AMD Radeon PRO V620 vs NVIDIA Tesla V100 PCIe 32 GB Comparison
AMD Radeon PRO V620
Tesla V100 PCIe 32 GB
PERFORMANCE BENCHMARKS
Analysis: AMD Radeon PRO V620 vs NVIDIA Tesla V100 PCIe 32 GB
Head-to-Head Benchmarks
The benchmark data presents a split decision between the NVIDIA Tesla V100 PCIe 32 GB and the AMD Radeon PRO V620. In the Geekbench OpenCL test, the NVIDIA card delivers a score of 168,763, while the AMD card trails significantly at 128,580. That is a 31.3% advantage for the Tesla V100, a commanding lead in raw compute throughput as measured by this particular workload. The gap is substantial enough that the V100 sits comfortably ahead, a result consistent with its position among its own nearest rivals.
However, the picture reverses when looking at the Geekbench Vulkan test. Here, the AMD Radeon PRO V620 scores 144,364, beating the NVIDIA Tesla V100's 131,847 by 8.7%. This is a notable turnaround. The Vulkan result suggests the RDNA 2.0 architecture has a distinct efficiency advantage in this API, likely due to lower-level hardware access and better utilization of its shader array. The delta of 8.7% is not trivial, but it is far smaller than the 31.3% gap seen in OpenCL.
The overall average benchmark score tells a similar story of divergence. The Tesla V100 averages 150,305 across the two tests, while the Radeon PRO V620 averages 136,472. The NVIDIA card's OpenCL dominance pulls its average higher, despite losing the Vulkan test. The data shows each card wins one head-to-head matchup, with the magnitude of the NVIDIA win being roughly 3.6 times larger than the AMD win in percentage terms.
To contextualize these scores, the Tesla V100's average places it just 1.1% behind the NVIDIA A10G, which scores 151,963. It is also 6.5% behind the AMD Radeon Pro W6800X (160,671) and 7.5% behind the NVIDIA A100 PCIe 40 GB (162,504). Meanwhile, it is 8.1% ahead of the AMD Instinct MI100 (139,035). The Radeon PRO V620, with its 136,472 average, sits in a tighter cluster: it is a mere 0.5% behind the AMD Radeon Pro W6800X Duo (135,774), 0.8% behind the AMD Radeon PRO W6800 (135,396), and 0.9% behind both the NVIDIA A10M (135,230) and the NVIDIA RTX 4000 Ada Generation (135,218). Both cards achieve a 96th percentile ranking against all GPUs, meaning they are in the top tier of available hardware, but their performance profiles are clearly oriented toward different tasks.
Architecture Differences
The fundamental architectural split between these two cards explains the benchmark results. The NVIDIA Tesla V100 is built on the Volta architecture using the GV100 chip, manufactured on a 12 nm process at TSMC. It packs 21,100 million transistors onto a die size of 815 mm², yielding a transistor density of 25.9 million per square millimeter. In contrast, the AMD Radeon PRO V620 uses the RDNA 2.0 architecture with the Navi 21 chip, fabricated on a more advanced 7 nm process, also at TSMC. This chip contains 26,800 million transistors on a smaller 520 mm² die, achieving a much higher density of 51.5 million per square millimeter.
The compute resources differ markedly. The Tesla V100 fields 5,120 shading units, 320 texture mapping units, and 128 ROPs. It also includes 640 tensor cores, which are specialized for AI and deep learning workloads, but it has no dedicated ray tracing cores. The Radeon PRO V620 counters with 4,608 shading units, 288 TMUs, and 128 ROPs. It includes 72 ray accelerators (RT cores) but has no tensor core equivalent. This is a clear philosophical divergence: NVIDIA's card is tailored for matrix math and neural network inference, while AMD's card brings hardware-accelerated ray tracing to the table.
