happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Vulkan: Default to 1GB allocations instead of 4GB to avoid fragmentation and driver issues

This commit is contained in:
0cc4m 2025-03-17 19:41:11 +00:00
parent b1b132efcb
commit 90f17bba01
1 changed files with 1 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@ -2512,13 +2512,9 @@ static vk_device ggml_vk_get_device(size_t idx) {
if (GGML_VK_SUBALLOCATION_BLOCK_SIZE != nullptr) { if (GGML_VK_SUBALLOCATION_BLOCK_SIZE != nullptr) {
device->suballocation_block_size = std::stoul(GGML_VK_SUBALLOCATION_BLOCK_SIZE); device->suballocation_block_size = std::stoul(GGML_VK_SUBALLOCATION_BLOCK_SIZE);
#if defined(_WIN32) } else {
} else if (device->vendor_id == VK_VENDOR_ID_NVIDIA) {
// Limit batching of allocations to 1GB by default to avoid fragmentation issues // Limit batching of allocations to 1GB by default to avoid fragmentation issues
device->suballocation_block_size = 1024*1024*1024; device->suballocation_block_size = 1024*1024*1024;
#endif
} else {
device->suballocation_block_size = device->max_memory_allocation_size;
} }
device->suballocation_block_size = std::min(device->suballocation_block_size, device->max_memory_allocation_size); device->suballocation_block_size = std::min(device->suballocation_block_size, device->max_memory_allocation_size);