From d326644d8b9f35201adbf9dc16fad94484384e96 Mon Sep 17 00:00:00 2001 From: psychocrypt Date: Fri, 17 Nov 2017 10:04:16 +0100 Subject: fix nvidia auto suggestion The lmem is still incalculably and crash the miner very often. Increase the potential lmem usage to 16kiB to respect lmem alignments, ... --- xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu b/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu index 9923cb2..12c935a 100644 --- a/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu +++ b/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu @@ -467,9 +467,9 @@ extern "C" int cuda_get_deviceinfo(nvid_ctx* ctx) // 200byte are meta data memory (result nonce, ...) size_t availableMem = freeMemory - (128u * byteToMiB) - 200u; size_t limitedMemory = std::min(availableMem, maxMemUsage); - // up to 1kibyte extra memory is used per thread for some kernel (lmem/local memory) + // up to 16kibyte extra memory is used per thread for some kernel (lmem/local memory) // 680bytes are extra meta data memory per hash - size_t perThread = hashMemSize + 1024u + 680u; + size_t perThread = hashMemSize + 16192u + 680u; size_t max_intensity = limitedMemory / perThread; ctx->device_threads = max_intensity / ctx->device_blocks; // use only odd number of threads -- cgit v1.1