AMD Radeon PRO V620 vs NVIDIA Tesla V100 SXM2 32 GB Comparison

AMD
RADEON

AMD Radeon PRO V620

CORE STATE Navi 21
VRAM 32 GB
CLOCK SPEED 2200 MHz
TDP 300 W
BUS WIDTH 256 bit
ARCHITECTURE RDNA 2.0
nm
PROCESS 7 nm
LAUNCH DATE 2021
VS
NVIDIA
GEFORCE

Tesla V100 SXM2 32 GB

CORE STATE GV100
VRAM 32 GB
CLOCK SPEED 1530 MHz
TDP 250 W
BUS WIDTH 4096 bit
ARCHITECTURE Volta
nm
PROCESS 12 nm
LAUNCH DATE 2018

PERFORMANCE BENCHMARKS

geekbench_opencl
128,580
131,696
geekbench_vulkan
144,364
143,765

Analysis: AMD Radeon PRO V620 vs NVIDIA Tesla V100 SXM2 32 GB

# NVIDIA Tesla V100 SXM2 32 GB vs AMD Radeon PRO V620

The NVIDIA Tesla V100 SXM2 32 GB and AMD Radeon PRO V620 are both end-of-life professional accelerators aimed at compute-heavy workloads, yet they represent fundamentally different design philosophies and eras of GPU architecture. The data shows a near-tie in overall average benchmark scores—the Tesla V100 posts an average of 137,731 against the Radeon PRO V620's 136,472, a margin of just 0.9%—but their performance profiles diverge sharply depending on the workload, and their underlying architectures could hardly be more different.

Where Each One Wins

The head-to-head benchmark results reveal a clean split: the NVIDIA Tesla V100 SXM2 32 GB wins the Geekbench OpenCL test with a score of 131,696, while the AMD Radeon PRO V620 takes the Geekbench Vulkan test with a score of 144,364. That is one win apiece, but the magnitude of the margins tells a more nuanced story. In OpenCL, the Tesla V100 leads by 2.4%, a modest but consistent advantage that reflects its mature compute stack and HBM2 memory subsystem. In Vulkan, the Radeon PRO V620 counters with a 0.4% edge—a much narrower margin, suggesting that the Vulkan API is a more level playing field for these two cards.

For compute workloads that rely heavily on OpenCL—common in scientific simulation, machine learning inference, and some rendering pipelines—the Tesla V100 is the safer bet. Its 2.4% lead in OpenCL, combined with its higher memory bandwidth (898.0 GB/s vs 512.0 GB/s), makes it particularly strong for memory-bandwidth-bound tasks. Conversely, the Radeon PRO V620's Vulkan win, while small, points to its advantage in workloads that leverage modern graphics APIs, ray tracing (it has 72 dedicated RT cores), and higher raw FP32 throughput (20.28 TFLOPS vs 15.67 TFLOPS). If your application is Vulkan-centric or benefits from ray tracing acceleration, the AMD card is the one that wins.

Architecture Differences

The architectural gulf between these two GPUs is wide. The Tesla V100 uses NVIDIA's Volta architecture on a 12 nm TSMC process, packing 21,100 million transistors into an 815 mm² die. Its transistor density is 25.9M per mm². The Radeon PRO V620, by contrast, uses AMD's RDNA 2.0 architecture on a 7 nm TSMC process, with 26,800 million transistors on a much smaller 520 mm² die—a transistor density of 51.5M per mm², nearly double that of the Volta chip. This density difference is a direct reflection of the three-and-a-half-year gap between their release dates (March 2018 for the Tesla, November 2021 for the Radeon).

Memory architecture is another major differentiator. The Tesla V100 uses 32 GB of HBM2 on a 4096-bit bus, delivering 898.0 GB/s of bandwidth. The Radeon PRO V620 also has 32 GB, but it uses GDDR6 on a 256-bit bus, yielding 512.0 GB/s—about 43% less bandwidth. The Tesla's massive memory bus is its key advantage for large datasets that need to stay resident in VRAM. The Radeon compensates with higher clock speeds: its base clock is 1825 MHz and boost is 2200 MHz, versus the Tesla's 1290 MHz base and 1530 MHz boost. The Radeon also has more raw compute throughput (20.28 FP32 TFLOPS vs 15.67 FP32 TFLOPS) and higher pixel/texture rates (281.6 GPixel/s and 633.6 GTexel/s vs 195.8 GPixel/s and 489.6 GTexel/s).

Feature sets diverge as well. The Tesla V100 has 640 tensor cores, purpose-built for AI and deep learning workloads; the Radeon PRO V620 has no tensor cores but instead includes 72 ray tracing cores for real-time ray tracing. The Radeon supports DirectX 12 Ultimate (12_2), while the Tesla is limited to DirectX 12 (12_1). Both support OpenGL 4.6 and Vulkan 1.4. Neither card has display outputs, so both are strictly compute or rendering accelerators.

