Fix nonce advancement logic and update CUDA miner env settings
This commit is contained in:
@@ -719,8 +719,9 @@ impl CudaEngine {
|
||||
}
|
||||
// Advance by covered window and continue to next batch, honoring cancel between batches
|
||||
let attempts = active_threads * (effective_iters as u64);
|
||||
// Advance by attempted nonces this batch; CPU handled +1 before the G2 loop
|
||||
hash_count = hash_count.saturating_add(attempts);
|
||||
current = current.saturating_add(U512::from(1u64 + attempts));
|
||||
current = current.saturating_add(U512::from(attempts));
|
||||
if cancel.load(AtomicOrdering::Relaxed) {
|
||||
break;
|
||||
}
|
||||
@@ -807,7 +808,9 @@ impl CudaEngine {
|
||||
}
|
||||
}
|
||||
|
||||
// Not found in this window (no additional advance; already advanced per batch)
|
||||
// Not found in this window
|
||||
// Advance by the single CPU step consumed at the beginning of this outer loop iteration
|
||||
current = current.saturating_add(U512::from(1u64));
|
||||
continue;
|
||||
} else {
|
||||
// G1 launch: computes y for (current + t + 1) for each thread t in [0, num_threads)
|
||||
|
||||
@@ -5,8 +5,9 @@ MINER_CUDA_IMAGE=ptx
|
||||
MINER_CUDA_ITERS=8192
|
||||
MINER_CUDA_THREADS=4096
|
||||
MINER_CUDA_BLOCK_DIM=256
|
||||
MINER_CUDA_BATCHES=32
|
||||
MINER_CUDA_BATCHES=4
|
||||
MINER_CUDA_SAMPLER=1
|
||||
MINER_CUDA_DESIRED_ITERS=1024
|
||||
MINER_CUDA_DESIRED_ITERS=64
|
||||
MINER_CUDA_VERIFY=2048
|
||||
MINER_CUDA_FORCE_WIN=0
|
||||
RUST_LOG=miner=debug
|
||||
|
||||
Reference in New Issue
Block a user