A diagram of the X-Gene platform, which combines four dual-core 64-bit ARM v8 chips.[/caption] Executives at Applied Micro Circuits Corporation (AMCC) said this week that its X-Gene platform is expected to ship this year, good news for those interested in taking advantage of the company’s low-cost ARM server chip. At the Hot Chips conference in Cupertino this week, Paramesh Gopi, AMCC’s chief executive, described the chip as necessary for a new class of scale-out machines in the data center, where the total cost of ownership (TCO) is paramount. “The real key is the need for the real category of purpose-built servers,” Gopi said. “The new cloud workload requires three things: an extremely nimble communications infrastructure, combined with performance-grade CPU architectures and the ability to leverage a very open source software ecosystem. All three are coming together.” The X-Gene combines four, dual-core 64-bit ARM v8 chips said to be running at 2.5 GHz, and which use a shared level-2 cache and a pooled L3. AMCC’s design goals include the ability to deliver “good” single-thread performance, but also the ability to scale out to many cores - 128 initially, and later 256 and 512. The key, however, is that the CPU and memory subsystem are coupled to a coherent network that runs at the full CPU frequency, with 15 ns latencies and bandwidth of 160 GB/s, with interfaces to network accelerators, 17 lanes of PCI Express 3.0 bandwidth and 10 Gigabit Ethernet ports. And it’s all on one piece of silicon. “This has never happened in the history of this particular industry,” Gopi said. “For once we are not talking about CPUs, we are talking about a complete server on a chip platform. And that to me in our minds is going to change the fundamental equation for TCO, and change the fundamentals of how the entire workload scales for cloud and enterprise.” Gopi brought a demonstration board of the X-Gene platform as further proof that the technology was viable, with 256 GB of physical memory on board. The real question, of course, is how much penetration ARM processors will achieve in the data center. A straw poll of attendees put it at about 10 percent. In April, AMCC said that it had demonstrated a Web server running on top of test silicon. The platform is capable of running a full LAMP software stack featuring open-source Linux, Apache, MySQL and PHP. At the time, the company added that test platforms had begun shipping to top-tier hardware customers and operating system vendors in both North America and Asia. Image: AMCC AMCC Readies ARM Server-on-a-Chip
[caption id="attachment_4076" align="aligncenter" width="613"] A diagram of the X-Gene platform, which combines four dual-core 64-bit ARM v8 chips.[/caption] Executives at Applied Micro Circuits Corporation (AMCC) said this week that its X-Gene platform is expected to ship this year, good news for those interested in taking advantage of the company’s low-cost ARM server chip. At the Hot Chips conference in Cupertino this week, Paramesh Gopi, AMCC’s chief executive, described the chip as necessary for a new class of scale-out machines in the data center, where the total cost of ownership (TCO) is paramount. “The real key is the need for the real category of purpose-built servers,” Gopi said. “The new cloud workload requires three things: an extremely nimble communications infrastructure, combined with performance-grade CPU architectures and the ability to leverage a very open source software ecosystem. All three are coming together.” The X-Gene combines four, dual-core 64-bit ARM v8 chips said to be running at 2.5 GHz, and which use a shared level-2 cache and a pooled L3. AMCC’s design goals include the ability to deliver “good” single-thread performance, but also the ability to scale out to many cores - 128 initially, and later 256 and 512. The key, however, is that the CPU and memory subsystem are coupled to a coherent network that runs at the full CPU frequency, with 15 ns latencies and bandwidth of 160 GB/s, with interfaces to network accelerators, 17 lanes of PCI Express 3.0 bandwidth and 10 Gigabit Ethernet ports. And it’s all on one piece of silicon. “This has never happened in the history of this particular industry,” Gopi said. “For once we are not talking about CPUs, we are talking about a complete server on a chip platform. And that to me in our minds is going to change the fundamental equation for TCO, and change the fundamentals of how the entire workload scales for cloud and enterprise.” Gopi brought a demonstration board of the X-Gene platform as further proof that the technology was viable, with 256 GB of physical memory on board. The real question, of course, is how much penetration ARM processors will achieve in the data center. A straw poll of attendees put it at about 10 percent. In April, AMCC said that it had demonstrated a Web server running on top of test silicon. The platform is capable of running a full LAMP software stack featuring open-source Linux, Apache, MySQL and PHP. At the time, the company added that test platforms had begun shipping to top-tier hardware customers and operating system vendors in both North America and Asia. Image: AMCC
A diagram of the X-Gene platform, which combines four dual-core 64-bit ARM v8 chips.[/caption] Executives at Applied Micro Circuits Corporation (AMCC) said this week that its X-Gene platform is expected to ship this year, good news for those interested in taking advantage of the company’s low-cost ARM server chip. At the Hot Chips conference in Cupertino this week, Paramesh Gopi, AMCC’s chief executive, described the chip as necessary for a new class of scale-out machines in the data center, where the total cost of ownership (TCO) is paramount. “The real key is the need for the real category of purpose-built servers,” Gopi said. “The new cloud workload requires three things: an extremely nimble communications infrastructure, combined with performance-grade CPU architectures and the ability to leverage a very open source software ecosystem. All three are coming together.” The X-Gene combines four, dual-core 64-bit ARM v8 chips said to be running at 2.5 GHz, and which use a shared level-2 cache and a pooled L3. AMCC’s design goals include the ability to deliver “good” single-thread performance, but also the ability to scale out to many cores - 128 initially, and later 256 and 512. The key, however, is that the CPU and memory subsystem are coupled to a coherent network that runs at the full CPU frequency, with 15 ns latencies and bandwidth of 160 GB/s, with interfaces to network accelerators, 17 lanes of PCI Express 3.0 bandwidth and 10 Gigabit Ethernet ports. And it’s all on one piece of silicon. “This has never happened in the history of this particular industry,” Gopi said. “For once we are not talking about CPUs, we are talking about a complete server on a chip platform. And that to me in our minds is going to change the fundamental equation for TCO, and change the fundamentals of how the entire workload scales for cloud and enterprise.” Gopi brought a demonstration board of the X-Gene platform as further proof that the technology was viable, with 256 GB of physical memory on board. The real question, of course, is how much penetration ARM processors will achieve in the data center. A straw poll of attendees put it at about 10 percent. In April, AMCC said that it had demonstrated a Web server running on top of test silicon. The platform is capable of running a full LAMP software stack featuring open-source Linux, Apache, MySQL and PHP. At the time, the company added that test platforms had begun shipping to top-tier hardware customers and operating system vendors in both North America and Asia. Image: AMCC