FAQ

Q: Which card has higher memory bandwidth?

A: The NVIDIA Tesla V100 SXM2 32 GB, with 898.0 GB/s from its HBM2 memory on a 4096-bit bus. The AMD Radeon PRO V620 offers 512.0 GB/s via GDDR6 on a 256-bit bus.

Q: Does the Radeon PRO V620 support ray tracing?

A: Yes, it has 72 ray tracing cores built into its RDNA 2.0 architecture. The Tesla V100 has no ray tracing cores.

Q: Which card is better for AI/deep learning workloads?

A: The Tesla V100 has 640 tensor cores, which are specifically designed for AI and deep learning operations. The Radeon PRO V620 has no tensor cores, so the Tesla is the clear choice for those workloads.

Q: What are the power requirements for each card?

A: The Tesla V100 SXM2 32 GB has a 250 W TDP and requires a 600 W suggested PSU; it uses an SXM module form factor with no power connectors. The Radeon PRO V620 has a 300 W TDP, requires a 700 W suggested PSU, and uses two 8-pin power connectors in a dual-slot design.

Q: How do their average benchmark scores compare?

A: The Tesla V100 has an average benchmark score of 137,731, while the Radeon PRO V620 scores 136,472. That puts the Tesla 0.9% ahead of the Radeon. Both sit at the 96th percentile among all GPUs.

Q: Which card has a smaller physical footprint?

A: The Radeon PRO V620 is a dual-slot card measuring 267 mm in length, 120 mm in height, and 50 mm in width. The Tesla V100 is an SXM module, which has no listed length/height/width dimensions in the data.

Specification Differences

| Specification | NVIDIA Tesla V100 SXM2 32 GB | AMD Radeon PRO V620 |

|---|---|---|

| Architecture | Volta | RDNA 2.0 |

| Process Node | 12 nm | 7 nm |

| Transistors | 21,100 million | 26,800 million |

| Die Size | 815 mm² | 520 mm² |

| Transistor Density | 25.9M / mm² | 51.5M / mm² |

| Base Clock | 1290 MHz | 1825 MHz |

| Boost Clock | 1530 MHz | 2200 MHz |

| Memory Type | HBM2 | GDDR6 |

| Memory Bus Width | 4096 bit | 256 bit |

| Memory Bandwidth | 898.0 GB/s | 512.0 GB/s |

| Memory Clock | 877 MHz (1754 Mbps effective) | 2000 MHz (16 Gbps effective) |

| Shading Units | 5120 | 4608 |

| TMUs | 320 | 288 |

| ROPs | 128 | 128 |

| Tensor Cores | 640 | None |

| RT Cores | None | 72 |

| FP32 Performance | 15.67 TFLOPS | 20.28 TFLOPS |

| FP16 Performance | 31.33 TFLOPS (2:1) | 40.55 TFLOPS (2:1) |

| Pixel Rate | 195.8 GPixel/s | 281.6 GPixel/s |

| Texture Rate | 489.6 GTexel/s | 633.6 GTexel/s |

| TDP | 250 W | 300 W |

| Suggested PSU | 600 W | 700 W |

| Slot Width | SXM Module | Dual-slot |

| Power Connectors | None | 2x 8-pin |

| Bus Interface | PCIe 3.0 x16 | PCIe 4.0 x16 |

| DirectX Support | 12 (12_1) | 12 Ultimate (12_2) |

| Release Date | 2018-03-26 | 2021-11-03 |

Head-to-Head Benchmarks

The two available benchmark comparisons show a split decision. In Geekbench OpenCL, the NVIDIA Tesla V100 SXM2 32 GB scores 131,696 against the AMD Radeon PRO V620's 128,580—a 2.4% win for NVIDIA. That margin is meaningful in a compute context: it suggests that for OpenCL-based workloads, the Tesla's higher memory bandwidth (898.0 GB/s vs 512.0 GB/s) more than compensates for its lower raw FP32 throughput (15.67 TFLOPS vs 20.28 TFLOPS). The Tesla's 4096-bit memory bus is a massive asset when the workload is memory-bound.

In Geekbench Vulkan, the tables turn. The Radeon PRO V620 scores 144,364 versus the Tesla's 143,765, giving AMD a 0.4% edge. The margin is small, but it is a win nonetheless. The Radeon's advantage here likely stems from its higher clock speeds (2200 MHz boost vs 1530 MHz) and its modern RDNA 2.0 architecture, which includes 72 ray tracing cores and DirectX 12 Ultimate support. Vulkan tends to favor architectures with higher shader throughput and more modern feature sets, which plays to the Radeon's strengths.

