3 ms·
> a 37 Ghz CPU with 16 TB of RAM now. Something common in the hypervisor admin space is expressing the compute capacity by multiplying the core count with the
by jiggawatts 5y ago
> a 37 Ghz CPU with 16 TB of RAM now.
Something common in the hypervisor admin space is expressing the compute capacity by multiplying the core count with the clock speed.
So for example a 64 core EPYC at 2.5GHz is written down in documentation as a 160 GHz processor.
So these Onyx desktops are directly comparable to artists using a high-end Mac Pro or Threadripper workstation.
- zozbot234 5y agoOTOH, those machines were already quad processors (at the high end). And I don't know if you can feasibly have 16TB of RAM in a single compute node. In that case you'd need a rack-scale multi-node system like the gear the Oxide folks are pushing for "hyperscaler" workloads. (My guess is that such a thing could be made useful to a single user - along the lines of Alan Kay's quote referenced in a sibling comment - but it would likely need to be something that involves chewing through humongous amounts of data, to really make use of that scale. Not sure if the art-creation use case has any real need for that nowadays. Some sort of compute-heavy data analytics for decision support and the like would be a lot closer to the mark.)
- jiggawatts 5y agoYou can have NVMe disk arrays that are faster than the main memory speed of even quite recent desktop PCs these days for reasonable money. My laptop can do 7 GB/s from one drive. I've seen demos of 30GB/s by simply using a 4-port PCIe x16 adapter card. There isn't even a need to venture into exotic technologies like NVRAM DIMMs..
- tomxor 5y agoIf you want to use that metric then the Onyx would be: 4x 250MHz R4400s + 12x 40MHz Intel i860s = 1480 MHz when maxed out. But clock speed is a very poor metric for comparing older CPUs. The size of these CPUs were much smaller, the number of instructions they could cram into each clock cycle and complexity of those instructions was no doubt far lower, and the variety of instructions and ability to specialise far lower (i'm not even considering the whole RISC and MIPS thing, just the fact that they are working with much smaller dies and far fewer, larger gates)... instructions per second might be a better comparison, but then you still lack that qualitative difference in the variety of instruction as i said.