Gil Pitney demonstrates how Texas Instruments’ Keystone II ARM+DSP multicore SoCs are ideal for “green supercomputing”, performing demanding High Performance Computing (HPC) workloads at lower power. TI’s Mulicore SDK for HPC (MCSDK-HPC) examples show how TI’s OpenCL driver and the OpenMP 4.0 Accelerator Model allow demanding scientific computations to be easily offloaded and distributed to the 8 DSP cores.
Cavium launches the world’s fastest ARM Processor in their family of workload optimized ThunderX 64bit ARMv8 Server Processors (including ThunderX_CP for Cloud, ThunderX_ST for Storage, ThunderX_SC for Security and ThunderX_NT for Networking), for a range of applications in the cloud and data center. With 48 cores running at 2.5GHz each, ThunderX is the world’s highest performing low-power 64-bit ARMv8 SoC family of workload optimized processors with a range of SKUs and form factors for high performance volume compute, storage, secure compute and networking specific workloads. Analysts predict that the global data center infrastructure market, including servers, storage, networking, security and virtualization, will reach $128 billion in 2014. Cavium is hereby taking their share of that market by releasing their extremely high performance custom design ARM Server processor.
This product family is based on highly efficient full custom processor cores designed by Cavium in 28nm process technology under architectural license from ARM. It is fully compliant with ARMv8 architecture as well as ARM’s Server Base System Architecture (SBSA) standard while bringing to market dramatic enhancements that include:
-The first ARM based SoC that scales up to 48 cores with up to 2.5 GHz core frequency with 78K of I-Cache and 32K of D-Cache along with 16MB of L2 cache.
– The first ARM based SOC to be fully cache coherent across dual sockets using Cavium Coherent Processor Interconnect (CCPI™)
– Integrated I/O capacity with 100s of Gigabits of I/O bandwidth
– Four DDR3/4 72 bit memory controllers capable of supporting 2400 MHz memories with 1TB of memory in a dual socket configuration
– Hundreds of integrated hardware accelerators for security, storage, networking and virtualization applications.
– Standard based low latency Ethernet fabric interconnecting thousands of ThunderX™ nodes in 2D and 3D configurations and enabling fabric monitoring and SLA enforcements with awareness and policy enforcement for virtualized networks.
– Virtualization everywhere with Cavium virtSOC™ technology – Full system virtualization for low latency from virtual machine to I/O.
– Best in class performance per watt and performance per dollar for the target applications
Read more: press release
- Cavium announces 48-core ARMv8 server processor (techreport.com)
- Cavium packs 48 cores into ThunderX ARM chips (pcworld.com)
- Cavium Thunder X ups the ARM core count to 48 on a single chip (semiaccurate.com)
- Welcoming Cavium as latest Xen Project Advisory Board member (xen.org)
- Cavium ThunderX: 2.5 GHz, 48 Core ARMv8 SoC (phoronix.com)
- Cavium ThunderX ARM Chip Rumbles Into Hyperscale (enterprisetech.com)
- 48 Core ARM-64 processors from Cavium to challenge Xeons? (eetimes.com)
- 64-bit Cortex Platform To Take On x86 Servers In The Cloud (electronicdesign.com)
- Cavium sets sights on Intel with 48-core SoC (go.theregister.com)
- Cavium Announces Mainstream ARM Server Chips To Challenge Xeon (techweekeurope.co.uk)
Here’s the 16-core ARM Cortex-A15 processor from HiSilicon Huawei on a development board for ARM Powered Networking and Servers coming up. Hacked on in this video by Linaro Toolchain Engineer Rob Savoye (2), who now is climbing the Mount Everest. Linux kernel v3.13 is running on this board, with three SATA ports and two Gigabit ethernet ports driver ready. The BSP code will soon be upgraded to kernel v3.14 and be upstreamed in parallel. Source code and binaries are released through Linaro website. Ubuntu Server is verified on this board. In this demo, it runs a GCC toolchain native build. Linaro Toolchain Working Group plans to use this board to run multiple builds per board, to maximally saturate D01’s computing and storage capability.
Kernel source: http://git.linaro.org/landing-teams/working/hisilicon/kernel.git (branch: integration-hilt-d01)
Binary release: http://www.linaro.org/downloads/ (found ‘HiSilicon D01′)
WiKi page: https://wiki.linaro.org/Boards/D01
Linaro is working on implementing ACPI for general purpose servers using the ARMv8 architecture. This has been controversial as it is a competing technology to FDT which has been used now for the 32bit ARM world.
ACPI has been chosen on for the general purpose servers to allow standard distributions such as RHEL and Ubuntu server to boot on hardware which they have no special support in the same way as x86 world. ACPI is used to abstract the hardware to the level the standard distribution can boot to the point it can be useful.
There is a large overlap between FDT and ACPI but they actually do things a different way. FDT is currently holding fast the the mobile and tablets market for ARM. But with Intel implementing ACPI phones and tablets nothing is certain for the future.
The Linaro Enterprise Group (LEG) is dedicated to accelerate Linux ARM server ecosystem development and extends the list of Linaro members beyond ARM silicon vendors to Server OEM’s and commercial Linux providers.
Linaro Enterprise Group (ARM Servers) engineers Ed Nevill, Leif Lindholm, Andrea Gallo, Al Stone, Hanjun Guo share key achievements in the OpenJDK, HipHopVM, UEFI and ACPI areas, building on new hardware, defining plans towards upstream acceptance, solving bugs and more.
Cavium talks about and shows their latest enterprise, data center, wired and wireless networking OCTEON and OCTEON Fusion SoCs based on ARMv8 64bit and MIPS, making customized optimized core designs for each in use for cloud servers and base stations among other. CAVIUM claims that their ARMv8 64bit enterprise/server design, due to be released later this year, provides more performance at lower power consumption than Intel´s x86.
The Linaro-hosted “Enterprise Group” (LEG) is dedicated to accelerate Linux ARM server ecosystem development and extends the list of Linaro members beyond ARM silicon vendors to Server OEM’s and commercial Linux providers.
Showing off the latest Applied Micro 64bit X-Gene ARM Server Development Board (which Rob Savoye of Linaro eagerly wants to start playing with), Dell’s 64bit ARM Server solution running Fedora 19, working on a proof of concept for early 2014 for Dell’s key cloud server customers (Google? Amazon?) before going into mass production (Dell already did some 32bit ARM Server tests in Europe with some customers), some things like Oracle JDK still has to fully come over (needs some tuning) to the platform. This is just me walking kind of randomly around some of the ARM Server demo area at the ARM Techcon. Then checking out the HP ARM Server booth, showing off some of the latest HP Moonshot ARM Server solutions also talked about in HP’s Keynote at ARM Techcon, watch the official video of that keynote or my version (sitting on the front row).
AMD in 2014 will be delivering a 64bit ARM processor for servers. The ARM Architecture and Ecosystem enables servers to achieve greater performance per watt and greater performance per dollar. The code name for the product is Seattle. AMD Seattle is expected to reach mass market cloud servers in the second half of 2014.
Calxeda shows their new ARM Cortex-A15 based ECX-2000 supporting 16GB RAM on quad-core, used by HP in one of their new HP Moonshot ARM Server platform. Performance is 2x to 3x faster than their previous ARM Cortex-A9 Server platform. Calxeda also has 2 64bit ARM Server chips on their Roadmap with full production systems to be expected for early 2015 or so.
You can read more about Calxeda ECX-2000 at http://www.calxeda.com/ecx-2000-family/