Memory architecture is another major differentiator. The Tesla V100 uses 32 GB of HBM2 memory on a 4096-bit bus, delivering 897.0 GB/s of bandwidth. The memory clock runs at 876 MHz, with an effective rate of 1752 Mbps. The Radeon PRO V620 also has 32 GB, but it uses GDDR6 memory on a 256-bit bus, providing 512.0 GB/s of bandwidth. The GDDR6 runs at 2000 MHz with an effective rate of 16 Gbps. The NVIDIA card has nearly 75% more memory bandwidth, which is critical for data-heavy compute tasks, while the AMD card's narrower bus is compensated by higher clock speeds but still falls short on total throughput.
Clock speeds favor AMD. The Radeon PRO V620 has a base clock of 1825 MHz and a boost clock of 2200 MHz, compared to the Tesla V100's 1230 MHz base and 1380 MHz boost. This clock advantage helps explain the AMD card's higher pixel and texture rates: it achieves 281.6 GPixel/s and 633.6 GTexel/s, versus 176.6 GPixel/s and 441.6 GTexel/s for the NVIDIA card. However, the Tesla V100's FP32 throughput of 14.13 TFLOPS and FP16 throughput of 28.26 TFLOPS (2:1) are lower than the AMD card's 20.28 TFLOPS FP32 and 40.55 TFLOPS FP16 (2:1). The Radeon PRO V620 is clearly the faster card on paper for raw floating-point math, yet the OpenCL benchmark tells a different story, suggesting that the V100's tensor cores or memory bandwidth play a decisive role in that specific workload.
Where Each One Wins
The data points to a clear use-case split. The NVIDIA Tesla V100 PCIe 32 GB wins decisively in OpenCL, a general-purpose compute API widely used in scientific simulation, image processing, and machine learning frameworks. Its 31.3% lead in this test is not marginal; it is a dominant performance gap. The card's 640 tensor cores, while not directly benchmarked here, are a known accelerator for AI workloads and likely contribute to its strong showing in compute-heavy tasks. The massive 897.0 GB/s memory bandwidth also favors data-intensive applications where moving large datasets quickly is paramount. For users running OpenCL-based workloads, the V100 is the clear choice based on benchmark results.
The AMD Radeon PRO V620 wins in Vulkan, an API that is more prevalent in gaming, real-time rendering, and certain professional visualization tasks. Its 8.7% lead in this test indicates better optimization for draw calls and graphics pipelines. The presence of 72 ray tracing cores gives it a hardware advantage in ray-traced rendering scenarios, a feature entirely absent on the Tesla V100. The higher clock speeds (2200 MHz boost) and higher FP32 throughput (20.28 TFLOPS) also suggest better performance in tasks that are latency-sensitive or rely on shader execution rather than memory bandwidth. For Vulkan-based rendering or ray tracing workloads, the Radeon PRO V620 is the stronger performer.
It is importantly the AMD card's wins are narrower in percentage terms than the NVIDIA card's. The 8.7% Vulkan advantage is real but not overwhelming, while the 31.3% OpenCL advantage is a landslide. This asymmetry matters for users who prioritize compute over graphics.
Specification Differences
The table below highlights only the fields where the two cards differ, based on the FACT PACK data.
