NVIDIA GeForce RTX 5070 vs NVIDIA GeForce RTX 4070 Ti SUPER

差异规格基准褒奖

差异

需要考虑的原因
NVIDIA GeForce RTX 5070

报告一个错误

需要考虑的原因
NVIDIA GeForce RTX 4070 Ti SUPER

报告一个错误

更高的时钟速度

左右 0% 更好的时钟速度

2.34 GHz 2.33 GHz
更多内存

33更多内存百分比

16 GB 12 GB

规格

技术数据

NVIDIA GeForce RTX 5070

NVIDIA GeForce RTX 4070 Ti SUPER

一般信息

显卡的基本技术数据比较 NVIDIA GeForce RTX 5070 和 NVIDIA GeForce RTX 4070 Ti SUPER, 芯片，信息处理单元.

建筑学

Blackwell Ada Lovelace
基于

NVIDIA GeForce RTX 5070 NVIDIA GeForce RTX 4070 Ti SUPER
CUDA Cores

6144 8448
着色器

6144 8448

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

672 GB/s -
Memory Bus

192 bit 256 bit
内存大小

12 GB 16 GB
存储器类型

GDDR7 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

2330 MHz 2340 MHz
最大频率

2510 MHz 2610 MHz

能源消耗

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

Maximum GPU Temperature

85 °C 90 °C
System Power Supply

650 W (System power requirement) 700 W (System power requirement)
TDP

250 W 285 W

接口

连通性和连接.

PCIe Interface

5.0 4.0

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 / Encode Decode / Encode

基准

性能测试 GPUs

Shared tests with results for both models: 13

3DMark graphics tests

Each row is a separate graphics subtest. Scores are not game FPS.

Test and settings	NVIDIA GeForce RTX 5070	NVIDIA GeForce RTX 4070 Ti SUPER	Result difference¹
3DMark – Steel Nomad — Steel Nomad Score / DirectX 12	4838 points Base GPU test: Nvidia GeForce RTX 5070 VRAM: 12 GBOS: Windows 11Graphics API: DirectX 12HVCI: enabledApplication version unreported	5581 points Base GPU test: Nvidia GeForce RTX 4070 Ti Super VRAM: 16 GBOS: Windows 11Graphics API: DirectX 12HVCI: enabledApplication version unreported	+15.4 %
3DMark – Time Spy — Graphics Score / DirectX 12	22267 points Base GPU test: Nvidia GeForce RTX 5070 VRAM: 12 GBTest CPU: AMD Ryzen 9 7950X3DOS: Windows 11Graphics API: DirectX 12HVCI: enabledApplication version unreported	24404 points Base GPU test: Nvidia GeForce RTX 4070 Ti Super VRAM: 16 GBTest CPU: AMD Ryzen 9 7950X3DOS: Windows 11Graphics API: DirectX 12HVCI: enabledApplication version unreported	+9.6 %

Blender Cycles

Each row uses one Blender version, operating system and compute backend.

Test and settings	NVIDIA GeForce RTX 5070	NVIDIA GeForce RTX 4070 Ti SUPER	Result difference¹
Blender 5.2.0 / OPTIX / Windows	5767.65523 samples/min Submissions: 157	6961.227734 samples/min Submissions: 42	+20.7 %
Blender 5.1.1 / OPTIX / Linux	6377.213977 samples/min Submissions: 36	7417.43604 samples/min Submissions: 7	+16.3 %
Blender 5.1.1 / OPTIX / Windows	5932.112448 samples/min Submissions: 226	7058.756179 samples/min Submissions: 57	+19.0 %
Blender 5.1.0 / OPTIX / Windows	5879.888246 samples/min Submissions: 85	7146.252511 samples/min Submissions: 14	+21.5 %
Blender 5.0.0 / OPTIX / Windows	5621.534179 samples/min Submissions: 31	6540.235565 samples/min Submissions: 9	+16.3 %
Blender 4.5.0 / OPTIX / Linux	6583.233133 samples/min Submissions: 71	7401.239808 samples/min Submissions: 30	+12.4 %
Blender 4.5.0 / OPTIX / Windows	6156.449706 samples/min Submissions: 843	7086.688923 samples/min Submissions: 223	+15.1 %
Blender 4.4.0 / OPTIX / Linux	6599.280337 samples/min Submissions: 31	7462.023189 samples/min Submissions: 16	+13.1 %
Blender 4.4.0 / OPTIX / Windows	6161.53544 samples/min Submissions: 405	7102.956728 samples/min Submissions: 224	+15.3 %
Blender 4.3.0 / OPTIX / Windows	6082.558299 samples/min Submissions: 35	6996.671596 samples/min Submissions: 494	+15.0 %
Blender 4.2.0 / OPTIX / Windows	6064.893658 samples/min Submissions: 9	6996.166673 samples/min Submissions: 358	+15.4 %

¹ Difference of the second result relative to the first in that row. No percentage is calculated for different upscaling algorithms. This compares published results, not a universal performance index.

A base GPU test is a reference for cards using that GPU and the same memory capacity. Board overclocking, cooling and power limits can change the result.

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.