AMD Radeon PRO V620 vs NVIDIA Tesla V100 SXM2 16 GB Comparison

# The Verdict

The AMD Radeon PRO V620 is the clear overall winner from the data, taking both head-to-head benchmark wins and posting a 19.3% higher average benchmark score (136,472 vs 114,395). The V620 sits in the 96th percentile of all GPUs, while the Tesla V100 SXM2 16 GB sits at the 95th — a narrow percentile gap that conceals a substantial performance delta in raw compute workloads.

For compute-heavy OpenCL tasks, the V620 is decisively ahead, delivering a 47% higher score (128,580 vs 87,454). For Vulkan workloads, the gap narrows dramatically to just 2.1% (144,364 vs 141,336), suggesting the V100 remains competitive in graphics-oriented APIs despite its older architecture. The V620 should be the pick for users prioritizing raw FP32 throughput, modern API support, and double the memory capacity. The V100 retains relevance only for those specifically requiring HBM2 memory bandwidth (897.0 GB/s vs 512.0 GB/s) or the 640 tensor cores that the V620 lacks entirely.

# Architecture Differences

The two GPUs come from fundamentally different architectural generations. The AMD Radeon PRO V620 uses the Navi 21 chip based on RDNA 2.0, fabricated on TSMC's 7 nm process with 26,800 million transistors packed into a 520 mm² die — a transistor density of 51.5M per mm². The NVIDIA Tesla V100 SXM2 16 GB uses the GV100 chip based on the older Volta architecture, built on TSMC's 12 nm process with 21,100 million transistors on a much larger 815 mm² die, yielding a lower density of 25.9M per mm².

The memory subsystems are radically different. The V620 uses 32 GB of GDDR6 across a 256-bit bus, delivering 512.0 GB/s of bandwidth. The V100 uses 16 GB of HBM2 across a 4096-bit bus, delivering 897.0 GB/s — 75% more bandwidth despite half the capacity. The V620 compensates with significantly higher clocks: 1825 MHz base and 2200 MHz boost versus the V100's 1312 MHz base and 1530 MHz boost.

Feature sets diverge sharply. The V620 includes 72 ray tracing cores and supports DirectX 12 Ultimate (12_2), while the V100 has no RT cores and only reaches DirectX 12 (12_1). Conversely, the V100 includes 640 tensor cores — a feature completely absent from the V620. Both support OpenGL 4.6 and Vulkan 1.4. The V620 uses a PCIe 4.0 x16 interface, while the V100 is limited to PCIe 3.0 x16. The V620 is a dual-slot card requiring 2x 8-pin power connectors and a 700 W PSU; the V100 is an SXM module with no power connectors and a 600 W PSU suggestion.

# Where Each One Wins

The AMD Radeon PRO V620 dominates in raw compute throughput. Its FP32 performance of 20.28 TFLOPS exceeds the V100's 15.67 TFLOPS by 29.4%. Its FP16 output of 40.55 TFLOPS (2:1) also surpasses the V100's 31.33 TFLOPS (2:1) by 29.4%. The V620 also wins on pixel rate (281.6 GPixel/s vs 195.8 GPixel/s) and texture rate (633.6 GTexel/s vs 489.6 GTexel/s). Its shading unit count of 4608 is lower than the V100's 5120, yet the V620's higher clocks more than compensate.

The V100's advantages are narrower but real. Its 897.0 GB/s memory bandwidth is a major asset for memory-bound workloads, and its 640 tensor cores provide dedicated hardware for deep learning inference and training that the V620 cannot match. The V100 also has more TMUs (320 vs 288) and ROPs are tied at 128 each. The V100's HBM2 memory operates at 876 MHz with 1752 Mbps effective, while the V620's GDDR6 runs at 2000 MHz with 16 Gbps effective — but the V100's 4096-bit bus width is the decisive factor in bandwidth.

In the benchmark results, the V620 wins both head-to-head tests. The OpenCL gap of 47% is enormous; the Vulkan gap of 2.1% is marginal. This suggests the V620 is the better all-rounder, but the V100 remains viable for Vulkan workloads and any application that leverages tensor cores or requires extreme memory bandwidth.

# FAQ

Q: Which GPU has higher average benchmark performance?

A: The AMD Radeon PRO V620 has an average benchmark score of 136,472, which is 19.3% higher than the NVIDIA Tesla V100 SXM2 16 GB's average of 114,395.

Q: How large is the OpenCL performance gap?

A: The V620 scores 128,580 in geekbench_opencl versus 87,454 for the V100, giving the AMD card a 47% advantage in that workload.

