NVIDIA A40 vs Intel Gaudi

NVIDIA A40
Intel Gaudi
10752 Shaders
48GB GDDR6
1740MHz
2048 Shaders
32GB HBM2
610MHz
Peak AI Performance
1.2 POPS
INT4 Tensor Sparse
Peak AI Performance
159.91 TFLOPS
BF16 Tensor
FP32
37.42 TFLOPS
-
-
FP16
37.42 TFLOPS
-
-
Form Factor
PCIe Card
2.0-Slots
Form Factor
OAM Module
-
TDP
300W
TDP
350W
Power Connectors
-
-
-
-
-
-
-
-
-
GB6 OpenCL N/A
0%
GB6 Metal N/A
0%
GB6 Metal N/A
0%
GB6 Vulkan N/A
0%
GB5 OpenCL N/A
0%
GB5 OpenCL N/A
0%
GB5 CUDA N/A
0%
GB5 CUDA N/A
0%
GB5 Metal N/A
0%
GB5 Metal N/A
0%
GB5 Vulkan N/A
0%
GB5 Vulkan N/A
0%
OCT 2020.1 N/A
0%
OCT 2020.1 N/A
0%
OCT Metal N/A
0%
OCT Metal N/A
0%
Peak AI Performance
1.2 POPS
INT4 Tensor Sparse
Peak AI Performance
159.91 TFLOPS
BF16 Tensor
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
FP16
37.42 TFLOPS
149.67 TFLOPS Tensor (FP16 Accumulate)
299.34 TFLOPS Tensor (FP16 Accumulate) Sparse
149.67 TFLOPS Tensor (FP32 Accumulate)
299.34 TFLOPS Tensor (FP32 Accumulate) Sparse
-
-
-
-
-
-
FP32
37.42 TFLOPS
-
-
-
-
-
-
FP64
590 GFLOPS
-
-
-
-
BF16
37.42 TFLOPS
149.67 TFLOPS Tensor
299.34 TFLOPS Tensor Sparse
BF16
-
159.91 TFLOPS Tensor
-
TF32
74.83 TFLOPS Tensor
149.67 TFLOPS Tensor Sparse
-
-
-
INT4
598.67 TOPS Tensor
1.2 POPS Tensor Sparse
-
-
-
INT8
-
299.34 TOPS Tensor
598.67 TOPS Tensor Sparse
-
-
-
-
INT32
18.71 TOPS
-
-
Ray Tracing
73.1 TOPS
-
-
Pixel Fillrate
194.88 GPixel/s
Pixel Fillrate
-
-
-
-
-
Texture Fillrate
584.64 GTexel/s
Texture Fillrate
-
Manufacturer
NVIDIA
Manufacturer
Intel
Chip Designer
NVIDIA
Chip Designer
Intel
Architecture
Ampere
Architecture
Gaudi
Family
Server
Family
Gaudi
Codename
NV172
GA102
Variant
GA102-895-A1
Codename
Gaudi
HL-2000
Variant
HL-2000
Market Segment
Server
Market Segment
Server
Release Date
10/5/2020
Release Date
6/17/2019
Foundry
Samsung
-
Foundry
TSMC
-
Fabrication Node
8N
-
Fabrication Node
16FF
-
Die Size
628 mm²
-
-
-
-
Transistor Count
28.3 Billion
-
-
-
Transistor Density
45.04M/mm²
-
-
-
-
Form
PCIe Card
Form
OAM Module
Shading Units
10752 Shaders
-
Shading Units
2048 Shaders
-
Texture Mapping Units
336 TMUs
Texture Mapping Units
-
Render Output Units
112 ROPs
Render Output Units
-
Tensor Cores
336 T-Cores
Tensor Cores
8 T-Cores
Ray-Tracing Cores
84 RT-Cores
-
-
Streaming Multiprocessors
84 SMs
-
-
-
-
Compute Units
1 CU
-
-
-
-
-
-
-
-
-
-
1305MHz Base
1740MHz
-
-
-
610MHz
-
-
-
-
L1
64KB/SM Tex
128KB/SM
-
-
L1
-
-
-
Unknown
L2
6MB Shared
L2
Unknown
-
-
-
-
-
-
48GB
GDDR6
-
32GB
HBM2
ECC
Bus Width
384Bit
Bus Width
4096Bit
Clock
906MHz
Transfer Rate
7.2GT/s
Bandwidth
347.9GB/s
Clock
980MHz
Transfer Rate
2GT/s
Bandwidth
1003.5GB/s
-
-
-
-
-
-
-
-
-
-
-
-
-
eSRAM
24MB
3200GB/s
-
-
TDP
300W
TDP
350W
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
3x DisplayPort 1.3
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
No Ports
Max Resolution
7680x4320
Max Resolution
Unknown
Max Resolution Refresh Rate
60Hz
Max Resolution Refresh Rate
-
Variable Refresh Rate
G-Sync
FreeSync
-
Variable Refresh Rate
-
-
-
Display Stream Compression (DSC)
Supported
Display Stream Compression (DSC)
Not Supported
Multi Monitor Support
3
Multi Monitor Support
Unknown
Content Protection
HDCP 2.3
-
-
Model
NVENC 7
No Encoders
-
Codec
-
-
-
-
-
-
-
-
AVC (H.264)
HEVC (H.265)
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Model
NVDEC 5
No Decoders
Codec
MPEG-1
MPEG-2
MPEG-4
-
VC-1
VP8
VP9
-
AVC (H.264)
HEVC (H.265)
-
AV1
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Direct X
12
Direct 3D
12_2
-
-
-
-
OpenGL
4.6
OpenCL
3.0
Vulkan
1.2
-
-
OpenCL
3.0
-
-
Shader Model
6.6
CUDA
8.6
-
-
PureVideo HD
VP11
VDPAU
Feature Set K
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Not a Card
-
-
-
Power Connectors
-
-
-
-
1x 8-Pin EPS
-
-
-
-
-
-
-
-
-
-
Slots Required
2.0
PCIe Version
4.0
PCIe Lanes
16
-
-
PCIe Version
4.0
PCIe Lanes
16
Multi GPU Support
Supported
Type
NVLink
Multi GPU Support
Supported
Type
RoCE
Height
112 mm (4.41 in)
Width
267 mm (10.51 in)
Depth
40 mm (1.57 in)
-
-
-
-
-
-
Change Comparison