AMD unveils ARM Opteron a1100 chip

I'll take one development kit please. Address sent separately.
 
Please note these are dev kits, AMD has said that they want to put 10Gbps ethernet native on Opterons in 2015.

So if you want one for production use then your best bet is waiting.
 
Nonetheless, it seems like an interesting product. I'd definitely be up for some 64bit A57 action. This is definitely not your daddies Moonshot ARM card, this is a real and potentially supported ARM server model. Hmmm.
 
Yeah max 12W and the same Apache performance of an Intel Xeon.

1Gbps ethernet is so easy to max out we don't need high powered web servers.

Hopefully this will bring down the cost of web hosting.
 
Anandtech has a longer article. The performance looks a bit less impressive now, at least with the 8c 25W Opteron A1100 vs last year's 8c 20W Atom C2750 with the benchmark (estimate) AMD released so far. The Atom is 25% faster and uses 20% less power. Ouch. B-b-b-b-but ARM is supposed to be lower power! :p
 
pxc said:
Anandtech has a longer article. The performance looks a bit less impressive now, at least with the 8c 25W Opteron A1100 vs last year's 8c 20W Atom C2750 with the benchmark (estimate) AMD released so far. The Atom is 25% faster and uses 20% less power. Ouch. B-b-b-b-but ARM is supposed to be lower power! :p
Click to expand...

Where are you seeing that?
 
pxc said:
Anandtech has a longer article. The performance looks a bit less impressive now, at least with the 8c 25W Opteron A1100 vs last year's 8c 20W Atom C2750 with the benchmark (estimate) AMD released so far. The Atom is 25% faster and uses 20% less power. Ouch. B-b-b-b-but ARM is supposed to be lower power! :p
Click to expand...

Spazturtle said:
Where are you seeing that?
Click to expand...

I think he is high on crack as no real world performance numbers have been released and they just announced the dev kit board this afternoon to is not shipping till march. long story short he failed to RTFA.
 
Spazturtle said:
Where are you seeing that?
Click to expand...
Better question, where am I not seeing that? Surprise me! :p

When AMD released the quad core Jaguar-based server CPU, it was more than happy to compare the performance to the Centerton Atom, which has its heritage in the crappy 5 year old Silverthorne in-order execution core. I was surprised that AMD didn't compare what appears to be somewhat decent SPECint_rate performance* to what Intel has been shipping since last year. So I just checked Avoton Atom's performance for comparison. http://www.intel.com/content/www/us/en/benchmarks/atom/atom-microserver-specint-rate-base-2006.html

Easy to compare. It doesn't really get better. 106 is actually 32% higher than 80 and 25W is 25% higher than 20W. Forgive my lazy math above. I should have put the bigger numbers first in the comparison.

For the curious of how an 80 score in SPECint_rate stacks up, a dual core i3-2100 scores a bit higher (this was the very first score listed in a search for 'intel core i3'): http://spec.org/cpu2006/results/res2012q2/cpu2006-20120426-21390.html Viva la ARM revolution!

* no one reports ARM spec scores, at least not for SPECint 2006. Nvidia and others usually release SPEC CPU 2000 scores, which are considered obsolete.
 
Last edited:
CPU speed is rarely a factor in web servers, the speed of a web server takes into consideration much more then just CPU speed. You have to look at all the factors. And we don't even have CPU speed figures yet.

Also this chips has 2 * 10Gbps ethernet controller built in. You are not factoring the wattage for the separate ethernet controller needed for the x86 chip in. Your looking at another 10W for the ethernet controller (going by Intel's own specs).

The main issue with web servers is that they have CPUs that are too fast for any use and use too much wattage, and they only come with 1Gbps ethernet.
 
Spazturtle said:
Also this chips has 2 * 10Gbps ethernet controller built in. You are not factoring the wattage for the separate ethernet controller needed for the x86 chip in. Your looking at another 10W for the ethernet controller (going by Intel's own specs).
Click to expand...
Avoton (the Atom C2750 I linked above) has 10GbE, if the switch supports that standard. If a second 10GbE port is needed, it might use 10W if it's installed via a PCIe card, but another controller chip installed on a motherboard isn't going to use 10W. If 1GbE (or 2.5GbE) is used, Avoton supports 4 of those.

By the time the Opteron A1100 comes out, the next gen Atom for servers will also be out and it's unlikely that 10GbE will not be included.

If all Opteron A1100 has going for it is a second 10GbE port, um...

lol, the faithful.
 
Spazturtle said:
Yeah max 12W and the same Apache performance of an Intel Xeon.

1Gbps ethernet is so easy to max out we don't need high powered web servers.

Hopefully this will bring down the cost of web hosting.
Click to expand...

Um, webhosting is already penny's a day...

The issue i recall reading a while back that while these are okay for webhostnig and small loads, once you get past webhosting, they really perform poorly compared to a Xeon..will have to dig up that article and testing they did with an HP arm server i think it was.
 
