Microsoft MVP성태의 닷넷 이야기
VC++: 126. CUDA Core 수를 알아내는 방법 [링크 복사], [링크+제목 복사],
조회: 21787
글쓴 사람
정성태 (techsharer at outlook.com)
홈페이지
첨부 파일
 

(시리즈 글이 4개 있습니다.)
VC++: 125. CUDA로 작성한 RGB2RGBA 성능
; https://www.sysnet.pe.kr/2/0/11471

개발 환경 구성: 356. GTX 1070, GTX 960, GT 640M의 cudaGetDeviceProperties 출력 결과
; https://www.sysnet.pe.kr/2/0/11472

개발 환경 구성: 357. CUDA의 인덱싱 관련 용어 - blockIdx, threadIdx, blockDim, gridDim
; https://www.sysnet.pe.kr/2/0/11481

VC++: 126. CUDA Core 수를 알아내는 방법
; https://www.sysnet.pe.kr/2/0/11482




CUDA Core 수를 알아내는 방법

결론부터 말하면, 현재 공식적으로 Cuda Core 수를 알아내는 방법은 없습니다. (혹시 아시는 분은 덧글 부탁드립니다. ^^) cudaGetDeviceProperties로는 multiProcessorCount 까지는 알 수 있지만 core 수는 제공하지 않습니다.

검색해 보면,

How can I get number of Cores in cuda device?
; https://stackoverflow.com/questions/32530604/how-can-i-get-number-of-cores-in-cuda-device

아래의 자료를 인용하면서,

Compute Capabilities
; http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#compute-capabilities

cudaDeviceProp 구조체의 major 버전으로 계산하는 코드가 나옵니다. 그런데... 이게 실제로 과거 NVIDIA에서 제공하던 방식입니다. Cuda SDK 6.0의 예제에 helper_cuda.h 헤더 파일이 있었다는데,

What's the proper way to detect SP/CUDA cores count per SM? 
; https://devtalk.nvidia.com/default/topic/470848/what-39-s-the-proper-way-to-detect-sp-cuda-cores-count-per-sm-/

이 코드를 보면,

helper_cuda.h
; https://people.maths.ox.ac.uk/gilesm/cuda/prac1/helper_cuda.h

다음과 같습니다.

// Beginning of GPU Architecture definitions
inline int _ConvertSMVer2Cores(int major, int minor)
{
    // Defines for GPU Architecture types (using the SM version to determine the # of cores per SM
    typedef struct
    {
        int SM; // 0xMm (hexidecimal notation), M = SM Major version, and m = SM minor version
        int Cores;
    } sSMtoCores;

    sSMtoCores nGpuArchCoresPerSM[] =
    {
        { 0x20, 32 }, // Fermi Generation (SM 2.0) GF100 class
        { 0x21, 48 }, // Fermi Generation (SM 2.1) GF10x class
        { 0x30, 192}, // Kepler Generation (SM 3.0) GK10x class
        { 0x32, 192}, // Kepler Generation (SM 3.2) GK10x class
        { 0x35, 192}, // Kepler Generation (SM 3.5) GK11x class
        { 0x37, 192}, // Kepler Generation (SM 3.7) GK21x class
        { 0x50, 128}, // Maxwell Generation (SM 5.0) GM10x class
        { 0x52, 128}, // Maxwell Generation (SM 5.2) GM20x class
        {   -1, -1 }
    };

    int index = 0;

    while (nGpuArchCoresPerSM[index].SM != -1)
    {
        if (nGpuArchCoresPerSM[index].SM == ((major << 4) + minor))
        {
            return nGpuArchCoresPerSM[index].Cores;
        }

        index++;
    }

    // If we don't find the values, we default use the previous one to run properly
    printf("MapSMtoCores for SM %d.%d is undefined.  Default to use %d Cores/SM\n", major, minor, nGpuArchCoresPerSM[index-1].Cores);
    return nGpuArchCoresPerSM[index-1].Cores;
}

보는 바와 같이, 장치로부터 값을 제공받는 방식이 아닌 버전 별로 구분해 반환하는 방식입니다.




[이 글에 대해서 여러분들과 의견을 공유하고 싶습니다. 틀리거나 미흡한 부분 또는 의문 사항이 있으시면 언제든 댓글 남겨주십시오.]







