AMD Announces Opteron A1100 Series 64-bit ARM Processors for the Datacenter

[btarunr]

AMD marks a major step towards delivering choice and innovation in the data center with the launch of the AMD Opteron A1100 System-on-Chip (SoC), formerly codenamed "Seattle". Jointly with its software and hardware partners, AMD is accelerating time-to-deployment of ARM-based systems and driving forward ecosystem support for ARM in the data center.

"The ecosystem for ARM in the data center is approaching an inflection point and the addition of AMD's high-performance processor is another strong step forward for customers looking for a data center-class ARM solution," said Scott Aylor, corporate vice president and general manager, Enterprise Solutions, AMD. "The macro trend of convergence between networking, storage and servers is an important catalyst in this evolution. Customers now have access to 64-bit ARM processors from the only silicon provider that also has decades of experience delivering professional enterprise and embedded products."



The AMD Opteron A1100 SoC represents a key milestone for establishing ARM in the data center as well. "The AMD Opteron A1100 processor brings a new choice in scalability across network infrastructure and data centers," said Lakshmi Mandyam, director of server systems and ecosystems, ARM. "AMD brings recognized expertise in the server and embedded markets, making them an ideal partner to deliver a 64-bit ARM processor with the impressive balance of performance and power-efficiency to address an increasingly diverse set of workloads."

The AMD Opteron A1100 Series SoC is the first 64-bit ARM Cortex-A57-based platform from AMD. Utilizing ARM Cortex-A57 processors with high-speed network and storage connectivity and outstanding energy efficiency, the AMD Opteron A1100 Series SoC delivers a balanced total cost of ownership for storage, web and networking workloads.



AMD Opteron A1100 Series SoC specifications:

• Up to eight ARM Cortex-A57 cores with 4MB shared Level 2 and 8MB of shared Level 3 cache
• 2x 64-bit DDR3/DDR4 channels supporting up to 1866 MHz with ECC
• 2x 10Gb Ethernet network connectivity
• 8-lane PCI-Express Gen 3
• 14x SATA 6 Gb/s ports

The AMD Opteron A1100 SoC is powering enterprise-class systems from SoftIron with its Overdrive3000 system for developers, as well as an upcoming lineup of Software Defined Storage solutions taking advantage of the processor's rich feature set.

"The secret of the AMD Opteron A1100 SoC's appeal is not just the cores, it's everything around the cores," said Norman Fraser, CEO of SoftIron. "If you've got an application where you need to move large amounts of data around quickly, you're going to love it."

AMD is also collaborating with Silver Lining Solutions (SLS) to integrate SLS' fabric technology in innovative dense server designs featuring the Opteron A1100 Series, targeted at streaming, web, and storage workloads for cloud and hyperscale datacenters. The SLS Fabric Interconnect incorporates a low-latency, energy efficient 60Gbps switching fabric and is available as a PCI Express expansion card or a standalone ASIC for custom server applications.

"We are very excited about working with AMD to bring power efficient fabric-based computing to market," said Dr. Ping-Kank Hsiung, Managing Director of Silver Lining Systems. "Combining the efficient AMD Opteron A1100 processor with our unique fabric will help drive down costs and power requirements of hyper scale computing and storage."

In addition to silicon innovation, AMD has been instrumental in supporting the 64-bit ARM software ecosystem, a critical component to any new processor, and has been working closely with Enterprise Linux leaders Red Hat and SUSE on operating system and application support.

"Red Hat and AMD share a vision of building an open, standards-based software ecosystem for highly converged designs based on 64-bit ARM architecture. As one of the first participants in Red Hat's ARM Partner Early Access Program, AMD has been instrumental in the testing and porting of the world's leading enterprise Linux platform to 64-bit ARM architecture", said Ranga Rangachari, vice president and general manager, Storage, Red Hat. "The arrival of the AMD Opteron A1100 SoC represents a major milestone to the ecosystem interested in driving converged infrastructure for storage, networking, and compute."

The AMD Opteron A1100 SoC has been in advanced development with technology partners and customers for several quarters and is available in mass production quantities today.

