GPU Comparison
RADEON
AMD Radeon Pro VII
CORE STATE
Vega 20
VRAM
16 GB
CLOCK SPEED
1700 MHz
TDP
250 W
BUS WIDTH
4096 bit
ARCHITECTURE
GCN 5.1
PROCESS
7 nm
LAUNCH DATE
2020
VS
GEFORCE
L4
CORE STATE
AD104
VRAM
24 GB
CLOCK SPEED
2040 MHz
TDP
72 W
BUS WIDTH
192 bit
ARCHITECTURE
Ada Lovelace
PROCESS
5 nm
LAUNCH DATE
2023
PERFORMANCE BENCHMARKS
geekbench_opencl
geekbench_vulkan
DETAILED SPECIFICATIONS
SPECIFICATION
Pro VII
L4
Core Specs
Shading Units
3,840
7,424
+93.3%
Shaders
3,840
7,424
+93.3%
TMUs
240
240
0.0%
ROPs
64
80
+25.0%
Compute Units
60
—
SM Count
—
60
Clocks
Base Clock
1400 MHz
795 MHz
Boost Clock
1700 MHz
2040 MHz
Memory Clock
1000 MHz
2 Gbps effective
1563 MHz
12.5 Gbps effective
Memory
Memory Size
16 GB
24 GB
VRAM (MB)
16,384
24,576
+50.0%
Memory Type
HBM2
GDDR6
Memory Bus
4096 bit
192 bit
Bandwidth
1.02 TB/s
300.1 GB/s
Cache
L1 Cache
16 KB (per CU)
128 KB (per SM)
L2 Cache
4 MB
48 MB
Performance
Pixel Rate
108.8 GPixel/s
163.2 GPixel/s
Texture Rate
408.0 GTexel/s
489.6 GTexel/s
FP32 (TFLOPS)
13.06 TFLOPS
30.29 TFLOPS
FP64 (TFLOPS)
6.528 TFLOPS (1:2)
473.3 GFLOPS (1:64)
FP16 (TFLOPS)
26.11 TFLOPS (2:1)
30.29 TFLOPS (1:1)
AI/RT
RT Cores
—
60
Tensor Cores
—
240
Power
TDP
250 W
72 W
TDP (W)
250
72
-71.2%
Suggested PSU
600 W
250 W
Power Connectors
1x 6-pin + 1x 8-pin
None
Architecture
Architecture
GCN 5.1
Ada Lovelace
GPU Name
Vega 20
AD104
Generation
Radeon Pro Vega
(Vega II Series)
Server Ada
(Lxx)
Process Size
7 nm
5 nm
Transistors
13,230 million
35,800 million
Die Size
331 mm²
294 mm²
Foundry
TSMC
TSMC
Density
40.0M / mm²
121.8M / mm²
API Support
DirectX
12 (12_1)
12 Ultimate (12_2)
OpenGL
4.6
4.6
Vulkan
1.3
1.4
OpenCL
2.1
3.0
CUDA
—
8.9
Shader Model
6.7
6.8
Physical
Slot Width
Dual-slot
Single-slot
Length
305 mm
12 inches
169 mm
6.7 inches
Height
111 mm
4.4 inches
56 mm
2.2 inches
Outputs
6x mini-DisplayPort 1.4a
No outputs
Bus Interface
PCIe 4.0 x16
PCIe 4.0 x16
Other
Launch Price
1,899 USD
—
Production
End-of-life
Active
Predecessor
Radeon Pro Polaris
Server Ampere
Successor
Radeon Pro Navi
Server Hopper