I’ve worked on several systems that had enough flop/s to make it in the top 5-10, but for which we never submitted benchmarks. Sometimes their backend network layout technically would make them several smaller clusters for an HPL run, sometimes it’s because the cluster is too heterogeneous to get a good benchmark result, and sometimes it’s because the employer wants to keep a low profile.
Most of the time, it just that it’s a hassle. It takes a while to prep and tune a big hero run for benchmarking, and if you spend a billion dollars on a cluster, it’s making you a lot more than that. Taking it down for a day or two stops the money printers.
You still need to traverse physical segments in the wireless path: think receiving dish to the next transmitting dish, the end of the path to get from the trading systems onto the roof and into to the first dish, etc. Every nanosecond counts.
> I'm sure some HFT traders are going to make an absolute fortune, but I doubt it'll have a huge impact for most other people.
They’ve been using hollow core fiber (and funding research into it) for nearly a decade. I know it goes back further than the 2017 spinoff mentioned in the article, but https://optics.org/news/11/9/52 talks about it a bit.
> But what about Jump Capital? They have a crypto division that also does market making. The difference here is there are given a large chunk of tokens to market make with, as they please.
It may be a little early to make that comparison. Jump is still being investigated for its crypto shenanigans.
I think it’s a way of keeping the museum’s single most popular piece of art on display whilst working it. I think most museums would remove it for a while, but so many people come specifically to see this painting that they want to keep it viewable, so they make a little show of its restoration.
I dunno; I’ve been through that floor 5 or 6 times since they started work, and people always seem to love the spectacle of it.
Or have them set up in a way that makes them hard to run full-system benchmarks on. I can think of a couple of financial firms that have clusters that would rate on the Top 10, but a) as you suspected, they’re too busy running the money printers to take them down for a few days to run benchmarks, and b) they’re split up into more manageable little clusters, so the high speed fabrics don’t connect and allow every node to talk to every other node, which you need for an HPL run.
Most of the time, it just that it’s a hassle. It takes a while to prep and tune a big hero run for benchmarking, and if you spend a billion dollars on a cluster, it’s making you a lot more than that. Taking it down for a day or two stops the money printers.