bench: record the memory clock alongside the SM clock (schema 2) #18

grenade · 2026-09-03T15:47:29Z

grenade commented

2026-09-03 15:47:29 +00:00

The memory clock is a confound like the power limit: locking it at 810 MHz gave +13% on the 4090 at the same watts (lair/quantus#10, applied to benjy and quadbrat). Records from before and after that change must be distinguishable, so the harness now captures clocks.mem in the GPU state and prints it in the summary. Record schema bumps to 2.

The bench run on this PR is also the first full-protocol measurement with the lock in place on benjy; the number to compare is 408 MH/s from actions/runs/33.

🤖 Generated with Claude Code

https://claude.ai/code/session_01CBgs2nSi4H2mdh8kD8vMX5

The memory clock is a confound like the power limit: locking it at 810 MHz gave +13% on the 4090 at the same watts (lair/quantus#10, applied to benjy and quadbrat). Records from before and after that change must be distinguishable, so the harness now captures `clocks.mem` in the GPU state and prints it in the summary. Record schema bumps to 2. The bench run on this PR is also the first full-protocol measurement with the lock in place on benjy; the number to compare is 408 MH/s from actions/runs/33. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_01CBgs2nSi4H2mdh8kD8vMX5

grenade added 1 commit 2026-09-03 15:47:30 +00:00

bench: record the memory clock alongside the SM clock (schema 2)

ci / fmt (pull_request) Successful in 36s

Details

bench / build (pull_request) Successful in 1m37s

Details

ci / doc (pull_request) Successful in 1m42s

Details

ci / clippy (pull_request) Successful in 2m5s

Details

bench / measure (pull_request) Successful in 3m48s

Details

ci / test (pull_request) Successful in 7m3s

Details

a7a8ff0532

The memory clock is a confound like the power limit: locking it at 810 MHz
gave +13% on the 4090 at the same watts (lair/quantus#10). Records from
before and after that change must be distinguishable.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CBgs2nSi4H2mdh8kD8vMX5

gitea-actionsbot commented

2026-09-03 15:53:46 +00:00

quantus-bench `a7a8ff0532` on benjy.hanzalova.internal

worker	median MH/s	spread	windows
0	452.15	0.2%	452.1, 452.6, 452.9, 451.9, 452.2

total median: 452.15 MH/s (batch 1000000, 5 x 30s, engine gpu-cuda)

gpu 0 NVIDIA GeForce RTX 4090: driver 580.173.02, limit 250 W, draw 247 W, sm 2175 MHz, mem 810 MHz, 59 C
parity: OK (25/25 jobs found solutions, all verified against CPU)
--- resume miner ---
active

## quantus-bench a7a8ff0532a27b978b421e5da65fa8be99dc10d6 on benjy.hanzalova.internal | worker | median MH/s | spread | windows | | --- | --- | --- | --- | | 0 | 452.15 | 0.2% | 452.1, 452.6, 452.9, 451.9, 452.2 | total median: **452.15 MH/s** (batch 1000000, 5 x 30s, engine gpu-cuda) - gpu 0 NVIDIA GeForce RTX 4090: driver 580.173.02, limit 250 W, draw 247 W, sm 2175 MHz, mem 810 MHz, 59 C - parity: OK (25/25 jobs found solutions, all verified against CPU) --- resume miner --- active

grenade merged commit f061ccfeea into main

2026-09-03 15:54:48 +00:00

grenade referenced this issue from a commit

2026-09-03 15:54:49 +00:00

Merge pull request 'bench: record the memory clock alongside the SM clock (schema 2)' (#18) from bench/record-memory-clock into main

grenade deleted branch bench/record-memory-clock

2026-09-03 15:54:49 +00:00

Sign in to join this conversation.

2 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: blackbeard/miner#18

bench: record the memory clock alongside the SM clock (schema 2) #18

quantus-bench a7a8ff0532 on benjy.hanzalova.internal

quantus-bench `a7a8ff0532` on benjy.hanzalova.internal