HPE is the first major OEM to adopt Ampere compute arm chips
Hewlett Packard Enterprise was an early and enthusiastic proponent of alternative processor architectures aside from the standard Xeon X86 processors which account for the vast majority of its revenue and shipments, especially with Arm server chips from 2011 onwards. Be that now, the fourth time will be the charm as it adds Ampere Computing’s Altra Max Arm “Mystique” server chip as the engine in a standard ProLiant rackmount server.
The CPU vendor deal with Ampere Computing is definitely a watershed event for this particular Arm server chipmaker, landing its first major OEM server maker after successfully convincing Baidu, Tencent, Alibaba, Oracle and now Microsoft to join. adopt its 80-core Altra and 128-core Altra Arm processors in their custom server infrastructure – primarily for inward-facing workloads, but increasingly for outward-facing ones sold as raw infrastructure for customers from their clouds who want to know more about Arm servers.
The announcement, made at the HPE Discover 2022 customer and partner event, is quite modest as things stand, with just one machine – the single-socket 1U rack server known as the ProLiant RL300 – in the range. And given HPE’s history with Arm servers, you can probably see why the company is playing a little more conservatively.
In November 2011, the company was known as Hewlett-Packard and adopted Calxeda’s quad-core EnergyCore ECX-1000 Arm server chip, which was effectively a system-on-chip with a single DDR3 memory controller, a management controller, four PCI-Express 2.0 controllers, one SATA 2.0 controller for flash and disk storage, and a fabric switch with eight 10 Gb/sec Ethernet channels (three for interconnecting with three other ECX-100 chips on a map, and five to talk to the outside world, including other interconnected maps in a mesh). The “Redstone” server boards using the Calxeda processor were part of the “Moonshot” hyperscale server design effort launched with great fanfare by Hewlett-Packard at the same time Calxeda unveiled. But the Calxeda chips only had 32-bit memory processing and addressing, and while 48-bit chips were on the roadmap, it was going to take years to add 64-bit chips and that was just too much long for Calxeda or Moonshot to get ground.
In 2017, Hewlett Packard Enterprise, as the company was called after splitting off its PC and printer business, named Cavium’s ThunderX2 server chip as Arm’s next big hope. (Well, it was really Broadcom’s recycled Arm “Vulcan” server chip.) The ThunderX2 was placed in HPE’s dense Apollo 70 “Comanche” barebone servers. A bunch of these systems were sold, but nothing close to volumes, and some time after Marvell bought Cavium for $6 billion and a few HPC systems were up and running to test the ThunderX2, Marvell was gearing up to launch the ” Triton” ThunderX3 follow-on then, at the end of 2020, Marvell spiked ThunderX3 like Broadcom spiked ThunderX2, and this time no one picked up the pieces. Although there have been rumors that Microsoft could do it around the same time.
And then, just over two years later, Microsoft tapped Ampere Computing as its Arm server chip supplier. We have performed an in-depth analysis of Altra instances running on the Azure cloud and calculated that they offer a quarter to a third better price/performance ratio than instances based on AMD “Milan” Epyc 7003 processors on Azure . And for all the reasons we’ve discussed in the past about the Altra line of chips, and primarily because it’s a chip aimed at hyperscalers and cloud-native workloads that performs more consistently on a core with a set clock speed and no simultaneous multithreading, HPE is adding Altra chips to its consumer ProLiant machines.
(An aside: As we’ve discussed here in the past, HPE is also an enthusiastic supporter of Fujitsu’s A64FX Arm processor, designed with large vector math units and used in the “Fugaku” supercomputer at RIKEN Lab in Japan, which has been added to its Apollo 80 systems.)
It was this deterministic performance and price/performance ratio that attracted HPE to Altra processors, and the fact that HPE does not sell it as a special type of Moonshot server, or one of its Cloudline the cheap hyperscaler enclosures, but in a standard ProLiant machine that comes with an iLO or OpenBMC motherboard management controller commonly used in enterprises, shows that HPE is serious.
“HPE and ProLiant have brought 30 years of computing leadership and innovation to a sea of followers,” said Neil MacDonald, executive vice president and general manager of the Compute Group at HPE. “We are deeply committed to compute innovation, whether it is managing automation, performance, security or energy efficiency. And this is a very, very natural next step in our journey to bring compute solutions with cloud-native silicon to support these emerging markets and meet changing customer needs. We’re excited to work with Ampere as they deliver this next-generation, cloud-driven compute for the data center, and Ampere is modernizing this core compute for cloud software and cloud workloads.
There aren’t many streams and speeds available for the ProLiant RL300 server, but it’s a single-socket server in a 1U chassis that has 16 DDR4 memory slots with up to 4TB of memory main, which is a lot for a single socket machine. The 80-core Altra and 128-core Altra Max chips will plug into the enclosure socket, which has three PCI-Express 4.0 slots and two OCP 3.0 slots. The machine has up to two NVM-Express M.2 USB drives and ten small form factor NVM-Express SSDs.
The ProLiant RL300 will be available in the third quarter. Pricing hasn’t been disclosed, but we assume it will be aggressively priced compared to ProLiant machines using Intel Xeon SP and AMD Epyc processors – and that in this case more of the margin will hit the HPE bottom line than what has happened in the past with Xeon SPs.
We also assume that HPE will offer a broader product portfolio, including a capable two-socket machine that rivals the flagship ProLiant DL380 using Xeon SPs and its rival, the ProLiant DL385 using Epycs. We are also expecting more compact sled servers for the Apollo line for those who want more density than what these ProLiants offer. A full-width 1U case with a single server socket isn’t all that dense, and it’s reasonable to expect a 2U chassis with four sleds at some point. If HPE is as serious about Ampere Arm server chips as MacDonald would have us believe. Asked about delivering such a comprehensive range, MacDonald smiled and said:
“When we make decisions about technology, we make them very thoughtfully with a view to the future. And there are other activities already underway, which I’m not going to talk about today in terms of things we can do in the future. But we are very excited about the opportunity and the ecosystem that is in place around this platform.
Like we said: when we see a ProLiant RL380 and an Apollo 2000 running Ampere Computing’s Altra processors, we’ll know HPE really means business.