Life's Good with HSA and the AMD Opteron ARMy



Author: James Prior
Editor: Charles Oliver
Date: October 31st, 2012

LG Joins the HSA

This week is a big one for AMD, host to the non-profit Heterogeneous System Architecture foundation, as another founder member is added - LG. This is another huge addition to the HSA - LG is a massive OEM that not only designs and sells those designs to companies, but manufacture products to sell directly to consumers - like Samsung, the only HSA Foundation Member OEM who also builds x86 PC's. LG has a massive range of products that leverage embedded ARM as well as their mobility products; the advantages of HSA will help them get SoC designs tailored for their needs to better differentiate their products in the market.

Life's Good With HSA

The HSA Foundation now includes most of the big SoC players and, speaking with Manju Hegde again this week, we hear there are another half dozen new members at different levels to come. This bodes well for standards that the HSA Foundation is looking to create; the more of industry agrees to contribute and use them the better chance of mass adoption and success. Implicit in that is agreement to compete at the product level, using the marketing funds to promote products rather than possibly restrict OEMs or System Integrators from also using competing designs or products from other HSA members. The goal here is to give everyone a slice from the largest pie, rather than compete for the big slice of the smallest pie and capped in how much there can be overall.

Opteron ARMy

In keeping with that, AMD announced this week a second licensing agreement with ARM, this time for the new 64-bit core design: Cortex-A50 series. AMD currently licenses the ARMv5 core for use as a security co-processor, implementing ARM Trust Zone technology; in keeping with the Fusion spirit, the new ARM 64-bit core license will be delivered as part of the Opteron product line blending Bulldozer x86 and AMD64, with Coretex-A57. AMD touts they will be in a unique position, offering x86, ARM and GPU compute prowess to high performance compute, megascale datacenter and behemoth enterprise customers and the timeframe offered is 2014 for the Opteron ARM to be shipping. This is what NVIDIA is hoping to do with CUDA platforms powered by their Maxwell architecture, although lacking x86.

The timing of this announcement is not coincidental or reactionary to recent events - this week is ARM TechCON 2012 in Santa Clara. ARM is breaking out of the ultra low power mobile device world of the public perception and becoming part of the background processing we all take for granted, powering the amorphous cloud. Just last week, in another non-coincidental action, Calxeda unveiled their EnergyCore SoC for building datacenter infrastructure out of ARM cores, albeit 32-bit currently and 64-bit coming. HSA Foundation president Phil Rogers is speaking at TechCON on Thursday, talking about sensor integration and improved user experiences at even lower power and how the HSA will help provide the platform to make the compute of the future possible.

Freedom Fabric

AMD is citing their SeaMicro expertize and proven interconnect, Freedom Fabric, as being the key ingredient to their winning in the cloud datacenter space. AMD is combining big cores and little cores with AMD's design strengths and knowledge of the needs of OEMs, System Integrators and Enterprises. At the HSA announcement in Santa Clara on Monday, AMD hosted Facebook, Red Hat, Dell and ARM senior representatives, plus a video from Amazon, to show their support for AMD's approach. Red hat already have a 32-bit version of Fedora running on ARMv8, and Facebook certainly sound comfortable talking about the need for ARM processing power. Dell are already in the business of providing ARM datacenters, and Amazon are not likely to look that other way while more hosting and providing opportunities appear - mobile development and testing in the cloud, anyone?

All large datacenter operators, mega-datacenter builders, are becoming more and more sensitive to the Power Utilization Effectiveness index offered by commodity servers. Current x86 servers are very similar, no room for customization or personalization other than superficially, which keeps prices and margins low and throws focus on the associated costs of running the machines - the power used isn't returned by the work done. This is sometimes expressed as performance per watt, but is more based in actual workloads in customer applications rather than theoretical FLOP throughput.

Why ARM in the Datacenter?

Big datacenters are looking for new ways to deploy, upgrade and finetune their compute power, and the blade by blade or box by box approach isn't their favored approach anymore, with financial write down periods driving refresh cycles. Instead, companies like Facebook would like to refresh DRAM, storage, processors or network when it makes sense and independently of each other. AMD can deliver this with their SeaMicro approach, and importantly AMD becomes an OEM and competitor to some of their HSA partners with this scenario.

Why AMD with ARM?

AMD President and CEO Rory Read informs us that the engineering teams for the new ARMv8 blended Opterons are in place and the progress well underway, part of a long term ambidextrous strategy for AMD. However, the need for little cores, more power efficient and highly adaptable, would seem to speak to the existing Bobcat and Jaguar architecture families, rather than Cortex-A57. AMD is not architectural designer licensed, meaning they won't be designing their own ARM cores, and Coretex-A57 isn't going to be used in client devices (that would be A53) - AMD's Bobcat and Jaguar based devices are going to compete against the ARM onslaught in the phone, tablet, convertible, and notebook spaces.

Why not bring those x86 small cores in to the datacenter, as was first suggested when the Bobcat architecture was first disclosed? Reading through the meandering responses, it would appear that the choice for certain big providers has been made and AMD is now hoping to be a part of it. However, AMD needs to make up for the last 3 years and show how the engineering teams will be bolstered with new/imported talent and resources, why the Bluestacks investment makes any kind of sense now, and what kind of devices we're going to see in the 2014 datacenter powered by AMD.

Heterogeneous System Architecture Foundation

2014 is an interesting date for when we'll see the Opteron ARMy and might be indicative of when we can expect the long-rumored-never-seen Windows Server on ARM OS to appear. The HSA Foundation has done well in getting SoC designer and IP holders to the table, along with some major OEMs but now the focus needs to come on ISVs - Google, Microsoft and Apple need to be signatory to the HSA Foundation for it to be the ultimate success that is envisioned. AMD took pains not to say they will design their own ARM cores but also to hint that their partnership with AMD is going to grow, meaning more possibilities for designs in future and optomizing Cortex-A57 will be crucial to AMD's success. LG is a major and welcome addition to the HSA Foundation and it'll be interesting to see how the everyday consumer benefits from HSA powered devices, and if those designs come from AMD or another SoC company. Will AMD be part of the noise of SoC vendors but destined to fade away, or will certain key high profile wins keep them in prominence in the consumer's minds?