Q: Is the V100 competitive in any benchmark?

A: Yes. In geekbench_vulkan, the V100 scores 141,336 versus the V620's 144,364, a difference of only 2.1%. The V100 also offers 897.0 GB/s of memory bandwidth, which is 75% higher than the V620's 512.0 GB/s.

Q: What are the memory capacity differences?

A: The V620 has 32 GB of GDDR6 memory, while the V100 has 16 GB of HBM2 memory. The V100's bus width is 4096-bit versus the V620's 256-bit.

Q: Which GPU supports ray tracing?

A: Only the AMD Radeon PRO V620 includes 72 ray tracing cores. The NVIDIA Tesla V100 SXM2 16 GB has no ray tracing cores.

Q: What about tensor cores?

A: The NVIDIA V100 includes 640 tensor cores, while the AMD V620 does not list tensor cores as a feature. This makes the V100 the only one of the two with dedicated tensor processing hardware.

# Head-to-Head Benchmarks

The geekbench_opencl test delivers the most decisive result of the comparison. The AMD Radeon PRO V620 scores 128,580 against the V100's 87,454, producing a 47% delta in AMD's favor. This is a massive margin that reflects the V620's architectural advantages: a 7 nm process versus 12 nm, higher clocks (1825 MHz base / 2200 MHz boost vs 1312 MHz / 1530 MHz), and 20.28 TFLOPS of FP32 throughput versus 15.67 TFLOPS. The V620's 32 GB of GDDR6 also doubles the V100's 16 GB capacity, which can matter in large datasets even if the raw bandwidth is lower.

The geekbench_vulkan test tells a different story. The V620 scores 144,364, while the V100 trails closely at 141,336 — a delta of just 2.1%. This narrow margin suggests that the V100's Volta architecture, despite its age, holds up well in Vulkan workloads. The V100's superior memory bandwidth (897.0 GB/s vs 512.0 GB/s) likely helps close the gap, as does its higher shading unit count (5120 vs 4608) and greater TMU count (320 vs 288). The result shows that while the V620 is the better overall GPU, the V100 is not obsolete in every scenario.

The win count is 2 for the AMD Radeon PRO V620 and 0 for the NVIDIA Tesla V100 SXM2 16 GB across the head-to-head tests. However, the average benchmark scores — 136,472 for the V620 and 114,395 for the V100 — indicate that the V100's performance is closer to its rival in aggregate than the OpenCL result alone suggests. The V620's nearest rivals include the AMD Radeon Pro W6800X Duo (0.5% behind), the AMD Radeon PRO W6800 (0.8% behind), the NVIDIA A10M (0.9% behind), and the NVIDIA RTX 4000 Ada Generation (0.9% behind). The V100's nearest rivals are the NVIDIA RTX 4000 SFF Ada Generation (2.3% ahead), the NVIDIA GB10 (2.6% ahead), the NVIDIA RTX A5500 Mobile (0.4% behind), and the AMD Radeon PRO W7900 (3.3% behind).

# Specification Differences

| Specification | AMD Radeon PRO V620 | NVIDIA Tesla V100 SXM2 16 GB |

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

| Chip | Navi 21 | GV100 |

| Architecture | RDNA 2.0 | Volta |

| Process Node | 7 nm | 12 nm |

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

| Die Size | 520 mm² | 815 mm² |

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

| Base Clock | 1825 MHz | 1312 MHz |

| Boost Clock | 2200 MHz | 1530 MHz |

| Memory Clock | 2000 MHz (16 Gbps effective) | 876 MHz (1752 Mbps effective) |

| Memory Size | 32 GB | 16 GB |

| Memory Type | GDDR6 | HBM2 |

| Memory Bus Width | 256 bit | 4096 bit |

| Memory Bandwidth | 512.0 GB/s | 897.0 GB/s |

| Shading Units | 4608 | 5120 |

| TMUs | 288 | 320 |

| ROPs | 128 | 128 |

| RT Cores | 72 | None |

| Tensor Cores | None | 640 |

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

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

| FP32 | 20.28 TFLOPS | 15.67 TFLOPS |

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

| TDP | 300 W | 250 W |

| Slot Width | Dual-slot | SXM Module |

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

| Suggested PSU | 700 W | 600 W |

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

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

| OpenGL | 4.6 | 4.6 |

| Vulkan | 1.4 | 1.4 |

| Length | 267 mm (10.5 inches) | Not specified |

| Release Date | 2021-11-03 | 2017-06-20 |

| Predecessor | Radeon Pro Vega | Tesla Pascal |

| Successor | None | Tesla Turing |