Resource network
General-purpose CPU capacity — the work that does not need a GPU and should not pay for one.
Metered in vCPU-seconds
The same resource, the same meter, three answers to “whose machine is this?” You choose per workload, and you can move a workload inward when the rules tighten.
On your device
Available
Inside your gram
Available
On a stranger’s machine — not available
Available
Each node benchmarks itself and signs the result with its own key. Where a benchmark does not exist yet, the node says so and emits nothing in its place. So do we.
Signed metric
cpu.throughput
Unit
hashes/s
The node hashes a fixed 1 MB buffer with SHA-256 for a fixed duration on one core, then repeats it on every logical core at once, and signs both — with its core count, architecture and operating system inside the same signature.
The headline is single-threaded by design, so it compares two machines fairly regardless of core count. The whole machine is measured too, and signed beside it — along with the ratio between them, which is usually nearer 0.6 than 1.0. Cores are not free to add: they share memory bandwidth, cache and a thermal budget.
Most workloads are not model training. Rendering, analysis, transcoding and everything a normal institution runs are CPU-bound, and pricing them against a GPU SKU is how cloud bills get absurd.
Free on your own hardware. The meter only runs when you consume someone else’s.
There is no 24/7 operations centre, because we do not employ one. Instead a node that cannot prove it is healthy is evicted from the ring rather than quietly serving your work. Fail-closed, not fail-silent.
Standard containers on k3s. Nothing to port.
Every resource above reaches your workload through the same substrate, whichever ring it was drawn from.
Each node runs k3s. One orchestration layer schedules all seven resources, so a workload moves between rings without being rewritten.
Workloads that cannot be containerised run as virtual machines on the same cluster, through the open-source KubeVirt project. Same scheduler, same meter.
Standard containers, standard VMs, content-addressed objects, exportable receipts. The cost of leaving is the reason to trust the platform.
Compute is metered in vCPU-seconds. Each unit of work produces a receipt naming the node that performed it and the price it was charged at, on the one ledger the whole platform shares. We publish no hourly rate table on this page, because a price is meaningless without the signed unit it is counting.
See how we price compute →Not a specification, not a vendor sheet. The best verified result any node on this network has signed for compute — and, below it, everything else that node put inside the same signature.
cpu.throughput
2,839 hashes/s
A benchmark reports one number and signs several. These travelled inside the same signature, so they are as verifiable as the figure above — and they are usually the ones that decide whether the machine suits your work.
Instruction set
arm64
arch
amd64 and arm64 do not execute the same SHA-256 at the same speed.
L1 data cache
64
cache_l1_data_kb
The fastest memory on the chip. This benchmark’s 1 MB buffer does not fit; almost nothing real does.
L1 instruction cache
128
cache_l1_instruction_kb
Where the code itself lives while it runs.
L2 cache
4,096
cache_l2_unified_kb
Per-core, and a few times slower than L1.
Logical cores
10
cores_logical
Including hyper-threads, which are not whole cores and do not behave like them.
Physical cores
10
cores_physical
Real cores, as opposed to hardware threads.
Processor
Apple M4
cpu_model
Two machines can post identical throughput and be nothing alike — one a laptop chip at its thermal limit, the other a server part idling.
Sockets
1
cpu_sockets
Physical processor packages. Two sockets means two pools of memory, and a thread that reaches across pays a latency penalty no benchmark headline will show.
Duration
500
duration_ms
How long the benchmark ran. Short runs flatter a processor that throttles under sustained load.
All cores at once
18,239
multicore_hashes_per_sec
The same hash on every logical core simultaneously. This is the whole machine working, and it is the number a scheduler can actually spend.
Operating system
darwin
os
The scheduler and the filesystem underneath every number on this page.
Work performed
sha256(1MB) single-threaded
payload
Exactly what the processor was asked to do.
Scaling efficiency
0.642
scaling_efficiency
All-cores throughput divided by (one core × core count). 1.0 would mean cores are free to add. They never are: they share memory bandwidth, cache and a thermal budget. This ratio is how much of the core count you paid for actually arrives.
Sustained throughput
0.883
sustained_ratio
One core’s speed measured again after every core has been busy, divided by its speed when cold. Near 1.0 means the machine holds its clocks under load; well below means it overheats and slows — which a short benchmark would never have shown you. No thermal sensor is required to see it, which matters, because most machines will not give us one.
Threads per core
1
threads_per_core
Above 1 means SMT: two threads sharing one core’s execution units. It is why a “16-core” machine rarely does sixteen cores of work, and it is most of the reason the scaling efficiency above is not 1.0.
Temperature, cache sizes and socket count appear only on Linux nodes. Other platforms hide them behind privileged interfaces, and an unread field is left out rather than estimated.
Nothing recorded against compute. That is a statement about our backlog, not a claim of completeness — the full list lives on the proof page.