AMD Radeon PRO V620 vs NVIDIA Tesla V100 PCIe 16 GB Comparison
AMD Radeon PRO V620
Tesla V100 PCIe 16 GB
PERFORMANCE BENCHMARKS
Analysis: AMD Radeon PRO V620 vs NVIDIA Tesla V100 PCIe 16 GB
The NVIDIA Tesla V100 PCIe 16 GB and AMD Radeon PRO V620 are both end-of-life workstation accelerators with no display outputs, but they sit at opposite ends of the architectural spectrum. The data shows a clean split: NVIDIA wins the OpenCL workload, AMD wins the Vulkan workload, and the average benchmark scores land within 1.2% of each other. This is a rare case where the choice is dictated almost entirely by the software API in use rather than by raw performance dominance.
Head-to-Head Benchmarks
The two accelerators each take one of the two recorded benchmark tests. In Geekbench OpenCL, the NVIDIA Tesla V100 PCIe 16 GB scores 163063, while the AMD Radeon PRO V620 scores 128580. That is a 26.8% advantage for the NVIDIA card, and it is the single largest margin in this comparison. The V100’s OpenCL result is also higher than its own average benchmark score of 138063, indicating that this test plays to the Volta architecture’s strengths. By contrast, the V620’s OpenCL score is below its average of 136472, which suggests the RDNA 2.0 design is less efficient in this particular workload.
The situation reverses completely in Geekbench Vulkan. Here, the AMD Radeon PRO V620 scores 144364, beating the NVIDIA Tesla V100 PCIe 16 GB’s 113062 by 21.7%. This is a substantial swing, and it means that the V620’s Vulkan result is well above its average, while the V100’s Vulkan result is well below its own average. The V100’s Vulkan score is roughly 30% lower than its OpenCL score, whereas the V620’s Vulkan score is about 12% higher than its OpenCL score. This pattern suggests that the newer RDNA 2.0 architecture handles Vulkan’s explicit, low-level compute model far more naturally than the older Volta design.
Looking at the aggregate numbers, the NVIDIA card has an average benchmark score of 138063, and the AMD card has an average benchmark score of 136472. The V100 leads by 1.2% in this combined measure, according to the nearestRivals data. That is a razor-thin margin, and it is smaller than the delta between the V100 and its own closest rival, the AMD Instinct MI100, which sits 0.7% ahead of the V100. In practical terms, the two cards are statistically equivalent on average, but the per-test results are anything but equivalent. The winner is determined entirely by whether your application uses OpenCL or Vulkan.
The Verdict
Choose the NVIDIA Tesla V100 PCIe 16 GB if your workload is built on OpenCL. The data shows a 26.8% lead over the AMD Radeon PRO V620 in that specific benchmark, and that is a decisive margin. The V100 also has a higher average benchmark score overall, at 138063 versus 136472, and it ranks at the 96th percentile among all GPUs, same as the V620. For legacy compute stacks that still rely on OpenCL, the V100 is the safer bet, and its 16 GB of HBM2 memory with 897.0 GB/s of bandwidth is clearly sufficient for the tasks that appear in this benchmark data.
Choose the AMD Radeon PRO V620 if your application is Vulkan-based. The V620 beats the V100 by 21.7% in Geekbench Vulkan, which is a larger margin than the V100’s OpenCL win in percentage terms? No, it is slightly smaller — 21.7% versus 26.8%. Still, the V620’s Vulkan score of 144364 is its stronger result, and it lands in the same 96th percentile as the V100. The V620 also offers double the memory capacity at 32 GB, though it uses GDDR6 with a narrower 256-bit bus and 512.0 GB/s of bandwidth. If your software has moved to Vulkan, the V620 is the clear choice, and the extra memory is a bonus that the benchmark data does not directly evaluate but is present in the specification sheet.
There is no universal winner here. The average scores are within 1.2% of each other, and the head-to-head results split one win each. The decision should be made on API compatibility, not on raw score hunting.
Where Each One Wins
The NVIDIA Tesla V100 PCIe 16 GB wins in OpenCL compute. Its 163063 score is 26.8% ahead of the V620’s 128580, and this is the only test where the V100’s score exceeds its own average. This suggests that the Volta architecture, with its 640 tensor cores and 5120 shading units, is well suited to the general-purpose compute patterns that OpenCL benchmarks typically exercise. The V100 also wins on average benchmark score, at 138063 versus 136472, but that is a 1.2% margin that is arguably within noise.
The AMD Radeon PRO V620 wins in Vulkan compute. Its 144364 score is 21.7% ahead of the V100’s 113062, and this is the V620’s best result by a wide margin. The RDNA 2.0 architecture, with 72 ray accelerators and 4608 shading units, appears to handle Vulkan’s explicit command queues and async compute features much more efficiently. The V620 also wins on raw throughput metrics: it has a higher fp32 rate at 20.28 TFLOPS versus 14.13 TFLOPS, a higher fp16 rate at 40.55 TFLOPS versus 28.26 TFLOPS, and a higher pixel rate at 281.6 GPixel/s versus 176.6 GPixel/s. These numbers are not directly benchmarked, but they align with the Vulkan result.
