From d3b8a0c81b41c4ba5ef8a5a3e0e33925fad139dd Mon Sep 17 00:00:00 2001 From: Toiletslayer <221773375+toiletslayer@users.noreply.github.com> Date: Wed, 7 Oct 2026 11:58:19 -0400 Subject: [PATCH 1/3] Add outside hardware testing guide --- HARDWARE_TESTING.md | 129 ++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 129 insertions(+) create mode 100644 HARDWARE_TESTING.md diff --git a/HARDWARE_TESTING.md b/HARDWARE_TESTING.md new file mode 100644 index 0000000..a99d2ce --- /dev/null +++ b/HARDWARE_TESTING.md @@ -0,0 +1,129 @@ +# Hardware Testing + +The most useful outside contribution to this project right now is **hardware +diversity**. + +The existing correctness work is heavily validated on NVIDIA/Windows. Results +from AMD, Intel, Apple, other NVIDIA architectures, different drivers, and +different current browsers help determine which observations are general and +which are machine-specific. + +## What this test is + +This is a local browser compute test. + +It does **not**: + +- connect to a mining pool; +- connect to a wallet; +- submit blocks; +- send CAP; +- run hidden/background mining. + +GPU work starts only when the tester explicitly starts a verification or +profiling action. + +## Easiest test + +Open the deployed lab: + + + +Use a current browser with WebGPU enabled. + +Run the full Core-vector correctness verification. + +The important result is: + +```text +294 / 294 +0 mismatches +``` + +A failure is useful evidence too. Do not retry repeatedly until it disappears; +record the first reproducible failure and the environment. + +## What to record + +Please record: + +- operating system and version; +- browser and version; +- GPU/adapter name; +- GPU vendor; +- driver version when easily available; +- whether WebGPU was available normally or required an experimental flag; +- full Core-vector result; +- mismatch count; +- pipeline/device error text, if any; +- optional exported result JSON. + +Do not include account credentials, wallet data, browser profiles, or unrelated +system logs. + +## Optional performance test + +Only after correctness passes, a tester may run the synthetic profiling or +workgroup comparison workflows. + +Performance is secondary. + +Browser timing can vary because of: + +- browser scheduling; +- CPU validation time; +- driver behavior; +- thermal/power state; +- background system load. + +A high-variance result is valid evidence. Do not turn a noisy result into a +claimed recommendation. + +## Result template + +```text +Result: PASS / FAIL / INCOMPLETE +OS: +Browser: +GPU: +Vendor: +Driver: +WebGPU available without special flag: yes / no + +Core vectors: + matched: + total: 294 + mismatches: + +Pipeline/device error: + +Optional profiling: + workflow: + samples: + mean throughput: + variability/CV: + recommendation produced by app: yes / no + +Notes: +``` + +## Priority hardware + +Especially useful environments include: + +1. AMD Radeon on Windows or Linux; +2. Intel Arc / Intel integrated graphics; +3. Apple Silicon / Safari or another WebGPU-capable browser; +4. NVIDIA architecture other than the currently recorded Blackwell result; +5. Linux on any supported discrete GPU. + +One clean correctness result from a new hardware/browser family is more useful +than many repeated performance runs on the already-tested machine. + +## Pass meaning + +A 294/294 pass means the tested browser/GPU stack produced outputs identical to +the committed CapStash Core vectors for this verification workload. + +It does **not** prove profitable mining performance, universal browser support, +production mining safety, or compatibility with every driver version. From 95bc7dfe4e89e52fb95a51212bde9ba674a9d6cf Mon Sep 17 00:00:00 2001 From: Toiletslayer <221773375+toiletslayer@users.noreply.github.com> Date: Wed, 7 Oct 2026 11:58:21 -0400 Subject: [PATCH 2/3] Add hardware test result template --- HARDWARE_TEST_RESULT_TEMPLATE.md | 33 ++++++++++++++++++++++++++++++++ 1 file changed, 33 insertions(+) create mode 100644 HARDWARE_TEST_RESULT_TEMPLATE.md diff --git a/HARDWARE_TEST_RESULT_TEMPLATE.md b/HARDWARE_TEST_RESULT_TEMPLATE.md new file mode 100644 index 0000000..8f8403c --- /dev/null +++ b/HARDWARE_TEST_RESULT_TEMPLATE.md @@ -0,0 +1,33 @@ +# Hardware Test Result + +- Result: `PASS | FAIL | INCOMPLETE` +- Test date (UTC): +- Repository/deployment version or commit, if known: +- Operating system: +- Browser/version: +- GPU/adapter: +- Vendor: +- Driver: +- WebGPU available without special flag: `yes | no` + +## Correctness + +- Core vectors matched: +- Core vectors total: `294` +- Mismatches: +- Pipeline/device error: + +## Optional profiling + +- Workflow: +- Samples: +- Mean throughput: +- Median throughput: +- Variability/CV: +- Application recommendation: `winner | no recommendation | not run` + +## Notes + +Record reproducible browser/device behavior here. Do not include wallet +information, account credentials, browser-profile data, or unrelated system +logs. From 95635c30217486a21892f1c09ece75f3308f2e75 Mon Sep 17 00:00:00 2001 From: Toiletslayer <221773375+toiletslayer@users.noreply.github.com> Date: Wed, 7 Oct 2026 11:58:24 -0400 Subject: [PATCH 3/3] Link outside hardware test guide --- README.md | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/README.md b/README.md index 8b50eaa..0d2ed19 100644 --- a/README.md +++ b/README.md @@ -158,6 +158,13 @@ loads idle with no automatic GPU workload. See [DEPLOYMENT.md](./DEPLOYMENT.md). - Already-submitted WebGPU dispatches cannot necessarily be interrupted, so the interface does not claim fake cancellation. +## Outside hardware testing + +Results from hardware/browser families not represented in the existing evidence +are especially useful. Start with correctness, not speed. See +[HARDWARE_TESTING.md](./HARDWARE_TESTING.md) and +[HARDWARE_TEST_RESULT_TEMPLATE.md](./HARDWARE_TEST_RESULT_TEMPLATE.md). + ## Reproducibility and tests ```powershell