View at TechPowerUp Main Site

 

[geon2k2]

That looks fantastic, there are so many workloads in the industry which can benefit from very high parallelism and for which per core performance is not so much a concern as the system is anyway waiting for filesystem, database or user input most of the time.

Deploying 64 or maybe more cores could create a pretty powerful system.

I wonder though if so many cores can be deployed together and what is the performance of such a core?

 

[jabbadap]

4-core tdp 25W, 8-core tdp 32W, quite high for arm processors. What manufacturing process are these made of(tsmc 28nm?). That integrated I/O is great, I presume there aren't any igp on them so pure server socs?

I would like to know if these are coming to consumer space too Linaro's 96boards miniITX huskyboard look nice, but is it for consumers or industrial?

 

[truth teller]

i like the kerning on those slides
that is all

 

[silentbogo]

4-core tdp 25W, 8-core tdp 32W, quite high for arm processors. What manufacturing process are these made of(tsmc 28nm?). That integrated I/O is great, I presume there aren't any igp on them so pure server socs?

I would like to know if these are coming to consumer space too Linaro's 96boards miniITX huskyboard look nice, but is it for consumers or industrial?

That's all that extra logic for I/O and cache.

AMD Opteron 1150 has 8xA57 cores, 4MB L2 and 8MB L3.

Samsung Exynos 7420 has 4xA57 cores with 2MB L2, 4xA53 cores with 256KB L2 and no L3 cache.

Plus you get all the yummy stuff, like SATA-III controllers, eight PCI-E 3.0 lanes, pair of 10GbE ports etc. etc. etc.

 

[HumanSmoke]

That's all that extra logic for I/O and cache.

AMD Opteron 1150 has 8xA57 cores, 4MB L2 and 8MB L3.

Samsung Exynos 7420 has 4xA57 cores with 2MB L2, 4xA53 cores with 256KB L2 and no L3 cache.

Apples and Oranges.
You would be better off comparing Seattle with a comparable environment. AMD has it's work cut out for it. Over a year late, missing promised features, average performance envelope, and having to compete with entrenched competitors like Cavium's Thunder-X and Applied Micro's X-Gene at the higher....and lower end of the markets.

Plus you get all the yummy stuff, like SATA-III controllers, eight PCI-E 3.0 lanes, pair of 10GbE ports etc. etc. etc.

All that (and more) has been available for some time in what amounts to a relatively budget server system.

 

[silentbogo]

Apples and Oranges.
You would be better off comparing Seattle with a comparable environment. AMD has it's work cut out for it. Over a year late, missing promised features, average performance envelope, and having to compete with entrenched competitors like Cavium's Thunder-X and Applied Micro's X-Gene at the higher....and lower end of the markets.

My comment goes in response to 25W TDP comment. I picked Exynos one of less power-hungry ARM CPUs for comparison and to give a little explanation as to where that power goes.

 

[$ReaPeR$]

4-core tdp 25W, 8-core tdp 32W, quite high for arm processors. What manufacturing process are these made of(tsmc 28nm?). That integrated I/O is great, I presume there aren't any igp on them so pure server socs?

I would like to know if these are coming to consumer space too Linaro's 96boards miniITX huskyboard look nice, but is it for consumers or industrial?

"
The quad-core A1120 is a 25W chip at 1.7GHz, the A1150 is a 32W core at 2GHz, and the A1170 is a 32W chip at 2GHz. The last two flavors are both eight-core processors, but all three SKUs share the same memory capabilities and L3 cache. If AMD follows its past history on evaluating TDP, the 25-32W figures will represent a worst-case scenario for the chip, rather than its performance in a general workload. AMD historically rates TDP in the first fashion, while Intel uses the second — and that’s why you can’t usually directly compare TDPs between the two companies.
"
http://www.extremetech.com/extreme/...d-processor-the-opteron-a1100-is-finally-here

lets hope this brings them some revenue..

