AMD Expected to Unveil ARM-Based Server on Monday

CommanderFrank

Cat Can't Scratch It
Joined
May 9, 2000
Messages
75,399
Amid speculation of the ‘doom and gloom’ rumors concerning the nature of AMD’s announced news conference on Monday, informed sources have said that the news conference will be to announce the plans for an ARM-based SeaMicro 64-bit processor design server.

It probably wouldn't be the only company with such a product. Hewlett-Packard and Dell are just two of the vendors that are also pursuing ARM-based servers, as the industry looks to reduce power consumption in data centers.
 
I understand the want to reduce energy use...but in doing so aren't you also greatly reducing processing power? I mean, only way I see this as being a good thing is if companies are going to upgrade their old "fast enough" servers to something equally as fast but ARM based.

No ARM CPU anything that I know of can equal the power of a hex-core Xeon.
 
No ARM CPU anything that I know of can equal the power of a hex-core Xeon.

A lot of servers out there don't need a lot of processing power, unless they are running CPU-heavy services of course.
A lot of CPUs simply sit idle, sapping a lot of power for no reason when most of the services running are more memory or disk bound.

I could think of quite a few servers that could be replaced by these for a huge drop in power consumption, but then again, the programs and daemons would need to be compiled for ARM as well... so this may be a ways out.
 
A lot of servers out there don't need a lot of processing power, unless they are running CPU-heavy services of course.
A lot of CPUs simply sit idle, sapping a lot of power for no reason when most of the services running are more memory or disk bound.

I could think of quite a few servers that could be replaced by these for a huge drop in power consumption, but then again, the programs and daemons would need to be compiled for ARM as well... so this may be a ways out.

Virtualization already accomplishes this. Just share the resources on one machine.
 
Virtualization already accomplishes this. Just share the resources on one machine.

You could, but what if ALL your servers didn't need a lot of CPU power? Then you could either virtualize on a single platform that consumes 300 watts or buy maybe five platforms that only use 200 total. Virtualization is great, but it doesn't fit in every single situation. Sometimes, having dedicated physical hardware is a better idea and this will fill the problem of needing stuff that is really power efficient and dedicated to a task.
 
You can't fault AMD for trying. lol

There certainly are some specialized, but small niches, which this could be successful in. Is it really worth it to develop the hardware and ARM-optimized software stacks? Imma thinking break even for early vendors is a little optimistic (and worse for straggling vendors, if there are any).
 
Virtualization already accomplishes this. Just share the resources on one machine.

What if there are only a few services that need to be run?
What I'm saying is that for systems that don't need to have full x86 processors, let alone virtualization, could do to benefit by moving to ARM-based servers simply do to their lower power and sufficient processing power.
 
You could, but what if ALL your servers didn't need a lot of CPU power? Then you could either virtualize on a single platform that consumes 300 watts or buy maybe five platforms that only use 200 total.
That's not what a commercial ARM server platform is about. It's about core count density at lower power. The thing is that more cores + low power does not necessarily mean efficiency (as in performance/watt... see Calxeda's mega ARM server losing to plain old Xeon in performance and performance/Watt). There are some (few) applications where core count is more important than overall performance or "latency" when completing tasks.

Lower core count ARM servers may certainly be adequate for low load servers, but I doubt many sysadmins will choose to get several low core count ARM servers over a single multi-socket faster server.

Also, AMD will be moving its low end, low power Hondo/Temash or whatever cores into the same space to compete against high density Atom-based servers and Xeon Phi.

Lastly, for the A15 to be cranked up in performance to compete against other platforms, the power consumption goes much, much higher than a phone CPU core. Nothing is free.
 
What if there are only a few services that need to be run?
What I'm saying is that for systems that don't need to have full x86 processors, let alone virtualization, could do to benefit by moving to ARM-based servers simply do to their lower power and sufficient processing power.

You create a VPS container and allow it to use one core and however much memory and storage space it needs, that core is still shared amongst the other containers. If this container isn't doing anything, it uses literally no power.
 
That's not what a commercial ARM server platform is about. It's about core count density at lower power. The thing is that more cores + low power does not necessarily mean efficiency (as in performance/watt... see Calxeda's mega ARM server losing to plain old Xeon in performance and performance/Watt). There are some (few) applications where core count is more important than overall performance or "latency" when completing tasks.

Lower core count ARM servers may certainly be adequate for low load servers, but I doubt many sysadmins will choose to get several low core count ARM servers over a single multi-socket faster server.

Also, AMD will be moving its low end, low power Hondo/Temash or whatever cores into the same space to compete against high density Atom-based servers and Xeon Phi.

