If they send billions of dollars of GPU cards into space, how are they going to secure it, physically?
Starcloud
131–140 of 234 posts
Re: Starcloud
#132This is absolute nonsense. The first thing to consider is that this thing won’t be stationary! Geosynchronous orbit is much more expensive to reach per kg launched, even for Starship… when it starts working properly. Lower orbits… aren’t stationary. Who wants a data centre that’s “over the horizon” from the owning country most of the time!? If you think AWS egress costs are bad? Just add some zeroes! No, more zeroes…
Why can't it be geostationary? Laser communication can get you gigabit speeds today. That would take a month to transmit GPT-5's estimated 280TB training corpus, which is acceptable. Latency does not matter.
With geostationary orbit you won't ever get less than 200ms round-trip latency from the ground (at the speed of light).
Fine for some applications, but a massive regression from modern fiber infrastructure and definitely not suitable for everything (just think how slow the modern web is even with 15ms connections to datacenters). There's a reason why Starlink & co are trying to set up communication satellites closer to the ground.
Re: Starcloud
#133Last time these folks were mentioned on HN, there was a lot of skepticism that this is really possible to do. The issue is cooling: in space, you can't rely on convection or conduction to do passive cooling, so you can only radiate away heat. However, the radiator would need to be several kilometers big to provide enough cooling, and obviously launching such a large object into space would therefore eat up any cost s…
Not sure if I follow really. Cooling from it's own generated heat? Are we even sure the system would get that hot in the first place? The temperatures can plunge up to -200 degrees. If needed, they'd cool it just like they keep the James Webb Telescope cool.
Keeping things cool in space is very hard. On earth we usually transfer heat from one medium to another (water to water, water to air, etc.). In space that's not possible because even though the matter in space is quite cold, there is very little. Therefore the only real way to get rid of heat in space is to radiate it away (think infrared light bulb). The James Webb Telescope does the same thing.
Re: Starcloud
#134Last time these folks were mentioned on HN, there was a lot of skepticism that this is really possible to do. The issue is cooling: in space, you can't rely on convection or conduction to do passive cooling, so you can only radiate away heat. However, the radiator would need to be several kilometers big to provide enough cooling, and obviously launching such a large object into space would therefore eat up any cost s…
Not sure if I follow really. Cooling from it's own generated heat? Are we even sure the system would get that hot in the first place? The temperatures can plunge up to -200 degrees. If needed, they'd cool it just like they keep the James Webb Telescope cool.
The Webb telescope is a _wildly_ different apparatus, designed from the ground up to run as cool as possible, and with an effectively unlimited budget. It lives in the shadow of the Earth behind multiple layers of shielding. These "data centers" need to live in direct sunlight and operate as cheaply as possible _at scale._ Very little of Webb's tech is applicable.
Re: Starcloud
#135Earlier quoted context omitted.
> This seems bad e.g., Cianide seems bad, but it won't kill you if the relative volumes are small. tl;dr: You haven't characterized the denominator.
See my edit. Just one starcloud would represent an increase in a risk factor of over 300 c.f. status quo. Then multiply that by the number of starclouds you think would be deployed.
You still keep playing with the numerator.
> increase in a risk factor of over 300
Even with a numerator-only view, I suspect it's not fair to characterize the "risk factor" as going up 300x. There's a lot more nuance about orbits in space.
Re: Starcloud
#136Earlier quoted context omitted.
I wonder if "normal" RDIMM ECC would be enough to mitigate most of those radiation bit-flipping issues. If so it wouldn't really make a difference to earth-based servers since most enterprise servers use RDIMM ECC too
You'll get bitflips elsewhere besides just in RAM. A bitflip in L1 or L3 cache will be propagated to your DIMM and noone will be the wiser.
Those do ECC already
Re: Starcloud
#137Everyone who puts up a persistent bright dot in the night sky should compensate everyone who has to see it with 1 cent for the sensory pollution.
Re: Starcloud
#138> In 10 years, nearly all new data centers will be being built in outer space,” Johnston predicts. Can I bet on the contrary odds? Could throw down my whole retirement with confidence
Re: Starcloud
#139Shameful to see this on Nvidia's site. They have real engineers and business prowess. This is really shaking my assumptions about the company.
Why is this shameful?
Re: Starcloud
#140Earlier quoted context omitted.
See my edit. Just one starcloud would represent an increase in a risk factor of over 300 c.f. status quo. Then multiply that by the number of starclouds you think would be deployed.
You still keep playing with the numerator. > increase in a risk factor of over 300 Even with a numerator-only view, I suspect it's not fair to characterize the "risk factor" as going up 300x. There's a lot more nuance about orbits in space.
Tell me the nuance then. If people have concerns about Kessler syndrome at the starlink scale then why wouldn't something literally 1000x bigger be even more concerning.