Looking at the broader rival context, both cards slot into a tight competitive cluster. The Tesla V100's nearest rival is the AMD Instinct MI100, which scores 139,035—that is 0.9% higher than the Tesla's average. The Radeon PRO V620, meanwhile, sits just 0.5% above the AMD Radeon Pro W6800X Duo (135,774) and 0.8% above the AMD Radeon PRO W6800 (135,396). Neither card dominates its peer group; they are both mid-pack performers in a very dense field.

The Verdict

The data points to a simple conclusion: choose based on workload, not on overall scores. The NVIDIA Tesla V100 SXM2 32 GB is the pick if your primary workloads are OpenCL-based, memory-bandwidth-hungry, or involve AI/deep learning where its 640 tensor cores are essential. Its 2.4% OpenCL lead over the Radeon, combined with 898.0 GB/s of bandwidth, makes it the more capable compute engine for large datasets. It also runs cooler and requires less power—250 W TDP versus 300 W—and its SXM module form factor is designed for dense server integration.

The AMD Radeon PRO V620 is the better choice for Vulkan-centric workloads, ray tracing, or any application that benefits from higher raw FP32 throughput (20.28 TFLOPS) and modern graphics API support (DirectX 12 Ultimate). Its 0.4% Vulkan win over the Tesla shows it can hold its own, and its 72 ray tracing cores open up rendering paths that the Tesla simply cannot handle. It is also built on a much more efficient 7 nm process with nearly double the transistor density (51.5M/mm² vs 25.9M/mm²), which explains its higher clock speeds despite a higher TDP.

For a system builder deciding between these two, the question is not which card is "better"—it's which workload matters more. OpenCL and AI? Take the Tesla. Vulkan, ray tracing, and raw shader throughput? Take the Radeon. Both are end-of-life products, but they still deliver competitive performance in their respective niches, sitting at the 96th percentile among all GPUs. The final average benchmark scores are within 0.9% of each other, but that statistical tie hides a functional split that should drive your purchasing decision.

DETAILED SPECIFICATIONS

SPECIFICATION
PRO V620
Tesla V100 SXM2 32 GB
Core Specs
Shading Units
4,608
5,120 +11.1%
Shaders
4,608
5,120 +11.1%
TMUs
288
320 +11.1%
ROPs
128
128 0.0%
Compute Units
72
—
SM Count
—
80
Clocks
Base Clock
1825 MHz
1290 MHz
Boost Clock
2200 MHz
1530 MHz
Memory Clock
2000 MHz 16 Gbps effective
877 MHz 1754 Mbps effective
Memory
Memory Size
32 GB
32 GB
VRAM (MB)
32,768
32,768 0.0%
Memory Type
GDDR6
HBM2
Memory Bus
256 bit
4096 bit
Bandwidth
512.0 GB/s
898.0 GB/s
Cache
L1 Cache
128 KB per Array
128 KB (per SM)
L2 Cache
4 MB
6 MB
L3 Cache
128 MB
—
L0 Cache
32 KB per WGP
—
Performance
Pixel Rate
281.6 GPixel/s
195.8 GPixel/s
Texture Rate
633.6 GTexel/s
489.6 GTexel/s
FP32 (TFLOPS)
20.28 TFLOPS
15.67 TFLOPS
FP64 (TFLOPS)
1,267.2 GFLOPS (1:16)
7.834 TFLOPS (1:2)
FP16 (TFLOPS)
40.55 TFLOPS (2:1)
31.33 TFLOPS (2:1)
AI/RT
RT Cores
72
—
Tensor Cores
—
640
Power
TDP
300 W
250 W
TDP (W)
300
250 -16.7%
Suggested PSU
700 W
600 W
Power Connectors
2x 8-pin
None
Architecture
Architecture
RDNA 2.0
Volta
GPU Name
Navi 21
GV100
Generation
Radeon Pro Navi (Navi II Series)
Tesla Volta (Vxx)
Process Size
7 nm
12 nm
Transistors
26,800 million
21,100 million
Die Size
520 mm²
815 mm²
Foundry
TSMC
TSMC
Density
51.5M / mm²
25.9M / mm²
API Support
DirectX
12 Ultimate (12_2)
12 (12_1)
OpenGL
4.6
4.6
Vulkan
1.4
1.4
OpenCL
2.1
3.0
CUDA
—
7.0
Shader Model
6.8
6.8
Physical
Slot Width
Dual-slot
SXM Module
Length
267 mm 10.5 inches
—
Height
120 mm 4.7 inches
—
Outputs
No outputs
No outputs
Bus Interface
PCIe 4.0 x16
PCIe 3.0 x16
Other
Production
End-of-life
End-of-life
Predecessor
Radeon Pro Vega
Tesla Pascal
Successor
—
Tesla Turing
View Radeon PRO V620 Details View Tesla V100 SXM2 32 GB Details