Gallery: AMD demos 'Shanghai'
AMD had seen its first effort at quad-core, the 65nm Barcelona, dogged with delivery problems and other issues. The chosen venue for the debut was a datacenter belonging to the second largest ISP in Europe, Strato, in Berlin.
Credit: Colin Barker, ZDNet.co.uk
The company was running a large server kitted with Barcelona processors. Next to it was a server that looked almost identical and was fitted with the new Shanghai processors — shown here. As Shanghai is a drop-in replacement for Barcelona, working in existing motherboards after a Bios refresh, most new Shanghai servers will be physically identical to their predecessors — at least until mid-2009, when new chipsets will enable new features.
The system powered by processors with the older 65nm Barcelona technology was started up and ran with an average loading. This chart of the live demonstration shows the machine running normally.
Four values are shown here. On the bottom-left is the CPU usage, showing the processor working at a level that is roughly the maximum a system manager would feel comfortable with.
The next value up shows electrical power consumed, in watts.
The third value form the bottom shows actual work being done, in this case a count of HTTP requests being served every second.
Finally, the value at the top of the screen shows the power being consumed against the work being undertaken. This is a measure of efficiency, and is useful both in terms of datacenter energy management and for judging green credentials.
The graph on the right-hand side of the middle row shows the performance of the Shanghai processor (in terms of HTTP requests) has increased noticeably.
The power-use graph [top right-hand side] shows the watts per request made, which has dropped substantially in comparison with the equivalent Barcelona graph. This is a raw illustration of the energy efficiency of the processor, and how much work it can do for the energy used.
Energy efficiency is the feature AMD is promoting as a step forward in design. The demonstration appeared genuine, however it was carried out and entirely controlled by AMD. In terms of independent checking, AMD points to the fact the demonstration was performed on a user site belonging to one of the company's largest customers, the ISP, Strato.
As the test progressed there was no change in the overall performance figures other than the power increasing very slightly before levelling off. All other values remained stable.