Seymour Cray was famous for packing, powering, and cooling circuits incredibly densely. Classic Crays were made obsolete by microprocessors, but we may yet do similar things at a larger scale. Hyperscale data centers and even national supercomputers are loosely coupled things \
13
20
1
334
today, but if challenges demanded it, there is a world with a zetta scale, tightly integrated, low latency matrix dissipating a gigawatt in a swimming pool of circulating fluorinert.

Aug 31, 2020 · 3:51 AM UTC

19
5
8
228
Replying to @ID_AA_Carmack
Have you been following the wafer-scale stuff that one company is working on, Cerebras?
1
Yes, it is interesting!
Replying to @ID_AA_Carmack
Surely big multiplayer sims are the killer apps for this architecture. Putting 12M players in a shared simulation with “just normal gameplay coding” requires something higher bandwidth and lower latency than Ethernet switches in data centers.
12
4
1
101
RDMA infiniband could probably get that job done today -- sub microsecond hops! A few serious low level people doing the local environment gathering primitives, and you could have a lot of normal-looking code running.
8
1
43
more replies
Replying to @ID_AA_Carmack
John did anyone tweet you about that recently found stash of the widescreen CRTs you used to use? We were talking about it over on Shacknews the other day since it made its way on to reddit. I thought you might buy one if you could find one again, others disagree.
2
1
8
I did not see that! I like hearing about finding stashes of old-new-stock, but the only monitor I have any sentimental feelings for is the original NeXT grey scale monitor.
1
1
1
10
more replies
Replying to @ID_AA_Carmack
What are your thoughts on Dojo, Tesla's supercomputer?
Replying to @ID_AA_Carmack
Yeah, but you can't service it and replace dead boards. And sure, you say the MTBF is huge, but take any length of time and apply the birthday problem to 100k nodes and you constantly have a whole bunch that need replacing.
6
7
Replying to @ID_AA_Carmack
This was the original premise of The Matrix
Replying to @ID_AA_Carmack
Still isn't going to beat Laundauer's limit. It has to be A: neuromorphic with memory at the point of computation. B: Quantum C: Time reversible billiard ball style computation. My bet is neuromorphic.
Load more