[최초 등록일: ]
[최종 수정일: 4/10/2018]

Creative Commons License
이 저작물은 크리에이티브 커먼즈 코리아 저작자표시-비영리-변경금지 2.0 대한민국 라이센스에 따라 이용하실 수 있습니다.
by SeongTae Jeong, mailto:techsharer at outlook.com

비밀번호

댓글 작성자
 




[1]  2  3  4  5  6  7  8  9  10  11  12  13  14  15  ...
NoWriterDateCnt.TitleFile(s)
13961정성태7/5/2025139디버깅 기술: 222. WinDbg 분석 사례 - IISreset 시점에 w3wp.exe의 crash 발생
13960정성태7/3/2025484개발 환경 구성: 752. ProcDump - C/C++ 예외 코드 필터를 지정한 덤프 생성
13959정성태6/25/20251295오류 유형: 966. Ubuntu - ping: connect: Network is unreachable
13958정성태6/21/20251624닷넷: 2339. C# - Phi-4-multimodal 모델의 GPU 가속 방법 (ORT 사용)파일 다운로드1
13957정성태6/20/20251472닷넷: 2338. C# / Foundry Local - Phi-4-multimodal 모델을 사용하는 방법
13956정성태6/19/20251656개발 환경 구성: 751. Triton Inference Server의 Python Backend 프로세스
13955정성태6/18/20251779오류 유형: 965. Hugging Face 모델 다운로드 시 "requests.exceptions.HTTPError: 401 Client Error: Unauthorized for url: ..." 오류
13954정성태6/18/20251604닷넷: 2337. C# - Hugging Face에 공개된 LLM 모델을 Foundry Local에서 사용하는 방법파일 다운로드1
13953정성태6/16/20251426스크립트: 78. 파이썬 - 소스 코드의 파일 경로를 지정한 모듈 로드
13952정성태6/15/20251767닷넷: 2336. C# - IValueTaskSource로 인해 주의가 필요한 ValueTask 호출파일 다운로드1
13951정성태6/15/20251631오류 유형: 964. Outlook - 일정이 "You cannot make changes to contents of this read-only folder." 오류 메시지로 삭제가 안 되는 경우
13950정성태6/12/20252129닷넷: 2335. C# - 간단하게 구현해 보는 IValueTaskSource 예제파일 다운로드1
13949정성태6/11/20251833오류 유형: 963. SignTool - "Error: SignerSign() failed." (-2146869243/0x80096005)
13948정성태6/10/20251431오류 유형: 962. 파이썬 - Linux 환경 + TCP 서버 소켓을 사용하는 프로세스 종료 후 재실행하는 경우 "OSError: [Errno 98] Address already in use" 오류 발생
13947정성태6/9/20251910개발 환경 구성: 750. 파이썬 - Azure App Service에 응용 프로그램 배포 후의 환경
13946정성태6/9/20252019개발 환경 구성: 749. 파이썬 - Azure App Service에 응용 프로그램 배포하기 전의 환경
13945정성태6/7/20251729오류 유형: 961. 파이썬 + conda - mysqlclient 사용 시 "NameError: name '_mysql' is not defined" 에러
13944정성태6/7/20255475오류 유형: 960. The trust relationship between this workstation and the primary domain failed. - 네 번째 이야기
13943정성태6/6/20252145개발 환경 구성: 748. Windows + Foundry Local - 로컬에서 AI 모델 활용 [1]
13942정성태6/5/20251694오류 유형: 959. winget 설치 시 "0x80d02002 : unknown error"
13941정성태6/2/20251533닷넷: 2334. C# - cpuid 명령어를 이용한 CPU 제조사 문자열 가져오기파일 다운로드1
13940정성태6/1/20251974C/C++: 188. C++의 32비트 + Release 어셈블리 코드를 .NET으로 포팅할 때 주의할 점파일 다운로드1
13939정성태5/29/20252502오류 유형: 958. NVIDIA Triton Inference Server - version `GLIBCXX_3.4.32' not found (required by /opt/tritonserver/backends/python/triton_python_backend_stub)
13938정성태5/29/20251933개발 환경 구성: 747. 파이썬 - WSL/docker에 구성한 Triton 예제 개발 환경
13937정성태5/24/20251885개발 환경 구성: 746. Windows + WSL2 환경에서 (tensorflow 등의) NVIDIA GPU 인식
[1]  2  3  4  5  6  7  8  9  10  11  12  13  14  15  ...