 

[Musaab]

So this is just another ARM CPU , And? Did they really spent part of their small budget to enter ARM market like it's not allready over flowing with names like Qualcomm, Mediatek,Free Scale, Cavium, and Applied Micro.
Why data centers? When a big company want a data center they will go to Oracle or IBM and thanks to open power foundation you can get cheaper systems from IBM partners, for the middle there is Intel with partners like HP, Dell and Lenovo and last but not least the small data centers where Intel still fight to control versus many many ARM supplier.
Wrong move AMD you had to put the money used to develop these CPUs in better place.

 

[HumanSmoke]

My comment goes in response to 25W TDP comment. I picked Exynos one of less power-hungry ARM CPUs for comparison and to give a little explanation as to where that power goes.

You might be better served comparing it to a competitor (most other SoC's in the server class have much higher core counts). 25W and the feature set puts it in direct competition (at least for the higher end of the microserver market) with Intel's old 8 core Avoton Atom (20W). Features/Power/Performance would put them close depending upon the users prioritization, but the Avoton would almost certainly win out in perf/watt for most workloads.

 

[GreiverBlade]

So this is just another ARM CPU , And? Did they really spent part of their small budget to enter ARM market like it's not allready over flowing with names like Qualcomm, Mediatek,Free Scale, Cavium, and Applied Micro.
Why data centers? When a big company want a data center they will go to Oracle or IBM and thanks to open power foundation you can get cheaper systems from IBM partners, for the middle there is Intel with partners like HP, Dell and Lenovo and last but not least the small data centers where Intel still fight to control versus many many ARM supplier.
Wrong move AMD you had to put the money used to develop these CPUs in better place.

i guess you don't understand that Opteron is aimed for server and general computing? (well maybe you understand it ... )

Qualcomm (principally smartphone and now, drones, and server but they are even newer than AMD in that sector, thus not carrying the same weight), Mediatek (mostly smartphones), Freescale(cortex A9 at the maximum) don't do server side, on the other hand you are right for Cavium and Applied Micro, so that make only 2(3) of the 5 "concurrent* mentioned, that's not what i call overflowing, not mentioning that AMD is already in that sector with the X86-64 Opteron.
AMD entering that sector might bring competition in it and also alternatives, depending the pricing, to the other brand already available

ofc any arm SOC can be used as a server cpu depending the demand (lots of home media server based on the Raspberry Pi 2 also ... )

who knows, that idea might not be so wrong and bring AMD some success they look after, at last i will pray for that, i wish them all the best.
(and i have a Mediatek SOC in my smartphone ... a intel CPU in my main and secondary rig and a Broadcom SOC in my Rpi2 )

 

[HumanSmoke]

Without wishing to get embroiled in the discussion re whether smartphone SoC's qualify as possible future server market application (which seems more like a curiosity than any meaningful push into the market from what I've seen)...

Freescale(cortex A9 at the maximum) don't do server side,

It wouldn't take much for them to get involved. They have the infrastructure, and it's probably a case of "when" rather than "if" they expand from their current server architectures

on the other hand you are right for Cavium and Applied Micro, so that make only 2(3) of the 5 "concurrent* mentioned, that's not what i call overflowing

There are also a few other players in the market - some with some serious backing and probably with an inside track to many markets that AMD could only dream about. Broadcom's Vulcan is also a contender once the dust from the Avago takeover settles. That's a lot of competitors in what amounts to a comparatively niche market.

not mentioning that AMD is already in that sector with the X86-64 Opteron.

Barely. At last count AMD had around 1.5% of the x86 server market, and those sales are entirely legacy support. A LOT of OEM's got burned by BD/PD, including some rather public (for server/HPC) dissatisfaction from some vendors (Cray being prominent)thanks to cancelled plans regarding OEM contracts with SeaMicro and the cancellation of Seattle's use of Freedom Fabric. I'm guessing AMD will need to reestablish some trust before any large scale contracts are signed.

AMD entering that sector might bring competition in it and also alternatives, depending the pricing, to the other brand already available

Any new name in a sector has to be good for competition, but after just reading Ars Technica's take on Seattle (which pretty much mirrors my previous post in the thread), it does look lukewarm - AMD wanting customers to compare Seattle with a 2+ year old Intel platform doesn't send a strong message - at least not a positive one:

AMD has been vague about both pricing and performance. The company says that the top-end part will cost around $150, with the others coming in below. This is quite a bit cheaper than Intel's Xeon D processors, which pack Broadwell-class processor cores with a bunch of I/O connectivity. These parts start at $199 and go up as high as $675. However, AMD concedes that the Xeon Ds are considerably faster. The company says that Intel's 2013-era Atom-based C2000 series systems-on-chips are a better comparison for these new models.

