GeekRadar / The comparison desk

COMPARE.

GPU

NVIDIA

NVIDIA CMP 70HX

Graphics processor

Video memory: 8 GB^*
Memory bus: 256 bit^*
Rated memory bandwidth: 608.30 GB/s^*

Benchmark results

Geekbench GPU (reported compilation) · version unreported · OpenCL42,211^*Platform unreported Compare results

Processing & platform

Architecture: Ampere^*
Process node: 8 nm^*
GPU cores: 3,840^*
Graphics interface: PCIe 1.0 x4^*
Reported chip die area: 392 mm²^*
Transistors: 17,400 million^*
GPU L2 cache: 4 MB^*
Graphics processor: GA104^*
Texture mapping units: 120^*
Raster output units: 64^*
Streaming multiprocessors: 30^*
Tensor cores: 120^*
Ray tracing cores: 30^*
Chip foundry: Samsung^*
Card or module format: Dual-slot^*
Suggested system PSU: 200 W^*
External power connectors: 1x 12-pin^*
DirectX / feature level: 12 Ultimate (12_2)^*
OpenGL: 4.6^*
OpenCL: 3.0^*
Vulkan: 1.4^*
CUDA compute capability: 8.6^*
Shader model: 6.9^*

Memory & storage

Memory options: 8^*
Video memory: 8 GB^*
Video memory type: GDDR6X^*
Memory bus: 256 bit^*
Rated memory bandwidth: 608.30 GB/s^*

System configuration

Detailed specifications

Specifications 1 · 30 fields

proc transistors: 17,400 million^*
proc die size: 392^* mm²
cores: 3,840^*
tensor cores: 120^*
rt cores: 30^*
sm count: 30^*
base clock: 1,365^*
boost clock: 1,395^*
memory clock: 1,188^*
ram: 8^*
name: NVIDIA CMP 70HX^*
ram type: GDDR6X^*
bus width: 256^*
bus interface: PCIe 1.0 x4^*
ram bandwidth: 608.3^* GB/s
fp16 performance: 10.71^*
fp32 performance: 10.71^*
fp64 performance: 0.1674^*
suggested psu: 200^* W
slot width: Dual-slot^*
power connectors: 1x 12-pin^*
api cuda: 8.6^*
is retail board: No^*
gpi value: 7.67^*
vendor: nvd^*
manufacturer: NVIDIA^*
arch name: Ampere^*
card release date: 2021-03-11^*
proc foundry: Samsung^*
proc process size: 8^* nm

Specifications 2 · 39 fields

CUDA compute capability

8.6^*

DirectX / feature level

12 Ultimate (12_2)^*

OpenCL

3.0^*

OpenGL

4.6^*

Shader model

6.9^*

Vulkan

1.4^*

Reported theoretical BF16 peak

21.43^* TFLOPS

Scope and conditions

Arithmetic density: Not stated^*

Graphics interface

PCIe 1.0 x4^*

Memory bus

256^* bit

Reported shader processors

3,840^*

Reported theoretical FP16 peak

10.71^* TFLOPS

Scope and conditions

Arithmetic density: Not stated^*

Reported theoretical FP32 peak

10.71^* TFLOPS

Scope and conditions

Arithmetic density: Not stated^*

Reported theoretical FP64 peak

0.1674^* TFLOPS

Scope and conditions

Arithmetic density: Not stated^*

Declared L1 cache configuration

128 KB (per SM)^*

Declared GPU L2 cache

4 MB^*

Reported display outputs

No outputs^*

Theoretical pixel fill rate

89.28 GPixel/s^*

External power connectors

1x 12-pin^*

Reported chip die area

392 mm²^*

Chip foundry

Samsung^*

Process node declaration

8 nm^*

Transistors

17.4 billion^*

Graphics processor

GA104^*

Reported video memory

8^* GB

Rated memory bandwidth

608.3 GB/s^*

Video memory type

GDDR6X^*

Raster output units

64^*

Ray tracing cores

30^*

Reported shading units

3,840^*

Card or module format

Dual-slot^*

Streaming multiprocessors

30^*

Suggested system PSU

200 W^*

Tensor cores

120^*

Theoretical texture fill rate

167.4 GTexel/s^*

Reported theoretical TF32 peak

10.71^* TFLOPS

Scope and conditions

Arithmetic density: Not stated^*

Texture mapping units

120^*

GPU cores

3,840^*

Video memory

8^* GB

Video memory type

GDDR6X^*

Specifications 3 · 2 fields

Device

NVIDIA CMP 70HX^*

Scope and conditions

benchmark_context

suite: Geekbench GPU (reported compilation)^*
backend_columns: CUDA^*
Metal^*
OpenCL^*
Vulkan^*
test_configuration_verified: No^*
comparable: No^*

OpenCL

42,211^*

Scope and conditions

benchmark_context

suite: Geekbench GPU (reported compilation)^*
backend_columns: CUDA^*
Metal^*
OpenCL^*
Vulkan^*
test_configuration_verified: No^*
comparable: No^*

* Provisional data