2 ms·

In what sense? The ranking is still based on LINPACK scores, and it's intended to represent the fastest machines for scientific computing. We can argue about

In what sense? The ranking is still based on LINPACK scores, and it's intended to represent the fastest machines for scientific computing. We can argue about whether modern simulation codes look like LINPACK (they don't) but neither Google's nor Amazon's cloud is competitive for that type of workload. Even if the clouds are physically bigger, nobody at Google or Amazon is running really big numerical jobs of the sort that DOE does. Many are running smaller scale HPC and ML jobs, but it's expensive, and for latency-bound physics simulations (most of what we run at LLNL, ORNL, etc.), cloud networks will not perform at scale. The question of whether China's machines ever performed well for big multi-physics applications is up in the air, but they were certainly beating us on LINPACK. The Chinese set out to show that they're serious about HPC, and I think they accomplished at least that. I wouldn't be surprised if you saw this race go back and forth a few times in the coming years.

Retric 8y ago
As you say LINPACK is mostly useless as a benchmark. But it's a wider problem than that. Interconnects expand the possible workloads, but that's always job specific. A single overclocked CPU would crush this for a huge range of tasks, but that in no way means it's not a super computer. So, sure this thing is going to be useful, but Google still has vastly more compute and Google uses it for more economically valuable jobs.
jcranmer 8y ago
The problem with dense linear algebra as a metric of supercomputer performance is that linear algebra tends to have relatively little communication compared to the computation. This means that the architecture you can use to really optimize linear algebra don't necessarily generalize well to different kinds of HPC code. China's architectures have definitely been accused of this in the past, but it looks like some of their newer stuff at least works fairly well on more HPC code.

Retric 8y ago

As you say LINPACK is mostly useless as a benchmark. But it's a wider problem than that. Interconnects expand the possible workloads, but that's always job specific. A single overclocked CPU would crush this for a huge range of tasks, but that in no way means it's not a super computer. So, sure this thing is going to be useful, but Google still has vastly more compute and Google uses it for more economically valuable jobs.

jcranmer 8y ago

The problem with dense linear algebra as a metric of supercomputer performance is that linear algebra tends to have relatively little communication compared to the computation. This means that the architecture you can use to really optimize linear algebra don't necessarily generalize well to different kinds of HPC code. China's architectures have definitely been accused of this in the past, but it looks like some of their newer stuff at least works fairly well on more HPC code.