 

[IvantheDugtrio]

I'd like to know what chipset the 10Gb connections are based off of. My guess is Intel.

 

[JunkBear]

Do I see an extreme cruncher folder cpu??

 

[GreiverBlade]

Without wishing to get embroiled in the discussion re whether smartphone SoC's qualify as possible future server market application (which seems more like a curiosity than any meaningful push into the market from what I've seen)...

It wouldn't take much for them to get involved. They have the infrastructure, and it's probably a case of "when" rather than "if" they expand from their current server architectures

well based one all that, it's pretty much what i said ... any ARM SOC can be used for server, for a fraction of a "pro" chip

Barely. At last count AMD had around 1.5% of the x86 server market, and those sales are entirely legacy support. A LOT of OEM's got burned by BD/PD, including some rather public (for server/HPC) dissatisfaction from some vendors (Cray being prominent)thanks to cancelled plans regarding OEM contracts with SeaMicro and the cancellation of Seattle's use of Freedom Fabric. I'm guessing AMD will need to reestablish some trust before any large scale contracts are signed.

well they still would hold more weight than Qualcomm or Mediatek even given that situation, hum i must have mixed server and supercomputer market then ... wait ... even there only the Titan use the Opteron 6274, well he's second behind Tianhe-2 (altho most of the computing power come from the Tesla K20X )

Spoiler

semi out of topic fun fact Titan (Cray XK7 AMD/Nvidia) achieve more than the half TFlops of the Tianhe-2 (Intel/intel) using 5 time less core and consuming a bit less half of the KW it needs, so ... AMD does better in supercomputer than the rest? well all other SC under Titan either use more core (power BQC/alpha) or more power (Xeon) and produce less so ... a Xeon-269X 12C/16C is less efficient than a Opteron 6274 16C if at a same core count they produce less and at equal count they produce half (only Titan an Tianhe-2 use GPGPU iirc ... Tesla K20X for one and Xeon Phi 31S1P) at same core count the Titan would have 5 time more Tflops than the Intel/Intel Tianhe-2? not so sure tho it would be 1st rank undoubtedly

 

[HumanSmoke]

semi out of topic fun fact Titan (Cray XK7 AMD/Nvidia) achieve more than the half TFlops of the Tianhe-2 (Intel/intel) using 5 time less core and consuming a bit less half of the KW it needs, so ... AMD does better in supercomputer than the rest? well all other SC under Titan either use more core (power BQC/alpha) or more power (Xeon) and produce less so ... a Xeon-269X 12C/16C is less efficient than a Opteron 6274 16C if at a same core count they produce less and at equal count they produce half (only Titan an Tianhe-2 use GPGPU iirc ... Tesla K20X for one and Xeon Phi 31S1P) at same core count the Titan would have 5 time more Tflops than the Intel/Intel Tianhe-2? not so sure tho it would be 1st rank undoubtedly[/spoiler]

That's a product of the benchmark workload. Linpack favours GPU-heavy systems - although Tianhe-2's Xeon Phi's actual performance isn't great*. The higher the GPU (and GPU core) -to- CPU (and CPU core) ratio the more efficiency is gained**. Look for a high GPU per node count for efficiency. The Green500 TFOPS/watt metric is a good comparison (note that the top system uses PEZY-SC a dedicated multi-core co-processor unencumbered by a traditional graphics pipeline).

* Xeon Phi's efficiency drops markedly as the job size increases. By all accounts Tianhe-2's Xeon Phi's remain idle most of the time. No big deal since the whole system was basically gifted to China by Intel (including the custom Xeon SKU being used). Does make for good marketing though.

