>The Cray-1 team hopes to fill the currently empty second half of the tower through a crowdfunding campaign to secure 70–80 additional M4 Mac Minis. The goal: create a cluster capable of ranking on the TOP500 supercomputers list.
Is ~100 Mac Minis really all it takes to get into the TOP500 these days?
I don't think so, the lowest from 2026 was still 2.66 PFlop/s, so they'd need thousands of Mac Minis. The M4s in the Minis aren't on a different level than the Xeons in those old clusters (outside of power consumption), even if we let the Macs use FP32.
Pretty sure not, especially since it requires 64-bit floating-point, so the GPU cores don't help. #500 right now is 2PF. One Mac won't do anywhere near 20 TF. Google tells me you can get 350GF of DP from one 10-core M4, so you might need 10,000 to get there. Then there's the network...
Rather than rent it out (maybe some people would pay a few bucks a year for novelty web/mail hosting, but it's both too much and too little for most use cases, running the program of renting it probably eats up a lot of the revenue), maybe they could give compute time to universities for free, with the requirement that they provide some sort of display for museum visitors? There's probably some people with Monte Carlo simulations that could benefit from a lot of cores (if they get the M4 upgrade) while being on a student budget of $0. And if no one has requested it, they could run some BOINC project that has a nice visualization.
The Cray-1 was measured on 64-bit FLOPS, so you would have to do apples-to-apples (heh). A modern iPhone can do 300 GF of 64-bit, so only 2000x.
The iPhone 4 was probably the first that could beat a Cray-1, and it was by a narrow margin. However, that was only a 32-bit chip. The next year iPhones went 64-bit, and it was all over.
If you were an engineer working on this machine, it was visually clear to you that the modularity took up the majority of the space. The ICs themselves were already pretty dense.
Isn't supercomputing more about the data link speed than about the actual node compute speed?
I know it's pedantic but this is a tribute or a re-envisioning but it's not a replica.
>The Cray-1 team hopes to fill the currently empty second half of the tower through a crowdfunding campaign to secure 70–80 additional M4 Mac Minis. The goal: create a cluster capable of ranking on the TOP500 supercomputers list.
Is ~100 Mac Minis really all it takes to get into the TOP500 these days?
I don't think so, the lowest from 2026 was still 2.66 PFlop/s, so they'd need thousands of Mac Minis. The M4s in the Minis aren't on a different level than the Xeons in those old clusters (outside of power consumption), even if we let the Macs use FP32.
Pretty sure not, especially since it requires 64-bit floating-point, so the GPU cores don't help. #500 right now is 2PF. One Mac won't do anywhere near 20 TF. Google tells me you can get 350GF of DP from one 10-core M4, so you might need 10,000 to get there. Then there's the network...
They could rent out all that compute to subsidy the museum running costs.
Rather than rent it out (maybe some people would pay a few bucks a year for novelty web/mail hosting, but it's both too much and too little for most use cases, running the program of renting it probably eats up a lot of the revenue), maybe they could give compute time to universities for free, with the requirement that they provide some sort of display for museum visitors? There's probably some people with Monte Carlo simulations that could benefit from a lot of cores (if they get the M4 upgrade) while being on a student budget of $0. And if no one has requested it, they could run some BOINC project that has a nice visualization.
Yeah all good ideas, hope they take up some of these.
I doubt distributed compute in a bunch of old i7s is worth much once you factor in the electricity cost
The RAM shortage is bad, but not that bad (yet).
You think it's bad now?
New 48GB DDR5 ECC UDIMM chips are retailing for £2.5K each in the UK.
imagine showing even the first iphone to a Cray engineer in 1976
they'd lose their minds lol
(Cray-1 was 160 megaflops, smartphones are teraflops)
The Cray-1 was measured on 64-bit FLOPS, so you would have to do apples-to-apples (heh). A modern iPhone can do 300 GF of 64-bit, so only 2000x.
The iPhone 4 was probably the first that could beat a Cray-1, and it was by a narrow margin. However, that was only a 32-bit chip. The next year iPhones went 64-bit, and it was all over.
There is also the issue of cooling. The iPhone will need extra cooling in order to function at full speed for more than a few minutes.
I don't think they would be that surprised. Look at the inside of the Cray-1: https://thumbs.dreamstime.com/z/inside-red-yellow-cray-super...
If you were an engineer working on this machine, it was visually clear to you that the modularity took up the majority of the space. The ICs themselves were already pretty dense.
The picture with all the cables everywhere... ugh that sets off my OCD.