HGST Demonstrate Phase Change Non-Volatile Memory

HGST blew people away last year when they presented their amazing PCM PCIe SSD that could deliver a record-breaking three million IOPS. At this years Flash Summit in Santa Clara they’re planning to have the same effect with their newly developed breakthrough persistent memory fabric. HGST’s Phase Change Memory is expected to deliver performance very close to DRAM, but at a much lower cost of ownership, as well as offering much greater scalability.

DRAM is fast, but it isn’t an optimal medium to store data for longer periods of times. The data needs to be refreshed fairly often due to leaky capacitors and the refresh power consumption can be as much as 20 to 30 percent of the servers total power consumption. That means that PCM storage could create huge cost savings for server farms while they still deliver the same performance.

The new technology doesn’t require any BIOS modifications nor rewriting of applications, so it can be deployed as soon as it is ready. It isn’t entirely clear what kind of products we’ll see from this and if we will see any at all – at least as consumers. It is still great news as our consumer technology started as enterprise technology at some point and even if these drives won’t make it to our systems, then the online services that we use will benefit from them and be able to deliver even better services, hopefully.

No matter what comes out of it, the demonstration by HGST and Mellanox show random access latency of less than two microseconds for 512B reads and a throughput exceeding 3.5GB/s for two KB block sizes using RDMA over InfiniBand; very impressive!

U.S. Department of Energy to Spend $425 Million on Supercomputers

The US Government’s Department of Energy has announced it is to invest $425 million to build two supercomputers, which, when built, will be the fastest computers in the world. The ultimate aim is to research science projects, including nuclear weapons.

The two computers, named Summit and Sierra, will be installed at Oak Ridge National Laboratory, Tennessee, and Lawrence Livermore National Laboratory in California, respectively.

NVIDIA, IBM, and Mellanox have provided the components for use in the two computers. Summit will run at 150 petaflops, with Sierra operating at 100 petaflops. For comparison, the world’s current fastest supercomputer, the Chinese Tianhe-2, runs at 55 petaflops.

An extra $100 million will go to fund research into extreme-scale computing, under the project name FastFoward2.

Source: Reuters