** A CPUs floating point ops (even if using AVX /FMA to boost ops per cycle) is pitiful compared to a GPGPU co-processor. AMD's own numbers are in the Opteron presentation notes (Take the comparison with Intel's Xeon with a pinch of salt obviously due to the constricted nature of the instruction sets they are using - this and the previous link are more indicative of the current state of play)

Spoiler

 

[GreiverBlade]

That's a product of the benchmark workload. Linpack favours GPU-heavy systems - although Tianhe-2's Xeon Phi's actual performance isn't great*. The higher the GPU (and GPU core) -to- CPU (and CPU core) ratio the more efficiency is gained**. Look for a high GPU per node count for efficiency. The Green500 TFOPS/watt metric is a good comparison (note that the top system uses PEZY-SC a dedicated multi-core co-processor unencumbered by a traditional graphics pipeline).

* Xeon Phi's efficiency drops markedly as the job size increases. By all accounts Tianhe-2's Xeon Phi's remain idle most of the time. No big deal since the whole system was basically gifted to China by Intel (including the custom Xeon SKU being used). Does make for good marketing though.

** A CPUs floating point ops (even if using AVX /FMA to boost ops per cycle) is pitiful compared to a GPGPU co-processor. AMD's own numbers are in the Opteron presentation notes (Take the comparison with Intel's Xeon with a pinch of salt obviously due to the constricted nature of the instruction sets they are using - this and the previous link are more indicative of the current state of play)

Spoiler

well let's say that the couple Opteron/Tesla is better than Xeon/Xeon Phi, i suppose cost to cost and perf to perf AMD still has it high thanks to ... "irony" Nvidia

thanks for the info @HumanSmoke

also that's why i did put in parenthesis "tho all the computing power come from the K20X"

 

[Stefan Payne]

AMD does someting and you bash it.
Why do you do that?!

Are you happy to pay 250€ for an entry level 4 core CPU?!
Or do you rater pay 500€ for it??

I don't understand it...

AMD has found a nice and tries to make some good CPUs for that niche to get some money - and it's not right.
What the?!

Why do you always see the negative stuff when it comes to AMD and always highlight the good things with their competitors?! I really don't get it...
Must be some kind of sect or so...


Anyway:
It's an 8 core ARM with 2 Channel DDR3/DDR4 Memory controller, a massive amount of Cache (12MiB L2 and L3) and a massive amount of I/O things!
Not the low power things but high performance I/O stuff like PCIe 3.0 (x8) or a whopping 14 S-ATA 6G ports.
And not only that but also 2 10GBit Ethernet ports...

That this monster consumes a little bit more than a mobile performance chip with next to no I/O or rather just low power/low performance I/O Stuff wich are designed for use with no memory slots but rather memory onboard. That also saves power!!

This beast is less for computing and more for storage/database things. So do the other ARM Socs also have 14 S-ATA 6G Ports? And an x8 PCie 3.0 connector?
I think you underestimate the power modern high performance I/O stuff consumes!
That's where most of the power consumption comes from, not from the CPU cores...

The other ARM processors mentioned here are more for Supercomputer/Computing stuff that needs high memory bandwith and computing power. Doesn't seem that this AMD CPU has designed the A1100 series as a competitor for those...

But well, we are talking about AMD...
And when some people read the name AMD, they have to bash it...

 

[jabbadap]

That's all that extra logic for I/O and cache.

AMD Opteron 1150 has 8xA57 cores, 4MB L2 and 8MB L3.

Samsung Exynos 7420 has 4xA57 cores with 2MB L2, 4xA53 cores with 256KB L2 and no L3 cache.

Plus you get all the yummy stuff, like SATA-III controllers, eight PCI-E 3.0 lanes, pair of 10GbE ports etc. etc. etc.

Yes, I know it has those extra I/O to consume more power than mobile socs. But exynos is _14nm_ mobile soc and as I know this is _28nm_ server soc(correct me if I'm wrong). So how much lower power this could be if this had been manufactured with 20nm, 16nm ff+ or even 14nm processes?

 

[Stefan Payne]

You can't shrink that analogue stuff much if at all, so it doesn't matter if we talk about 28nm or 10nm, it would have the same size due to the needed current for driving the Signal.