For memory, the V620 wins on capacity with 32 GB versus 16 GB, but the V100 wins on bandwidth with 897.0 GB/s versus 512.0 GB/s. The V100 also has a wider 4096-bit bus versus 256-bit, which explains its bandwidth advantage despite using HBM2 instead of GDDR6.
FAQ
Q: Which card has the higher average benchmark score?
A: The NVIDIA Tesla V100 PCIe 16 GB has an average benchmark score of 138063, which is 1.2% higher than the AMD Radeon PRO V620’s 136472, according to the nearestRivals data.
Q: How large is the OpenCL performance gap?
A: The V100 scores 163063 in Geekbench OpenCL, while the V620 scores 128580. That gives the V100 a 26.8% advantage.
Q: How large is the Vulkan performance gap?
A: The V620 scores 144364 in Geekbench Vulkan, while the V100 scores 113062. That gives the V620 a 21.7% advantage.
Q: Do both cards support the same API versions?
A: Both support Vulkan 1.4 and OpenGL 4.6. However, the V100 supports DirectX 12 (12_1), while the V620 supports DirectX 12 Ultimate (12_2).
Q: What is the memory configuration difference?
A: The V100 has 16 GB of HBM2 on a 4096-bit bus with 897.0 GB/s bandwidth. The V620 has 32 GB of GDDR6 on a 256-bit bus with 512.0 GB/s bandwidth.
Q: Which card has a higher transistor density?
A: The AMD Radeon PRO V620 has a transistor density of 51.5M per mm², compared to the NVIDIA Tesla V100’s 25.9M per mm².
Architecture Differences
The NVIDIA Tesla V100 PCIe 16 GB is built on the Volta architecture using the GV100 chip, fabricated on a 12 nm process at TSMC. It houses 21,100 million transistors on an 815 mm² die, giving a transistor density of 25.9M per mm². The V100 includes 640 tensor cores, which are absent from the AMD card, and it has no ray accelerators. Its memory subsystem is HBM2 with a 4096-bit bus, and it supports PCIe 3.0 x16.
The AMD Radeon PRO V620 is built on the RDNA 2.0 architecture using the Navi 21 chip, fabricated on a 7 nm process at TSMC. It houses 26,800 million transistors on a 520 mm² die, giving a transistor density of 51.5M per mm². The V620 includes 72 ray accelerators, which are absent from the NVIDIA card, and it has no tensor cores. Its memory subsystem is GDDR6 with a 256-bit bus, and it supports PCIe 4.0 x16.
The process node difference is significant: 7 nm versus 12 nm allows the V620 to pack more transistors into a smaller die, and this is reflected in the shading unit count. The V100 has 5120 shading units, 320 TMUs, and 128 ROPs. The V620 has 4608 shading units, 288 TMUs, and 128 ROPs. The V620 has fewer shading units but a higher boost clock of 2200 MHz versus 1380 MHz, which contributes to its higher fp32 throughput.
Specification Differences
The two cards differ on several specification fields. The NVIDIA Tesla V100 PCIe 16 GB has a base clock of 1245 MHz and a boost clock of 1380 MHz, while the AMD Radeon PRO V620 has a base clock of 1825 MHz and a boost clock of 2200 MHz. Memory clocks are 876 MHz (1752 Mbps effective) for the V100 and 2000 MHz (16 Gbps effective) for the V620.
The V100 has 16 GB of HBM2 memory, while the V620 has 32 GB of GDDR6. The V100’s bus width is 4096 bit, versus 256 bit for the V620. Memory bandwidth is 897.0 GB/s for the V100 and 512.0 GB/s for the V620.
The V100 has 5120 shading units, 320 TMUs, and 128 ROPs. The V620 has 4608 shading units, 288 TMUs, and 128 ROPs. The V100 has 640 tensor cores and no ray accelerators; the V620 has 72 ray accelerators and no tensor cores.
Pixel rate is 176.6 GPixel/s for the V100 and 281.6 GPixel/s for the V620. Texture rate is 441.6 GTexel/s for the V100 and 633.6 GTexel/s for the V620. FP32 compute is 14.13 TFLOPS for the V100 and 20.28 TFLOPS for the V620. FP16 compute is 28.26 TFLOPS for the V100 and 40.55 TFLOPS for the V620, both at a 2:1 ratio.
The V100 uses PCIe 3.0 x16, while the V620 uses PCIe 4.0 x16. The V620 has physical dimensions of 267 mm in length, 120 mm in height, and 50 mm in width, while the V100’s dimensions are not listed. Both cards are dual-slot, use 2x 8-pin power connectors, have a 300 W TDP, and recommend a 700 W PSU. The V100 was released on 2017-06-20, while the V620 was released on 2021-11-03. The V100’s predecessor is Tesla Pascal and its successor is Tesla Turing; the V620’s predecessor is Radeon Pro Vega.