GPUs

Intel Gaudi 3 vs AMD Instinct MI250X Full Specs

16,384 Shaders
438MHz
14,080 Shaders
1.7GHz
128GB HBM2e3.7TB/s
128GB HBM2e3.28TB/s
··
-
··
47.87 TFLOPS
Form Factor
OAM Module
Form Factor
OAM Module
TDP
900W
TDP
560W

Gaudi 3Gaudi 31.84 PFLOPSFP8-32 Tensor
x4.80
Instinct MI250XInstinct MI250X383 TOPSINT4 Tensor
x1

Clock Speed
··
Clock Speed
···
Peak OPS
1.84 PFLOPSFP8-32 Tensor
Peak OPS
383 TOPSINT4 Tensor
Tensor FP8-32
1.84 PFLOPS
-
Tensor FP16-16
-
Tensor FP16-32
459.3 TFLOPS
Tensor FP16-16
383 TFLOPS
Tensor FP16-32
383 TFLOPS
Tensor BF16
1.84 PFLOPS
Tensor BF16
383 TFLOPS
Tensor TF32
459.3 TFLOPS
Tensor TF32
-
FP32
-
Tensor FP32
229.6 TFLOPS
FP32
47.87 TFLOPS
Tensor FP32
95.74 TFLOPS
-
FP64
47.87 TFLOPS
Tensor FP64
95.74 TFLOPS
Tensor INT4
-
Tensor INT4
383 TOPS
-
Tensor INT8
383 TOPS
Texture Rate
-
Texture Rate
1.5 TTexel/s

Shaders
16,384 Shaders
Shaders
14,080 Shaders
TMUs
-
TMUs
880 TMUs
Tensor Cores
64 T-Cores
Tensor Cores
880 T-Cores
CUs
8 CUs
CUs
220 CUs

Base Clock
-
Base Clock
1GHz
Boost Clock
438MHz
Boost Clock
1.7GHz

L2 Cache
-
L2 Cache
16.4MB shared

128GB HBM2e
128GB HBM2e
Memory Bus
8192-bit
Memory Bus
8192-bit
Memory Speed
3.6GT/s
Memory Speed
3.2GT/s
Memory Bandwidth
3.7TB/s
Memory Bandwidth
3.28TB/s
ECC
Yes
ECC
Yes
ESRAM
96MBSize
12800GB/sBandwidth
ESRAM
-

TDP
900W
TDP
560W

-
Encoder Model
2x VCN 2.6

Decoder Model
14x Gaudi
Decoder Model
2x VCN 2.6

Form Factor
OAM Module
Form Factor
OAM Module
Cooling
Passive
Cooling
Passive

Manufacturer
Manufacturer
Chip Designer
Chip Designer
Architecture
Architecture
Family
Family
Branding
Gaudi branding
Branding
Instinct branding
Codename
Gaudi 3
Codename
Aldebaran
Chip Variant
-
Chip Variant
Aldebaran XTX
Market Segment
Server
Market Segment
Server
Release Date
Apr 9, 2024
Release Date
Nov 8, 2021

Foundry
TSMC
Foundry
TSMC
Fabrication Node
N5
Fabrication Node
N6
Die Size
-
Die Size
1448mm²
Transistor Count
-
Transistor Count
56 Billion
Transistor Density
-
Transistor Density
38.67 MTr/mm²

No images available
No images available