Title: INDUSTRY view of Academic Architecture research
I listened to a class of Wen-Mei HWu today, he just came back from HPCA, as a member of Panel, he and
Several representatives of Industry exchanged about Academic Architecture research
Where is the view. Let us return to us. Here is an impression.
Oracle
Oracle is now doing Data (Storage) Center, that is, big companies (Fortune500)
All handed Database to Oracle maintenance without purchasing a database. It is a bit like a host hosted. Oracle
Concern is, Architecture's Fault Tolerance and Support for Software Debugging
It is very important to do it, it is worth doing, very promising.
It is also important to mention a interesting Topic, Virtual Hardware will be very important. For example, on a machine
You can run multiple operating systems at the same time, upgrade, and restart an operating system does not affect the other operating system. (A bit
Similar to VMware, this Virtual Machine is completely hardware, each operating system is independent of each other.
And the status equality seems to have similar techniques that are productive.
IBM (Cell)
Multi Core is a trend. Cell's Performance has far away from Intenral Purpose CPU and NVIDIA's GPU behind. But limited to the shortage of Programming Language and development tools, CELL is still unable to shake
Both are in their respective fields. For CPUs like Cell, the largest Concern is Power Consumption.
Almost all chip design companies in the industrial community now reach a consensus, the best way to reduce power consumption
It is to improve the PEAK Performance of the CPU, optimized by parallel, so that the software can be better Match hardware
The parallelism that can be provided, so that the hardware can run on a lower clock rate.
For example, Cell, its Clock Rate will be 4GHz, but if the program can be very good, it does not need to run
At 4GHz, perhaps running at hundreds of MHz, you can achieve the performance of the competitor's similar products in parallel to the multicore (in the original words: Beat the Crap Out of the Competitors), and power consumption can be greatly reduced.
Among them, the people who mentioned SONY caught CRAZY insisted that IBM's CELL and follow-up products should be good support.
Double Floating Point Data (Note: All GPUs, or processors for Video GAME
Only 32-bit floating point data is supported. IBM people, very reluctant, but Sony gets people
It is believed that they can develop the clarity of the physical world that is able to understand the human eye in the near future.
Image model, and 64-bit floating point data is necessary.
NVIDIA
GPU performance will continue to increase. The current GPU does not support parallel processing units to communicate with each other, and future GPU development is more developed towards Shared Multi Processor direction. In addition, Power Consumption is also
NVIDIA's largest convern.
Intel
Some about relatively disappointed if the study of Academia in recent years. Probably, academic resarch is exceeded
