- And 65nm process "Power6" in comparison, Power7 in what has improved?
Power6, compared with, Power7 kernel number increased to four times the number of threads per-core executable to a two-fold. In addition, imported to the kernel half of sleep, increasing an average of cache memory allocated to the remaining core of the "Turbo Core Mode" functions. Not only improve performance, but also uses a number of low-power technology. If only from a performance point of view, Power6 not less than Power7, but can not be called low-power processor. The Power7 unit of electricity consumed will be the performance of up to 4 times the Power6.
- The previously external L3 cache as a mixed set DRAM integrated into Power7, this point it has become a topic of concern.
Containing DRAM using mixed mode L3 cache, and is said to achieve our Power7 the largest technological innovations. We have about 10 years ago started to focus on the processor DRAM mixed set, this was finally achieved. It took 10 years to achieve, we can see the high technical difficulty. With the use of mixed composition containing L3 cache, SRAM, mixed containing DRAM memory cell area can be reduced to 1 / 3, the standby power consumption down to 1 / 5. When compared with the external L3 cache, latency (Latency) can be shortened to 1 / 6.
- With no less than Intel's processors do?
Less better than Intel's chips. Particularly in the processor core performance and memory access speed balance, as well as during operation, the average memory bandwidth, but also over other competitors, including Intel. Even if only from the point of view chip area also has advantages, such as: mixed set through the use of DRAM, making integration of L3 cache memory density is higher than the same process technology, Intel's processors.
- Following the Power7 after the "Power8" would be what kind of performance is to have a chip? In addition, your company has been put into operation when using Intel's 32nm manufacturing process processor?
Power8 is still in the conceptual stage of chip design, performance indicators can not open. However, the mixed containing DRAM, etc. We hope to continue to adopt good technology. With regard to commissioning date, you can refer to the past (every three years, a new architecture launch) cycle. While the 32nm production process of a date can not be published, but I can tell that the production has been incorporated into the field of vision.
- With the energy conservation become a global trend, the server processors compete with what changes occurred?
Low-power technology has become an important differentiating factor. IBM in 2001, production in the 180nm process of "Power4" when the power management techniques even responsible for the development team of professionals does not exist. At present, in the chips and systems at various levels have started to use low-power technology. In the future, into a similar power gating (Power Gating) as a low-power technology is a must.
- IBM to engage in one-stop until the server from the semiconductor process development. From the processor is responsible for the development of technical staff point of view, this is not achieved with the competitors to differentiate the key factor?
I very much agree with this view. Can improve the processor performance of "magic-like approach" does not exist. From the semiconductor process until the chip design, firmware, OS, servers, those responsible at all levels must work in their respective positions will do our best. IBM, with such an environment. For example, the company has senior engineers (technical director) to participate in the meeting. By this way with the semiconductor technology experts to explore this even if I am responsible for the design is also extremely useful.
没有评论:
发表评论