GPU bench table

The table

One row per card on the current class: the class v4 program (the latency-shadow block over the class v3 hash), or a class v3 row re-measured with its class v4 cost on 6 October 2026 or later. Click a column header to sort; the table opens by MH per watt. Integrated GPUs are not listed. The earlier classes sit below, collapsed.

Why the rate fell from the first bench to today. The genesis program did 104 dependent random 4-byte loads per hash over a 1 GiB dataset; the hourly program and class v3 do 128, with the mixer between them; class v4 adds about 100,000 integer operations per hash that ride in the memory wait. So the hash is bound by random-read bandwidth by design, and a card's MH/s is a relative number: the difficulty follows it, and the same card earns the same share of blocks at 136 MH/s on class v3 as it did at 228 MH/s on the genesis program. What a miner compares is hash per watt, and what the chain cares about is the chip edge, which the shadow work is there to cut.

Rows on the current class: 32. Each row names the engineering log entry or the job it came from.

The Hive flight sheet column. Where a card has a measured tune point, the column gives the core clock lock, the memory clock and the power limit to copy into a HiveOS flight sheet (core / mem / PL); the line under each row carries the label with the date, the class v4 cost in full, the miner and driver, the source and the note. Stock means no tune point has been measured yet. The Hive package mines at these settings through Hive's own overclock controls; the desktop app's Ember Tune lands on them by itself.

How a row gets here

Every row names the engineering log entry or the job it came from. "Measured by the team" means our own hardware and our own log. "Reported by the fleet" or "measured by the fleet" means a machine we rent or do not own, read from the status lines its miner uploads or from a bench run on it.

MH per watt needs the card's power draw during the run. The app reads it on NVIDIA cards through the driver. Rows get the figure when a run records it. The class v4 cost column is what the latency-shadow work costs that card against the class v3 control, in watts and rate, where it was measured. The tuned column is the card's state at the row: a full Ember Tune names its point; stock means the card as it came, bench only or mining.

There is no other Igneum miner to compare with yet, so this table compares cards, not miners. The app that produces these rows: the miner page.

Fleet tuning priors

Ember Tune runs on every card the app mines with: the power limit and the core clock are stepped on the live program and the card keeps the point with the best MH per watt within 1% of its top rate. Every finished tune is reported back without anything that identifies the owner, and the fleet's median point per card model, driver major and program class comes back down inside the signed update manifest as the starting point for the next card of that model. A model needs 5 reports before its prior is used.

No tune reports yet. The first rows appear once five machines with the same card model have reported.

Measured by the team

• NVIDIA GeForce RTX 5090 (driver 617, class l128w0, 2026-10-06, job ember-tune-pc1-6): 1854 MHz core cap, power limit 100% (575 W). 84 W saved for 0.15% of rate: 127.71 MH/s at 226.8 W (0.563 MH/W) against 127.9 MH/s at 311 W untuned (0.411 MH/W), 37% more hashes per watt. The chosen clock is the ladder's floor (60% of the maximum), not the optimum; the next cut's ladder goes to 45% and the climb walks on from here.
• NVIDIA GeForce RTX 4070 (driver 617, class l128w0, 2026-10-06, job ember-tune-pc1-6): 1863 MHz core cap, power limit 50% (100 W). 30 W saved for no rate lost: 28.78 MH/s at 75.6 W (0.381 MH/W) against 28.72 MH/s at 106 W untuned (0.271 MH/W), 41% more hashes per watt. The chosen clock is the ladder's floor (60% of the maximum), not the optimum; the next cut's ladder goes to 45%.

Rows: 0. Written from the fleet's tune records at build time.