Lastly, for the A15 to be cranked up in performance to compete against other platforms, the power consumption goes much, much higher than a phone CPU core. Nothing is free.

Efficiency and calculations per watt consumed aren't that important if you're using almost none of it to begin with. Xeons might be a lot more efficient in that respect, but total power consumed is sometimes more important. Why should I stick any Xeon into a file server that hosts data for a workgroup of ten people? A single socket Xeon server might respond better and even be more efficient, but if the server is doing almost nothing all the time, then it really can be beneficial to deal with a phone CPU which, even without ramping up performance, can certainly handle the limited demands placed on it and, despite being more heavily loaded when asked to do something, is still cheaper to feed electricity. :)
 
You're not understanding the purpose of the ARM "server" chip. And you don't seem to understand server offerings (hint: look at Dell's or HP's 1P server page... the low end uses desktop AMD or Intel processors and low end server or desktop chipsets). And the "server" A15 is not a power constrained device like models used in portable electronics. Anyways, this is not about a hypothetical* small server which uses 5W-10W under load, although it would be possible to make one.

The models HP and Dell are investigating and which Calxeda is selling and now SeaMicro is announcing, are all high core count/high density servers (many dozens to hundreds of cores). As cool as an enthusiast thinks a very low power, low socket count server could be, it has virtually no applications in the market which would offset development costs**, competing against low end servers which use desktop processors. The x86 servers would run rings around an ultra low power, low core count toy ARM server. A15 trying to get a piece of the very profitable segment in the server market, not to dwell in the low end ghetto. Why? Because each ARM core is relatively weak, even a newest and greatest A15 running at a cranked up clock speed. Read: it's uncompetitive in that configuration for anything performance sensitive.

* I haven't seen any commercial low socket count A8, A9 or A15 based servers announced or hinted at. Do you have a source for anything but hobbyist platforms or specialized embedded devices? Please share.

** basic utility appliances could contain an ARM-based processor, or MIPS-based or some esoteric model... it makes no difference. However, those items usually have integration problems if consistent security management is important and other management features just don't exist. Remember the time of the general linux server taking over offices? lol Masochists love them on Windows networks and most sane people give them a pass.
 
You create a VPS container and allow it to use one core and however much memory and storage space it needs, that core is still shared amongst the other containers. If this container isn't doing anything, it uses literally no power.

I don't really think you understand how processors work, but whatever, thanks for the tip. :p
 
You're not understanding the purpose of the ARM "server" chip. And you don't seem to understand server offerings (hint: look at Dell's or HP's 1P server page... the low end uses desktop AMD or Intel processors and low end server or desktop chipsets). And the "server" A15 is not a power constrained device like models used in portable electronics. Anyways, this is not about a hypothetical* small server which uses 5W-10W under load, although it would be possible to make one.

The models HP and Dell are investigating and which Calxeda is selling and now SeaMicro is announcing, are all high core count/high density servers (many dozens to hundreds of cores). As cool as an enthusiast thinks a very low power, low socket count server could be, it has virtually no applications in the market which would offset development costs**, competing against low end servers which use desktop processors. The x86 servers would run rings around an ultra low power, low core count toy ARM server. A15 trying to get a piece of the very profitable segment in the server market, not to dwell in the low end ghetto. Why? Because each ARM core is relatively weak, even a newest and greatest A15 running at a cranked up clock speed. Read: it's uncompetitive in that configuration for anything performance sensitive.

* I haven't seen any commercial low socket count A8, A9 or A15 based servers announced or hinted at. Do you have a source for anything but hobbyist platforms or specialized embedded devices? Please share.

** basic utility appliances could contain an ARM-based processor, or MIPS-based or some esoteric model... it makes no difference. However, those items usually have integration problems if consistent security management is important and other management features just don't exist. Remember the time of the general linux server taking over offices? lol Masochists love them on Windows networks and most sane people give them a pass.

I'm pretty sure that no one would bother doing it if it wasn't both technologically possible and potentially profitable. This sort of thing would not have gotten to the point where it is now if there wasn't some means of leveraging the advantages of a A15 processors in a manner that was optimal for certain workloads over conventional x86 servers.

Aside from the RPi, nothing for the hobbyist comes to mind. There are development kits containing ARM boards which you could probably build into a small server or cluster, but I don't think they're practical yet. This AMD effort is a top first approach so we're not going to see low core count ARM servers for a little while. Like you've said, there's a derth of both hardware and enterprise-worthy software. I'm pretty sure there's a market for them though. There are a lot of applications in which a backend system is necessary, but isn't under any particular demand and doesn't warrant the added burden of platform virtualization (though, I totally admit that VMWare's stuff is pretty much stupid-proof).
 
Back
Top