If this is a new server chip I wonder if this means there's a chance of a FX based derivative eventually showing up.
 
Ws60 said:
If this is a new server chip I wonder if this means there's a chance of a FX based derivative eventually showing up.
Click to expand...

Extremely Doubtful. Windows enthusiasts would want an x86 based processor. There would not be enough of a following of linux, bsd ... users for an FX chip.
 
pxc said:
Better question, where am I not seeing that? Surprise me! :p

When AMD released the quad core Jaguar-based server CPU, it was more than happy to compare the performance to the Centerton Atom, which has its heritage in the crappy 5 year old Silverthorne in-order execution core. I was surprised that AMD didn't compare what appears to be somewhat decent SPECint_rate performance* to what Intel has been shipping since last year. So I just checked Avoton Atom's performance for comparison. http://www.intel.com/content/www/us/en/benchmarks/atom/atom-microserver-specint-rate-base-2006.html

Easy to compare. It doesn't really get better. 106 is actually 32% higher than 80 and 25W is 25% higher than 20W. Forgive my lazy math above. I should have put the bigger numbers first in the comparison.

For the curious of how an 80 score in SPECint_rate stacks up, a dual core i3-2100 scores a bit higher (this was the very first score listed in a search for 'intel core i3'): http://spec.org/cpu2006/results/res2012q2/cpu2006-20120426-21390.html Viva la ARM revolution!

* no one reports ARM spec scores, at least not for SPECint 2006. Nvidia and others usually release SPEC CPU 2000 scores, which are considered obsolete.
Click to expand...

so you are comparing a not released arm v8 64bit chip to a intel product that has been released and the previous amd release mmmk its a different microarchitecture your point is invalid
 
maxius said:
so you are comparing a not released arm v8 64bit chip to a intel product that has been released and the previous amd release mmmk its a different microarchitecture your point is invalid
Click to expand...

I kinda have to agree on this one. I'm very impressed that this is an arm processor. Considering that, I think the a1100 looks very promising, I wish it would use a tad less power but freaking TWO 10gbps ports is fucking NICE and as the tech matures I'm sure the speed and power draw will improve as well. x86 servers are not in the least bit exciting to me, good on AMD for switching it up and moving forward. Here's hoping the price is reasonable compared to the atom counterpart. I'd bet it is... Definitely will be paying attention if/when this go's retail.
 
Last edited:
Microsoft should consider making full Windows for ARM if they want to stay is server OS business. Wouldn't mind full OS for end users. It would open up possibilities.
 
I would be on board for sure. The damn things just keep getting better and better. There really isn't a good reason for them not to. I'd bet they are developing something somewhere. That's just my opinion though.
 
Another competitive gem in the SPEC submissions: low end dual core Pentium G3420 also gets around the same score as that 8c Opteron. http://spec.org/cpu2006/results/res2013q4/cpu2006-20131104-27176.html Ouch again.

maxius said:
so you are comparing a not released arm v8 64bit chip to a intel product that has been released and the previous amd release mmmk its a different microarchitecture your point is invalid
Click to expand...
wut?

AMD has only given one benchmark number so far, which is uniformly used to measure CPU performance in that little market AMD hopes to compete in (servers).

SPEC is a cross platform/cross architecture suite of benchmarks and is the standard for measuring CPU performance in many different segments.

I'm not responsible for someone not understanding what's going on. I wouldn't normally have seen that post because lol ignore list works.
 
omega-x said:
I wonder how well this would perform scaled to 95W
Click to expand...
The A57 design spec is for 2.5GHz-ish clock speed on a 20nm process according to ARM, although it may hit close to that clock speed on TSMC's 28nm LP process. AMD is using a stock licensed A57 core, but performance may be a bit better in certain tasks than other licensed A57-based SoCs if AMD's memory controllers are superior.

If AMD had an architecture license (like Apple, Qualcomm, Samsung and others), and an ARM design team, it could stretch the clock speed and improve IPC, but AMD doesn't have those things. Even it hired a quality team today, the fruits of that labor would take years to emerge (see Nvidia and Project Denver).

There's really no point to making a 95W version, since trying to scale up clock speed would make it very inefficient, and packing more cores on it brings diminishing returns. AMD certainly had carefully weighed what it could control when deciding how many cores it could optimally include per node, and what speeds/voltage it could run the A57 core.

You wouldn't want this on a desktop. Performance is pretty crappy at the speeds a licensed A57 core could possibly run. Single threaded performance is atrocious, and Kaveri already provides performance in the same range while providing significantly better single threaded performance.
 
