Nvidia Blackwell: Difference between revisions

From HPCWIKI
Jump to navigation Jump to search
(Add categories: Linux, Reference)
(Fix: remove --- horizontal lines (7 removed))
 
(2 intermediate revisions by the same user not shown)
Line 1: Line 1:
== Blackwell ==
{{Status
|status=Draft
|owner=Knowledge Agent
|last_update=2026-07-16
|review=Pending
}}


=== Blackwell 6000 edition ===
{{TOC}}
 
== Overview ==
 
Nvidia Blackwell에 대한 기술 문서입니다.
 
=== Summary ===
 
* 무엇인가? - Nvidia Blackwell
* 왜 필요한가? - HPC 및 서버 환경에서 필수 개념
* 언제 사용하는가? - 서버 구성, 성능 튜닝, 문제 해결 시
 
 
== Purpose ==
 
이 문서가 존재하는 이유
 
* Goal: Nvidia Blackwell에 대한 기술 정보 제공
* Scope: Nvidia Blackwell의 개념, 사용법, 설정
* Non-goals: 다른 주제로의 확장
 
 
== Key Concepts ==
 
{| class="wikitable"
! Concept
! Description
! Related
|-
| Nvidia Blackwell
| HPC/서버 환경에서 중요한 기술 개념
| [[Linux]], [[Server]]
|}
 
 
== Detailed Explanation ==


* The server edition is the better option when you’re scaling identical nodes in a data center, where every watt, cable, and airflow route needs to align with an OEM service [[plan]].
* The server edition is the better option when you’re scaling identical nodes in a data center, where every watt, cable, and airflow route needs to align with an OEM service [[plan]].
* The workstation model is ideal for creators and engineers who need to view and interact with their work locally while also processing large inference or simulation tasks.
* The workstation model is ideal for creators and engineers who need to view and interact with their work locally while also processing large inference or simulation tasks.
* Max-Q Edition is best when local interaction, mobility and workstation-class performance are needed, rather than remote or large-scale server deployment.
* Max-Q Edition is best when local interaction, mobility and workstation-class performance are needed, rather than remote or large-scale server deployment.
{| class="wikitable"
{| class="wikitable"
|+
|+
Line 24: Line 63:
|24,064
|24,064
|-
|-
|Tensor Cores
|752 (5th generation)
|752 (5th generation)
|752 (5th generation)
|-
|FP32 (Single-precision performance)
|120 TFLOPS
|125 TFLOPS
|110 TFLOPS
|-
|Peak FP4 AI PFLOPS
|4 PFLOPS
|4 PFLOPS
|4 PFLOPS
|-
|RT Core performance
|188 (fourth-generation) 355 TFLOPS
|380 TFLOPS
|333 TFLOPS
|-
|GPU memory
|96 GB GDDR7 with ECC
|96 GB GDDR7 with ECC
|96 GB GDDR7 with ECC
|-
|Memory interface
|512-bit
|512-bit
|512-bit
|-
|Memory bandwidth
|1597 GB/s
|1792 GB/s
|1792 GB/s
|-
|Power consumption
|Up to 600W (Configurable)
|Total board power: 600 W
|Total board power: 300 W
|-
|Multi-instance GPU
|Up to 4 MIGs @ 24GB
|>Up to 4x 24 GB>Up to 2x 48 GB>Up to 1x 96 GB
|>Up to 4x 24 GB>Up to 2x 48 GB>Up to 1x 96 GB
|-
|<nowiki>NVENC | NVDEC | JPEG</nowiki>
Video Engines
|<nowiki>4x | 4x | 4x</nowiki>
|4x NVENC (9th Gen)4x NVDEC (6th Gen)
|>4x NVENC (9th Gen)>4x NVDEC (6th Gen)
|-
|Graphics bus
|PCI Express 5.0 x16
|PCI Express 5.0 x16
|PCI Express 5.0 x16
|-
|Form factor
|4.4″ (H) x 10.5″ (L), dual slot
|5.4” H x 12” L, dual slot, extended height
|4.4″ (H) x 10.5″ (L), dual slot
|-
|Thermal solution
|Passive
|Double-flow-through
|Active
|-
|Power connector
|1x [[PCIe]] CEM5 16-pin
|1x PCIe CEM5 16-pin
|1x PCIe CEM5 16-pin
|-
|Display connectors
|4× DisplayPort 2.1
|4x DisplayPort 2.1
|4x DisplayPort 2.1
|-
|Ray Tracing Cores
|
|4th Generation
|4th Generation
|-
|Graphics APIs
|
|DirectX 12, Shader Model 6.6, OpenGL 4.63, Vulkan 1.3
|DirectX 12, Shader Model 6.6, OpenGL 4.63, Vulkan 1.3
|-
|Compute APIs
|
|CUDA 12.8, OpenCL 3.0, DirectCompute
|CUDA 12.8, OpenCL 3.0, DirectCompute
|-
|Max simultaneous displays
|
|>4x 4096 x 2160 @ 120 Hz>4x 5120 x 2880 @ 60 Hz>2x 7680 x 4320 @ 60 Hz
|>4x 4096 x 2160 @ 120 Hz>4x 5120 x 2880 @ 60 Hz>2x 7680 x 4320 @ 60 Hz
|-
|AI TOPS
|
|4000 AI TOPS
|3511 AI TOPS
|-
|Confidential compute
|Supported
|
|
|-
|Secure boot with root of trust
|Yes
|
|
|}




