Originally posted by: Pastfinder
I want one of these...
This article says nothing about the performance of this computer. Sure, the info about the number of processors and memory size is nice, but that is not how you evaluate supercomputers.
In the meanwhile:
http://www.theinquirer.net/?article=11831
That fifth team handles two interesting projects: Japan's Open Bioinformatics Grid (OBIGrid), and a
1 PetaFLOPs Molecular Dynamics Machine. This is the first known public PetaFLOPs project,
exceeding in size even the IBM-based ASCI projects in the US announced late last year.
The architecture of this PetaFLOP machine is right now still a secret - more will be known at the SuperComputing 2003 to be held this November in dusty Phoenix, Arizona. To achieve PetaFLOP peak performance with a minimum number of processors, you need to use CPUs that have high peak FP throughput - something like G5 PPC980 (10 GFLOPs Rpeak per 2.5 GHz CPU) or POWER5, maybe? Good memory and I/O bandwidth, coupled with low power and high integration, plus a very fast interconnect with distributed shared memory capability, are the key to achieving such goal.
There could be other PetaFLOP-scale projects around, and we'll be sniffing on them aggressively in a prelude to the SuperComputing 2003 conference.
On the other hand, the Grid project has its usual usage candidates, such as "Embarrassingly Parallel Computation? (easy to parallelise code not too sensitive on bandwidth and latency penalties on grid), distributed resource sharing among organizations, and researcher collaboration
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)
Facebook
Twitter