Just a hunch but: AMD is at a point with ARM where they are late to the game and catching up rather well. Their long term moves are probably aimed at consumer applications, and where the A57 might be unfit right now, vs jaguar or SR or even an i3 dual, with ARM outpacing x86 sales, it's where code-monkeys will migrate to, and in 10 years time AMD could be fabbing high end ARM APU alongside a dying breed of x86 chips.
 
omega-x said:
Just a hunch but: AMD is at a point with ARM where they are late to the game and catching up rather well. Their long term moves are probably aimed at consumer applications, and where the A57 might be unfit right now, vs jaguar or SR or even an i3 dual, with ARM outpacing x86 sales, it's where code-monkeys will migrate to, and in 10 years time AMD could be fabbing high end ARM APU alongside a dying breed of x86 chips.
Click to expand...

This may very well be the case for the future of computing. X86 as an instruction is old, inefficient and you can talk to most if not all programmers and they will tell you similar things told different ways. Even my programmer friend knows about X86 and how behind and inefficient the instruction set is.

As we move towards more mobile devices, our future computers will be the Google Glass of the 24th century, the iPhone/Galaxy S of the future, and a tablet with a 16K advanced OLED display. :p And, we are getting there inch by inch, year by year.

Our servers are going to get smaller, powerful and more efficient, and ARM might be the direction to head into instead of POWER or X86-based Opterons/Xeons.
 
pxc said:
The A57 design spec is for 2.5GHz-ish clock speed on a 20nm process according to ARM, although it may hit close to that clock speed on TSMC's 28nm LP process. AMD is using a stock licensed A57 core, but performance may be a bit better in certain tasks than other licensed A57-based SoCs if AMD's memory controllers are superior.

If AMD had an architecture license (like Apple, Qualcomm, Samsung and others), and an ARM design team, it could stretch the clock speed and improve IPC, but AMD doesn't have those things. Even it hired a quality team today, the fruits of that labor would take years to emerge (see Nvidia and Project Denver).
Click to expand...

This feels like a foot in the door release to get their name out there along with the other vendors hitting the streets with their AArch64 chips this year. Given the limited time between AMD announcing they will be developing ARM products and the announcement that they have products in hand I think these will be built completely from vanilla designs, even the memory controllers can be taken off the shelf and probably have been: http://arm.com/products/system-ip/memory-controllers/corelink-dmc-520.php. I also believe they are using GLF 28nm to source these as that has been stated in a couple of articles covering the event although the offical PR did not say specifically, that could be a good or bad thing depending on how far behind GLF really is on sub 28nm processes and if AMD can easily port future/current designs to other foundries.

Beyond the actual product which itself isn't very exciting spec-wise, I think they are trying to sell the platform as a "secure" way to get into ARM-based microservers and at least test the waters. It's secure in the sense that because it's an open platform that is being supported by heavyweights it will be around for awhile with meaningful development and upgrades. The first goal will be to create a customer base with a solid platform, then they can begin to really invest in products to sell to that base.

As far as an ARM design teams go, they re-hired this guy a year and a half ago and he seems to have known a thing or two about creating winning custom ARM designs. I would expect to see more fruits of his knowledge and leadership with the successor to these A1100 chips. ARM doesn't make public their list of architectural licensees specifically, but AMD is listed as being a licensee on their site and if they have any desire to meet the lofty goals they just presented I would have to assume they bought into that top-tier.
 
omega-x said:
Just a hunch but: AMD is at a point with ARM where they are late to the game and catching up rather well.
Click to expand...
AMD is on pace to release Cortex-A57 based server chips with other manufacturers which also licensed a standard core. There are couple of ARM architecture licensees which have announced server chips too (Qualcomm's 3GHz 16nm model and Samsung stating it will enter the server market), which will have significant benefits over a plain licensed core in CPU performance. Then there's the literally dozens of companies which assemble ARM SoCs from licensed blocks, any of which could easily enter the same market and match features and performance point for point.
 
Tup3x said:
Microsoft should consider making full Windows for ARM if they want to stay is server OS business. Wouldn't mind full OS for end users. It would open up possibilities.
Click to expand...

well they have a base to work from

http://en.wikipedia.org/wiki/Windows_RT

but then again all your x86 apps woulden't work so it would be a tough transition... like when apple when PPC to x86
 
i'd love one of these for a home server running linux.

the best about it is the common platform standard announced alongside.
 
Tup3x said:
Microsoft should consider making full Windows for ARM if they want to stay is server OS business. Wouldn't mind full OS for end users. It would open up possibilities.
Click to expand...

ARM has it's place in the server market, but beyond VPS and webhosting, so far it doesn't stand a chance vs Xeon and larger loads and for me i use centos / nginx for all my web servers where possible. very few IIS servers and the ones that are host massive backend sites for our work that arm would choke.
 
"Seattle" demo.
https://www.youtube.com/watch?v=03jto1a1oFY

They go over their ambidextrous product road map and talk about Project Skybridge Will have x86 and ARM compute SoC's and is pin compatible with each other. Full HSA support and GCN. Puma+ x86 cores announced. Presentation lasted for 1.5 hours so there is a lot of information to check out.
 
You must log in or register to reply here.
Back
Top