Meanwhile, at the National Association of Broadcaster’s annual event in Las Vegas, Intel announced new Thunderbolt controllers and demonstrated a next-generation product still in development.
The Intel DSL4510/4410 Thunderbolt controllers add DisplayPort 1.2 capability when connecting to native DisplayPort systems, improve power management and reduce cost. Intel also previewed prototype silicon of Falcon Ridge, a next-generation controller that runs at 20 Gbits/second, enabling 4K video file transfer and display simultaneously. Initial production is expected before the end of this year.
Separately, Pericom Semiconductor announced six new products including a switch that handles Thunderbolt. The 10.3-Gbits/s Thunderbolt switch is designed to multiplex and de-multiplex Thunderbolt data paths into individual signals from the graphics processor and route them to a platform connector or end point.
A Pericom spokesman said Thunderbolt is “being deployed more in the professional video storage and video editing systems.”
The interface was initially seen as a possible mainstream PC interconnect and may still play a role in desktops and notebooks. “Given that we are moving to from 1080P to 2K/4K resolutions, having a 10.3G interface will ease the data movement between the PC and storage,” said the Pericom source.
Click on image to enlarge.
Intel's reference design for server racks, due next year, will include its silicon photonics interconnects.
It is interesting that the new E3 power ranges from 17W, 45W to 87W. They must be really hand-picking the 17W parts because the true power seems to be in the 60W range. Wonder how much yield they can get for the low power part.
The current one is 17W. The new one is down to 13W. More interesting will be the E5 10 core chips at 70W. It will be possible to get 4 of these in a 1U chassis. That's 40 2.5+ GHz cores, 80 hyper-threads, with at least Ivy Bridge performance levels, and at least half a TB of DRAM in a 1U. It is interesting. It is not clear to me that the ARM and Atom server chips will win here. Better performance per Watt is important, but only as long as it can maintain the same or better performance per volume unit. If a 2U space of ARM or Atom chips is needed to run as many VMs as a 1U of E5s, then which really costs less?
The lowest TDP Sandy bridge and corresponding performance is about same as lowest TDP of haswell and its corresponding performance. It can be found in published data that the 22 nm finfet transition took more of a performance hit in order to reduce power.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.