fix: ignore compute_cap if not present (#1866)

* fix: ignore compute_cap if not present * fix: correct gpu info * fix: remove check for toolkit version --------- Co-authored-by: vansangpfiev <[email protected]>
janhq · Jan 17, 2025 · 68094d9 · 68094d9
1 parent 0707069
commit 68094d9
Show file tree

Hide file tree

Showing 4 changed files with 25 additions and 23 deletions.
diff --git a/engine/services/engine_service.cc b/engine/services/engine_service.cc
@@ -404,17 +404,6 @@ cpp::result<bool, std::string> EngineService::DownloadCuda(
   auto suitable_toolkit_version =
       GetSuitableCudaVersion(engine, hw_inf_.cuda_driver_version);
 
-  // compare cuda driver version with cuda toolkit version
-  // cuda driver version should be greater than toolkit version to ensure compatibility
-  if (semantic_version_utils::CompareSemanticVersion(
-          hw_inf_.cuda_driver_version, suitable_toolkit_version) < 0) {
-    CTL_ERR("Your Cuda driver version "
-            << hw_inf_.cuda_driver_version
-            << " is not compatible with cuda toolkit version "
-            << suitable_toolkit_version);
-    return cpp::fail("Cuda driver is not compatible with cuda toolkit");
-  }
-
   auto url_obj = url_parser::Url{
       .protocol = "https",
       .host = jan_host,

diff --git a/engine/utils/hardware/gpu_info.h b/engine/utils/hardware/gpu_info.h
@@ -17,7 +17,7 @@ inline std::vector<GPU> GetGPUInfo() {
 
   for (size_t i = 0; i < nvidia_gpus.size(); i++) {
     for (size_t j = 0; j < vulkan_gpus.size(); j++) {
-      if (nvidia_gpus[i].uuid == vulkan_gpus[j].uuid) {
+      if (nvidia_gpus[i].uuid.find(vulkan_gpus[j].uuid) != std::string::npos) {
         vulkan_gpus[j].version =
             nvidia_gpus[0].cuda_driver_version.value_or("unknown");
         vulkan_gpus[j].add_info = NvidiaAddInfo{

diff --git a/engine/utils/system_info_utils.cc b/engine/utils/system_info_utils.cc
@@ -108,26 +108,32 @@ std::vector<GpuInfo> GetGpuInfoList() {
     auto [driver_version, cuda_version] = GetDriverAndCudaVersion();
     if (driver_version.empty() || cuda_version.empty())
       return gpuInfoList;
-
+    bool need_fallback = false;
     CommandExecutor cmd(kGpuQueryCommand);
     auto output = cmd.execute();
+    if (output.find("NVIDIA") == std::string::npos) {
+      need_fallback = true;
+      output = CommandExecutor(kGpuQueryCommandFb).execute();
+    }
 
-    const std::regex gpu_info_reg(kGpuInfoRegex);
+    std::string rg = need_fallback ? kGpuInfoRegexFb : kGpuInfoRegex;
+    const std::regex gpu_info_reg(rg);
     std::smatch match;
     std::string::const_iterator search_start(output.cbegin());
+    int rg_count = need_fallback ? 5 : 6;
 
     while (
         std::regex_search(search_start, output.cend(), match, gpu_info_reg)) {
       GpuInfo gpuInfo = {
-          match[1].str(),              // id
-          match[2].str(),              // vram_total
-          match[3].str(),              // vram_free
-          match[4].str(),              // name
-          GetGpuArch(match[4].str()),  // arch
-          driver_version,              // driver_version
-          cuda_version,                // cuda_driver_version
-          match[5].str(),              // compute_cap
-          match[6].str()               // uuid
+          match[1].str(),                        // id
+          match[2].str(),                        // vram_total
+          match[3].str(),                        // vram_free
+          match[4].str(),                        // name
+          GetGpuArch(match[4].str()),            // arch
+          driver_version,                        // driver_version
+          cuda_version,                          // cuda_driver_version
+          need_fallback ? "0" : match[5].str(),  // compute_cap
+          match[rg_count].str()                  // uuid
       };
       gpuInfoList.push_back(gpuInfo);
       search_start = match.suffix().first;

diff --git a/engine/utils/system_info_utils.h b/engine/utils/system_info_utils.h
@@ -25,6 +25,13 @@ constexpr static auto kGpuQueryCommand{
 constexpr static auto kGpuInfoRegex{
     R"((\d+),\s*(\d+),\s*(\d+),\s*([^,]+),\s*([\d\.]+),\s*([^\n,]+))"};
 
+constexpr static auto kGpuQueryCommandFb{
+    "nvidia-smi "
+    "--query-gpu=index,memory.total,memory.free,name,uuid "
+    "--format=csv,noheader,nounits"};
+constexpr static auto kGpuInfoRegexFb{
+    R"((\d+),\s*(\d+),\s*(\d+),\s*([^,]+),\s*([^\n,]+))"};
+
 struct SystemInfo {
   explicit SystemInfo(std::string os, std::string arch)
       : os(std::move(os)), arch(std::move(arch)) {}