systems over time,
hardware, software, servicing and other operating costs.

 

[mrmt]

I am also doubtful about the "1/10th of Xeon" solution cost they are promising.

That's a high end Xeon, and didn't specify the line (E, EP, EX...). Mr. Feldman also did not address how does the price compares against Avoton, you know, the ones you can buy right now if you want for a price far smaller than a conventional Xeon, let alone a high end Xeon and will be replaced by the 14nm Denverton in Q414.

 

[meloz]

Ubuntu, RedHat and SUSE will support ARM-64.
Also, Oracle, Java and GNU C has been ported to ARM-64

Ubuntu Server

And where's the link to the AArch64 ISO?

There isn't.

That's the point, all talk no substance. AArch64 was announded in late 2012, till date none of the major OS like Debian, Ubuntu Server and CentOS have a link on their website where you can download an ISO. This is history repeating itself, same nonsense we heard when Calxeda promised their revolution.

The closest thing to working ARM support in mainstream linux OS are the OMAP4 images for Ubuntu Server. I wonder what happened to any customers who bought into ARM and are running "OMPA4" Ubuntu Server 13.10, they must be loving their dead-end! I am sure they will all rally around to get raped a second time with AArch64.

With enemies like these, Intel do not need any friends. No wonder they charge so exorbitantly for Xeons.

 

[mrmt]

AMD didn't release this ARM chip because they were tired of not getting any market share with their x86 versions. It is a real advantage in time to market, cost and possibly margins.

AMD released an ARM chip because their Bulldozer replacement would be years by the time they committed to ARM, and Bobcat didn't have a roadmap strong enough to take on the new Atom, so instead to pick a fight they couldn't, they decided to try ARM, where the field will be more or less leveled despite the low margins and of the big guys developing custom cores.

And Mr. Feldman is tapdancing a lot. Sure, ARM chips costs less to develop, but there are some caveats: TAM is smaller, margins are low, you don't get rid of your x86 competition and the IP is open to mediateks to sell a cheaper competition to your product.

 

[meloz]

That's a high end Xeon, and didn't specify the line (E, EP, EX...). Mr. Feldman also did not address how does the price compares against Avoton, you know, the ones you can buy right now if you want for a price far smaller than a conventional Xeon, let alone a high end Xeon and will be replaced by the 14nm Denverton in Q414.

I see your point.

AMD are promising 1/10th solution cost, but if this reduction in cost is accompanied by a reduction in overall computing capacity, then it is no true saving. Customers will have to buy more of AMD solutions to get similar capacity as they have with their Xeons (whichever hypothetical Xeons these might be).

Improvements in performance per watt per $ are only impressive if they do not result in a reduction of capacity. Otherwise any comparision is meaningless.

 

[ancientarcher]

Sure, ARM chips costs less to develop, but there are some caveats: TAM is smaller, margins are low, you don't get rid of your x86 competition and the IP is open to mediateks to sell a cheaper competition to your product.

Yes, and it is going to be this way (lots of competition, lower than monopoly margins...) whether or not AMD participated in this market. At least, AMD has the uncore done for the server parts. Bring on HSA later with GCN and see what cooks up.

Talking of cooking up, what will it take to build a high performance ARM core with only AARCH64 compatibility. Clearly, it will be far easier (and faster and cheaper) to build this than on x86 where you have to also ensure compatibility for 8086, 286, SMM, 386, x64 and god knows what else... That's what i am looking to AMD for when they come out with their own ARM core.

 

[Abwx]

Clearly, it will be far easier (and faster and cheaper) to build this than on x86 where you have to also ensure compatibility for 8086, 286, SMM, 386, x64 and god knows what else...

Last time i checked it was about 1200 instructions , we re surely
above 1500 currently if not close to 2000 , and it s not over ,
even more are expected in the coming years.

 

[Vesku]

And where's the link to the AArch64 ISO?

There isn't.

That's the point, all talk no substance. AArch64 was announded in late 2012, till date none of the major OS like Debian, Ubuntu Server and CentOS have a link on their website where you can download an ISO. This is history repeating itself, same nonsense we heard when Calxeda promised their revolution.

The closest thing to working ARM support in mainstream linux OS are the OMAP4 images for Ubuntu Server. I wonder what happened to any customers who bought into ARM and are running "OMPA4" Ubuntu Server 13.10, they must be loving their dead-end! I am sure they will all rally around to get raped a second time with AArch64.

With enemies like these, Intel do not need any friends. No wonder they charge so exorbitantly for Xeons.

You probably have to dig around to find AArch64 stuff. Why would they have a direct consumer link to ISOs when there isn't any hardware available yet to non-developers?

 

[NTMBK]

And where's the link to the AArch64 ISO?

There isn't.

That's the point, all talk no substance. AArch64 was announded in late 2012, till date none of the major OS like Debian, Ubuntu Server and CentOS have a link on their website where you can download an ISO. This is history repeating itself, same nonsense we heard when Calxeda promised their revolution.

The closest thing to working ARM support in mainstream linux OS are the OMAP4 images for Ubuntu Server. I wonder what happened to any customers who bought into ARM and are running "OMPA4" Ubuntu Server 13.10, they must be loving their dead-end! I am sure they will all rally around to get raped a second time with AArch64.

With enemies like these, Intel do not need any friends. No wonder they charge so exorbitantly for Xeons.

The first ARM Server Platform hardware isn't out yet; why would they put up ISOs for a platform which isn't yet released? I expect that they will show up at the same time as Seattle.