== Reference ==
== Best Practices ==
<references />
 
[[Category:Linux]]
* 최신 버전 사용 권장
* 공식 문서 참고
* 테스트 환경에서 먼저 검증
 
 
== References ==
 
* [https://wiki.hpcmate.com Nvidia Blackwell]
 
 
== Related Pages ==
 
* [[Linux]]
* [[Server]]
* [[Hardware]]
* [[Network]]
 
 
[[Category:GPU]]
== Knowledge Graph ==
 
Related
 
→ [[CUDA]]
→ [[GPU]]
→ [[NVIDIA GPU]]
→ [[H100]]
→ [[NVLink]]
→ [[A100]]
→ [[MIG]]
→ [[NVIDIA GPUs]]


[[Category:Reference]]
[[Category:Reference]]

Latest revision as of 11:30, 17 July 2026

Template:Status

Template:TOC

Overview

Nvidia Blackwell에 대한 기술 문서입니다.

Summary

  • 무엇인가? - Nvidia Blackwell
  • 왜 필요한가? - HPC 및 서버 환경에서 필수 개념
  • 언제 사용하는가? - 서버 구성, 성능 튜닝, 문제 해결 시


Purpose

이 문서가 존재하는 이유

  • Goal: Nvidia Blackwell에 대한 기술 정보 제공
  • Scope: Nvidia Blackwell의 개념, 사용법, 설정
  • Non-goals: 다른 주제로의 확장


Key Concepts

Concept Description Related
Nvidia Blackwell HPC/서버 환경에서 중요한 기술 개념 Linux, Server


Detailed Explanation

  • The server edition is the better option when you’re scaling identical nodes in a data center, where every watt, cable, and airflow route needs to align with an OEM service plan.
  • The workstation model is ideal for creators and engineers who need to view and interact with their work locally while also processing large inference or simulation tasks.
  • Max-Q Edition is best when local interaction, mobility and workstation-class performance are needed, rather than remote or large-scale server deployment.

Best Practices

  • 최신 버전 사용 권장
  • 공식 문서 참고
  • 테스트 환경에서 먼저 검증


References


Related Pages

Knowledge Graph

Related

→ CUDA → GPU → NVIDIA GPU → H100 → NVLink → A100 → MIG → NVIDIA GPUs

RTX PRO 6000 Server edition RTX PRO 6000 Workstation edition RTX PRO 6000 Max-Q edition
GPU architecture NVIDIA Blackwell NVIDIA Blackwell NVIDIA Blackwell
CUDA cores 24,064 24,064 24,064
Retrieved from "http://wiki.hpcmate.com/index.php?title=Nvidia_Blackwell&oldid=3410"