Hewlett-Packard announced today two new “Apollo” server systems, the Apollo 6000 aimed at dense, hyper-virtual enterprise deployments and the Apollo 8000 aimed at HPC (high performance computing). I wanted to focus on the Apollo 6000 in this column as I think it is the most interesting competitively as it brings Hewlett-Packard squarely into the hyper-virtualized world of Cisco Systems UCS. If you want to do a deeper dive on the Apollo 6000, you can download our paper here. Let me start talk about “hyper virtual” environments as it’s a new term we’re introducing.
New “Hyper Virtual” Server Segment
The server world is getting quite complex as the number of form factors and specialized workloads are multiplying. As these needs specialize, it’s not just about how many processors that can be shoved a rack, although that’s still important for density, cooling and power. Throughput requirements are also very important for certain environments and workloads. Throughput is amount of data that can be taken out of the server chassis to the TOR (top of rack) switch. We (Moor Insights & Strategy) have created a new server segmentation that adds this new segment called “hyper virtual”. “Hyper virtual” environments are extremely dense and have very high throughput. You can learn more about our new rack scale server segmentation here; our “hyper virtual” segment includes Cisco Systems, Dell, and now Hewlett-Packard examples.
Hewlett-Packard Apollo 6000 System
Hewlett-Packard’s new Apollo 6000 fits right into the “hyper virtual” environment as it is suitably dense, as Hewlett-Packard demonstrated they were capable of with their Moonshot systems. What is new for Hewlett-Packard is that they have designed in very high chassis-to-TOR throughput to address the needs of lightly-threaded HPC applications like Electronic Design Automation (EDA) and Monte Carlo Simulations, used for financial risk modeling and various engineering/scientific applications. To tackle these, the system offers high per-thread performance, robust network bandwidth, and rack-level shared infrastructure for efficiency.
The Apollo 6000 System is the newest server in Hewlett-Packard’s Scalable Product line. The Apollo 6000 System is the first of this portfolio to use Intel Xeon E3 Processors V3 (aka Haswell) Family and does it in a single-socket configuration. It bridges the gap between the high performance HP ProLiant SL servers mainly built around Intel Xeon E5 processors (larger cores, dual socket) and the HP Moonshot servers that deliver higher density but use Intel Atom, AMD Opteron APU (accelerated processing unit), and ARM processors (smaller cores, single socket). Apollo takes a similar approach to the Moonshot family with servers that are purpose-built and modular to tackle the specific resource requirements of various workloads. With Apollo’s focus on larger cores and Moonshot designed for energy efficiency, these products target different workloads.
More Competition for Cisco Systems UCS?
Up until this point, Hewlett-Packard lacked a hyper virtual system to compete with high-throughput configurations of Cisco UCS B200, B220, and the B440. You can see this more clearly in this segmentation here. HP significantly increased throughput with the Apollo 6000 through what HP calls their “I/O Innovation Zone”, providing 1G, 10G, or InfiniBand depending on an enterprise’s needs. For workloads that need both the per-thread performance of a single socket Xeon-E3 and significant network bandwidth per rack, a system configuration with 10GbE could provide up to 1600 GbE Uplink bandwidth per rack.
While the Apollo 6000 today doesn’t provide more throughput than the highest configured Cisco Systems UCS, HP is certainly capable of adding fatter links. Net-net, customers buying Cisco UCS based on throughput and density have a new alternative.
As the vast majority of the workloads in the datacenter are running on dual CPU platforms, the Hewlett-Packard Apollo 6000 System is not necessarily a replacement for all of the servers in a datacenter, but it is an important additive element to help drive a better blended efficiency across workloads in the datacenter. The system’s high per-thread performance, robust network bandwidth, and rack-level shared infrastructure for efficiency make it a strong potential fit for lightly-threaded HPC applications, hosting, and hyper-virtualized environments.
HP’s planned introduction of future server trays with other capabilities could help expand the market opportunity for the Apollo 6000 System in the future. If you want a deep dive on the new “Hyper Virtual” segment you can see more here, and if you are looking for more on HP’s Apollo 6000, you can download our paper here.