| Specification | NVIDIA Tesla V100 PCIe 32 GB | AMD Radeon PRO V620 |
|---|---|---|
| Architecture | Volta | RDNA 2.0 |
| Chip | GV100 | Navi 21 |
| 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 | 1230 MHz | 1825 MHz |
| Boost Clock | 1380 MHz | 2200 MHz |
| Memory Clock | 876 MHz (1752 Mbps effective) | 2000 MHz (16 Gbps effective) |
| Memory Type | HBM2 | GDDR6 |
| Memory Bus Width | 4096 bit | 256 bit |
| Memory Bandwidth | 897.0 GB/s | 512.0 GB/s |
| Shading Units | 5120 | 4608 |
| TMUs | 320 | 288 |
| RT Cores | None | 72 |
| Tensor Cores | 640 | None |
| Pixel Rate | 176.6 GPixel/s | 281.6 GPixel/s |
| Texture Rate | 441.6 GTexel/s | 633.6 GTexel/s |
| FP32 | 14.13 TFLOPS | 20.28 TFLOPS |
| FP16 | 28.26 TFLOPS (2:1) | 40.55 TFLOPS (2:1) |
| TDP | 250 W | 300 W |
| Suggested PSU | 600 W | 700 W |
| Bus Interface | PCIe 3.0 x16 | PCIe 4.0 x16 |
| DirectX | 12 (12_1) | 12 Ultimate (12_2) |
| Release Date | 2018-03-26 | 2021-11-03 |
| Predecessor | Tesla Pascal | Radeon Pro Vega |
| Successor | Tesla Turing | None |
Both cards are dual-slot, use 2x 8-pin power connectors, have no display outputs, and share the same OpenGL (4.6) and Vulkan (1.4) API support. The AMD card has specified dimensions of 267 mm in length, 120 mm in height, and 50 mm in width; the NVIDIA card's dimensions are not listed in the data.
FAQ
Q: Which card has a higher average benchmark score?
A: The NVIDIA Tesla V100 PCIe 32 GB has a higher average benchmark score of 150,305, compared to the AMD Radeon PRO V620's 136,472.
Q: How large is the performance gap in the OpenCL test?
A: The NVIDIA Tesla V100 leads by 31.3% in Geekbench OpenCL, scoring 168,763 versus the AMD card's 128,580.
Q: Does the AMD card win any benchmark?
A: Yes, the AMD Radeon PRO V620 wins the Geekbench Vulkan test with a score of 144,364, which is 8.7% higher than the NVIDIA card's 131,847.
Q: What is the memory bandwidth difference?
A: The NVIDIA Tesla V100 offers 897.0 GB/s of memory bandwidth using HBM2 on a 4096-bit bus, while the AMD Radeon PRO V620 provides 512.0 GB/s using GDDR6 on a 256-bit bus.
Q: Which card has hardware ray tracing support?
A: The AMD Radeon PRO V620 includes 72 ray tracing cores, whereas the NVIDIA Tesla V100 has no RT cores.
Q: How do the cards compare on transistor density?
A: The AMD Radeon PRO V620 has a transistor density of 51.5M / mm², significantly higher than the NVIDIA Tesla V100's 25.9M / mm², due to the 7 nm process versus 12 nm.
The Verdict
Based strictly on the benchmark data, the choice between these two cards depends entirely on the target workload. The NVIDIA Tesla V100 PCIe 32 GB is the superior option for OpenCL-based compute tasks. Its 31.3% lead in that benchmark is decisive, and its higher average score (150,305 versus 136,472) reflects overall strength in general-purpose compute. The 640 tensor cores and substantially higher memory bandwidth (897.0 GB/s) make it well-suited for AI inference and large-scale data processing, even if those specific capabilities are not directly benchmarked here. For users running scientific simulations, machine learning training, or any OpenCL-heavy application, the V100 is the clear pick from the data.
The AMD Radeon PRO V620 is the better choice for Vulkan-based rendering workloads. Its 8.7% win in that test, combined with 72 ray tracing cores and higher raw FP32 performance (20.28 TFLOPS), positions it as a stronger candidate for ray-traced graphics and real-time visualization. The higher boost clock of 2200 MHz also suggests better responsiveness in latency-sensitive tasks. For professionals working in game development, architectural visualization, or any Vulkan-centric pipeline, the Radeon PRO V620 offers a measurable advantage.
The data does not support a universal winner. The Tesla V100's win is larger in magnitude, but the Radeon PRO V620's win is in a distinctly different domain. Both cards sit at the 96th percentile of all GPUs, indicating top-tier performance within their respective niches. If forced to choose based on the average benchmark score alone, the NVIDIA card wins, but that would ignore the Vulkan-specific strengths of the AMD card. The decision should be guided by the API and workload profile of the user's applications, not by a single aggregate number.