AMD Radeon PRO V620 vs NVIDIA A100 SXM4 80 GB Comparison
AMD Radeon PRO V620
A100 SXM4 80 GB
PERFORMANCE BENCHMARKS
Analysis: AMD Radeon PRO V620 vs NVIDIA A100 SXM4 80 GB
The NVIDIA A100 SXM4 80 GB and the AMD Radeon PRO V620 are both end-of-life server/workstation accelerators, but they target fundamentally different compute profiles. The data shows the A100 is a clear leader in the single available cross-platform benchmark, but the V620 counters with a distinct architectural feature set and a much lower power envelope. For buyers prioritizing raw compute throughput in a Vulkan workload, the A100 is the data-backed choice; for those needing a more compact, lower-power card with hardware ray tracing and broader API support, the V620 is the only option of the two.
The Verdict
The benchmark data is unambiguous in one respect: the NVIDIA A100 SXM4 80 GB wins the only head-to-head test available. In Geekbench Vulkan, the A100 scores 183,725 against the V620’s 144,364, a 27.3% margin. This places the A100 in the 98th percentile of all GPUs, while the V620 sits at the 96th percentile. The A100’s nearest rivals include the NVIDIA RTX 5000 Ada Generation (deltaPct -0.5%) and the NVIDIA A100 SXM4 40 GB (deltaPct -1.8%), showing it is statistically tied with the very top of the stack. The V620, meanwhile, is clustered with the AMD Radeon Pro W6800X Duo (deltaPct 0.5%) and the NVIDIA A10M (deltaPct 0.9%), indicating it trades blows with mid-range professional cards rather than competing with flagship accelerators.
The verdict splits by workload type. For general compute, AI inference, or any task that scales with FP32 and FP16 throughput, the A100 is the stronger pick: it delivers 19.49 TFLOPS FP32 and 77.97 TFLOPS FP16 (4:1), versus 20.28 TFLOPS FP32 and 40.55 TFLOPS FP16 (2:1) for the V620. The A100’s FP16 advantage is nearly 2x, a decisive factor for machine learning. However, the V620 is the only card with hardware ray tracing cores (72 RT cores), and it is the only one with a modern API stack (DirectX 12 Ultimate, OpenGL 4.6, Vulkan 1.4). If your software stack requires those APIs or ray-traced workloads, the V620 is the data-backed choice, despite losing the Vulkan benchmark. The A100 has no listed DirectX, OpenGL, or Vulkan support, making it a pure compute device.
Architecture Differences
The two GPUs come from different architectural philosophies. The A100 uses NVIDIA’s Ampere architecture on the GA100 chip, built on a 7 nm TSMC process. It packs 54,200 million transistors on an 826 mm² die, yielding a transistor density of 65.6M per mm². The V620 uses AMD’s RDNA 2.0 architecture on the Navi 21 chip, also on a 7 nm TSMC process, but with only 26,800 million transistors on a 520 mm² die (51.5M per mm²). The A100’s die is 58% larger and holds just over twice the transistors, which explains its higher compute ceiling.
The compute pipelines are fundamentally different. The A100 has 6,912 shading units, 432 TMUs, and 160 ROPs, plus 432 tensor cores for AI workloads. The V620 has 4,608 shading units, 288 TMUs, and 128 ROPs, but adds 72 ray tracing cores—a feature the A100 lacks entirely. The A100 has no tensor core counterpart on the AMD side, but the V620’s RT cores give it a hardware feature that the Ampere part cannot match. The A100’s FP16 throughput is listed as 4:1, meaning it uses specialized hardware to quadruple FP16 rates; the V620’s FP16 is 2:1, a more modest uplift.
Memory architecture also diverges sharply. The A100 uses 80 GB of HBM2e on a 5120-bit bus, delivering 2.04 TB/s of bandwidth. The V620 uses 32 GB of GDDR6 on a 256-bit bus, providing 512.0 GB/s. The A100 has 4x the bandwidth and 2.5x the capacity, which is critical for large datasets that exceed the V620’s 32 GB limit. The A100’s memory clock is 1593 MHz (3.2 Gbps effective), while the V620 runs at 2000 MHz (16 Gbps effective); the GDDR6 runs faster per pin, but the HBM2e’s massive bus width wins overall.
Head-to-Head Benchmarks
The only direct comparison in the data is Geekbench Vulkan, and the A100 wins decisively. The A100 scores 183,725 versus the V620’s 144,364, a delta of 27.3%. This is not a marginal victory; it is a category-level gap. The A100’s score positions it above the NVIDIA GeForce RTX 4090 D (178,050, deltaPct 3.2%) and just below the NVIDIA A100 SXM4 40 GB (187,147, deltaPct -1.8%). The V620’s score, by contrast, falls below all four of the A100’s nearest rivals, and its own nearest rivals are all within 0.9% of its score, indicating it is in a dense competitive pack.
The performance delta stems from the A100’s raw resources: more shading units, more TMUs, higher memory bandwidth, and tensor cores that can accelerate certain Vulkan compute paths. The V620’s higher boost clock (2200 MHz versus 1410 MHz) and higher pixel rate (281.6 GPixel/s versus 225.6 GPixel/s) do not compensate for its narrower memory bus and fewer compute units in this test. The V620’s texture rate is 633.6 GTexel/s, slightly ahead of the A100’s 609.1 GTexel/s, but that single metric does not translate into a benchmark win. The data shows a 27.3% lead for NVIDIA, and no other benchmark is listed to dispute it.
