NVIDIA H100 PCIe vs NVIDIA GeForce RTX 3090

差异规格基准褒奖

差异

需要考虑的原因
NVIDIA H100 PCIe

报告一个错误

更多内存

233更多内存百分比

80 GB 24 GB

需要考虑的原因
NVIDIA GeForce RTX 3090

报告一个错误

更高的时钟速度

左右 31% 更好的时钟速度

1.4 GHz 1.065 GHz

所有排名中的位置

共同立场 NVIDIA GeForce RTX 3090 在流行的基准测试中的GPU，用于与其他模型进行比较.

Passmark Performance Test GPU

2 place

规格

技术数据

NVIDIA H100 PCIe

NVIDIA GeForce RTX 3090

一般信息

显卡的基本技术数据比较 NVIDIA H100 PCIe 和 NVIDIA GeForce RTX 3090, 芯片，信息处理单元.

建筑学

Hopper Ampere
Execution Units

7296 82
GPU芯片

GH100 GA102-300-A1
渲染输出单位

24 112
纹理单位

456 328
CUDA Cores

- 10496
着色器

- 10496

内存设置

Video memory capacity, type, bus width and bandwidth describe different properties. These values alone do not determine graphics performance.

Memory Bandwidth

1,280 GB/s 936 GB/s
Memory Bus

5120 bit 384 bit
Memory Clock

1000 MHz 1219 MHz
Memory Data Rate

2 Gbps 19.5 Gbps
内存大小

80 GB 24 GB
存储器类型

HBM2e GDDR6X

时钟频率

Published GPU clock frequencies describe different operating modes. Clock speed alone does not determine graphics performance.

基准频率

1065 MHz 1400 MHz
最大频率

1650 MHz 1700 MHz

能源消耗

Graphics card power limits and recommended system power supply describe different specifications.

System Power Supply

750 W (PSU recommendation) 750 W (System power requirement)
TDP

350 W 350 W
Graphics Card Power

- 350 W
Maximum GPU Temperature

- 93 °C

接口

连通性和连接.

最大。显示

1 4
PCIe Interface

PCIe 5.0 x16 PCIe 4.0 x16
Power Connectors

8-pin EPS 2x PCIe 8-pin (adapter to 1x 12-pin included)
DisplayPort Ports

- 3 × DisplayPort
HDCP Version

- 2.3
Other Display Outputs (unparsed)

- HDMI

技术数据

在电脑游戏中被充分使用的技术数据.

CUDA

9.0 -
DirectX

没有 12_2
OpenCL

3.0 3.0
OpenGL

没有 4.6
Shader Model

没有 6.6
Vulkan

没有 1.3
最大分辨率

- 7680 × 4320

支持的视频编解码器

内置对视频和图像压缩标准的支持.

AV1

- Decode
h264

- Decode / Encode
h265 / HEVC

- Decode / Encode
VP8

- Decode
VP9

- Decode

更多信息

GPU的接口和发布日期的比较 NVIDIA GeForce RTX 3090 和 NVIDIA H100 PCIe.

Process Technology

4 nm 8 nm
发布日期

Q2/2022 Q3/2020
Manufacturer specification scope

- NVIDIA Founders Edition/reference design. Clock specifications apply to gaming use. AIB shipping specifications and complete-system power needs may differ.

Theoretical Performance

FP16 (theoretical)

96.31 TFLOPS 35.58 TFLOPS
FP32 (theoretical)

24.08 TFLOPS -
FP64 (theoretical)

12.04 TFLOPS -
Pixel Rate

39.60 GPixel/s 189.8 GPixel/s
Texture Rate

752.4 GTexel/s 556.0 GTexel/s

基准

性能测试 GPUs

Individual model tests

These results have no shared test of both models with the same version and settings. They are shown separately without a percentage difference.

NVIDIA H100 PCIe

NVIDIA GeForce RTX 3090

Blender 4.0.0 / OPTIX / Linux 5746.193538 samples/min
Blender 4.0.0 / OPTIX / Windows 5483.352619 samples/min
Blender 4.1.0 / OPTIX / Linux 5718.835863 samples/min
Blender 4.1.0 / OPTIX / Windows 5491.672029 samples/min
Blender 4.2.0 / OPTIX / Linux 5633.457024 samples/min
Blender 4.2.0 / OPTIX / Windows 5311.379423 samples/min
Blender 4.3.0 / OPTIX / Linux 5639.593959 samples/min
Blender 4.3.0 / OPTIX / Windows 5297.880883 samples/min

More results: 10

Blender 4.4.0 / OPTIX / Linux 5656.590332 samples/min
Blender 4.4.0 / OPTIX / Windows 5277.965639 samples/min
Blender 4.5.0 / OPTIX / Linux 5606.175997 samples/min
Blender 4.5.0 / OPTIX / Windows 5383.496578 samples/min
Blender 5.0.0 / OPTIX / Windows 4868.424378 samples/min
Blender 5.1.0 / OPTIX / Windows 5244.921791 samples/min
Blender 5.1.1 / OPTIX / Linux 5400.816266 samples/min
Blender 5.1.1 / OPTIX / Windows 5177.841757 samples/min
Blender 5.2.0 / OPTIX / Linux 5197.903823 samples/min
Blender 5.2.0 / OPTIX / Windows 5046.518892 samples/min

Individual system tests

Results were obtained on the reported systems. Test versions and settings remain separate; incomplete conditions do not support calculated model performance differences.

Cinebench 2024 / GPU

Model	Result	Reported conditions
NVIDIA GeForce RTX 3090	19512 points	System: GeForce RTX 3090 Graphics: GeForce RTX 3090 Source minimum runtime: 0 minutes. Exact test build and complete conditions are unreported; no ranking position.