FAQ
Q: Which card is faster in the Geekbench Vulkan test?
A: The NVIDIA A100 SXM4 80 GB scores 183,725, which is 27.3% higher than the AMD Radeon PRO V620’s 144,364.
Q: Does the AMD card have any hardware feature the NVIDIA card lacks?
A: Yes. The AMD Radeon PRO V620 includes 72 ray tracing cores, while the NVIDIA A100 SXM4 80 GB has no RT cores listed.
Q: Which card has more memory bandwidth?
A: The NVIDIA A100 SXM4 80 GB has 2.04 TB/s bandwidth from 80 GB of HBM2e, versus 512.0 GB/s from 32 GB of GDDR6 on the AMD card.
Q: What is the power consumption difference?
A: The AMD Radeon PRO V620 has a 300 W TDP, while the NVIDIA A100 SXM4 80 GB has a 400 W TDP. The AMD card also requires a lower suggested PSU (700 W versus 800 W).
Q: Which card is better for FP16 compute workloads?
A: The NVIDIA A100 SXM4 80 GB delivers 77.97 TFLOPS FP16 (4:1), nearly double the AMD card’s 40.55 TFLOPS FP16 (2:1).
Q: Does the AMD card support modern graphics APIs?
A: Yes. The AMD Radeon PRO V620 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The NVIDIA A100 SXM4 80 GB has no API support listed.
Where Each One Wins
The NVIDIA A100 SXM4 80 GB wins on raw compute and memory capacity. Its 27.3% Vulkan benchmark lead is backed by a hardware spec sheet that dominates in memory bandwidth (2.04 TB/s versus 512.0 GB/s) and FP16 throughput (77.97 TFLOPS versus 40.55 TFLOPS). The 80 GB of HBM2e is the standout feature; any workload that fits in 32 GB will run on the V620, but anything larger requires the A100. The tensor cores (432) give the A100 a hardware path for AI and deep learning tasks that the V620 cannot accelerate. The A100 also has a higher pixel rate? No, that is incorrect—the V620 has 281.6 GPixel/s versus 225.6 GPixel/s. The A100 wins on nearly every compute metric except pixel rate and texture rate, where the V620 leads slightly (633.6 GTexel/s versus 609.1 GTexel/s).
The AMD Radeon PRO V620 wins on power efficiency, physical size, and feature completeness. It draws 300 W versus the A100’s 400 W, and it fits in a dual-slot form factor with 2x 8-pin power connectors, versus the A100’s OAM module (which has no power connectors listed and is not a standard PCIe card). The V620 is 267 mm long, 120 mm high, and 50 mm wide, making it installable in a standard workstation chassis. The A100’s OAM form factor requires a specialized server board. The V620 also has 72 RT cores and full DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4 support—the A100 has no API support listed, meaning it cannot run graphics workloads that rely on those standards. For ray tracing or any DirectX/Vulkan graphics task, the V620 is the only functional choice.
Specification Differences
The two cards differ in nearly every major specification category. The chip and architecture are different: NVIDIA uses GA100 (Ampere) while AMD uses Navi 21 (RDNA 2.0). The A100 has 54,200 million transistors on an 826 mm² die, versus 26,800 million on 520 mm² for the V620. Clock speeds favor AMD: the V620 has a base clock of 1825 MHz and boost of 2200 MHz, while the A100 runs at 1275 MHz base and 1410 MHz boost. Memory is the biggest split: 80 GB HBM2e on a 5120-bit bus with 2.04 TB/s bandwidth versus 32 GB GDDR6 on a 256-bit bus with 512.0 GB/s. The A100 has 6,912 shading units, 432 TMUs, and 160 ROPs, while the V620 has 4,608 shading units, 288 TMUs, and 128 ROPs. The A100 adds 432 tensor cores; the V620 adds 72 RT cores.
Compute throughput differs by precision: the A100 offers 19.49 TFLOPS FP32 and 77.97 TFLOPS FP16 (4:1), while the V620 offers 20.28 TFLOPS FP32 and 40.55 TFLOPS FP16 (2:1). The V620 has a slightly higher FP32 figure, but the A100 dominates FP16. Pixel and texture rates: the V620 leads in pixel rate (281.6 GPixel/s versus 225.6 GPixel/s) and texture rate (633.6 GTexel/s versus 609.1 GTexel/s). Power and physical specs: the A100 has a 400 W TDP with an OAM slot width and no power connectors, while the V620 has a 300 W TDP, dual-slot width, and 2x 8-pin connectors. The suggested PSU is 800 W for the A100 and 700 W for the V620. The V620 has dimensions (267 mm x 120 mm x 50 mm) while the A100 lists none. The A100 has no display outputs; the V620 also has no display outputs. The A100 lacks any API support entries; the V620 lists DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. Production status is end-of-life for both, with release dates of 2020-11-15 for the A100 and 2021-11-03 for the V620.