AMD Radeon Instinct MI60 vs AMD Radeon PRO V620 Comparison

Head-to-Head Benchmarks

The benchmark data delivers a clear, decisive result across both recorded workloads. In the Geekbench OpenCL test, the AMD Radeon PRO V620 posts a score of 128,580 against the AMD Radeon Instinct MI60’s 92,488. That is a 39% advantage for the V620, a margin substantial enough to be felt in any compute-heavy task that scales well with raw throughput. The Vulkan result is even more lopsided: the V620 scores 144,364 while the MI60 manages 92,444, yielding a 56.2% lead for the newer card. In both cases, the V620 wins outright, giving it a 2–0 record in the head-to-head comparison.

What makes the Vulkan gap particularly noteworthy is that it widens beyond the already-large OpenCL margin. The V620’s Vulkan score is 56.2% higher than the MI60’s, whereas the OpenCL difference sits at 39%. This suggests the V620’s architecture is not just faster in absolute terms but also scales better in APIs that expose modern hardware features. The MI60, by contrast, shows near-identical scores across both APIs (92,488 OpenCL vs. 92,444 Vulkan), indicating that its performance is largely API-agnostic—consistent but stagnant. The V620’s Vulkan advantage of 15,784 points over its own OpenCL score further underscores that it is extracting extra performance from newer driver and API paths, while the MI60 appears to be hitting a hard ceiling regardless of workload type.

Looking at the broader competitive context, the V620’s average benchmark score of 136,472 places it in the 96th percentile of all GPUs. Its nearest rivals—the AMD Radeon Pro W6800X Duo (135,774), the AMD Radeon PRO W6800 (135,396), the NVIDIA A10M (135,230), and the NVIDIA RTX 4000 Ada Generation (135,218)—are all within a 0.9% delta. This means the V620 is effectively at parity with the top tier of workstation and data-center accelerators, sitting at the pinnacle of its class. The MI60, with an average score of 92,466, lands in the 93rd percentile, which is still respectable but a full three percentile points lower. Its nearest rivals include the AMD Radeon Pro VII (97,131, which is 4.8% faster) and the AMD Radeon RX 7900M (97,487, which is 5.2% faster), meaning the MI60 is actually trailing some of its own contemporaries. The NVIDIA RTX A4500 (91,671) and RTX A4500 Mobile (91,134) are slightly behind the MI60 by 0.9% and 1.5%, respectively, but the overall picture is that the MI60 sits in the middle of a pack that the V620 dominates from above.

The Verdict

The data is unambiguous: the AMD Radeon PRO V620 is the superior performer in every measured benchmark. It wins both head-to-head tests, holds a higher percentile ranking (96th vs. 93rd), and posts an average score that is roughly 47% higher than the MI60’s (136,472 vs. 92,466). Anyone choosing between these two purely on compute performance should select the V620 without hesitation. The 56.2% Vulkan lead and 39% OpenCL lead are not marginal differences; they represent a generational leap in capability.

However, the MI60 is not without its own merits, and the choice is not solely about raw speed. The MI60 offers 32 GB of HBM2 memory on a 4096-bit bus, delivering 1.02 TB/s of bandwidth. The V620 also has 32 GB, but it is GDDR6 on a 256-bit bus, yielding 512.0 GB/s. For workloads that are memory-bandwidth-bound rather than compute-bound, the MI60’s 1.02 TB/s could be a decisive factor, as it offers double the bandwidth of the V620. The MI60 also includes a display output (1x mini-DisplayPort 1.4a), whereas the V620 has no outputs at all, making the MI60 the only option of the two for any task requiring visual output. The V620 is strictly a compute accelerator.

The production status of both cards is end-of-life, so neither is a future-proof investment. But if the question is which card to pick today from the data alone, the V620 wins on nearly every compute metric. Choose the MI60 only if memory bandwidth or display output is a hard requirement, and even then, be aware that you are sacrificing significant compute performance to get it. The V620 is the benchmark champion; the MI60 is a niche solution for bandwidth-sensitive or display-equipped scenarios.

Where Each One Wins

The AMD Radeon PRO V620 wins in pure compute throughput across both OpenCL and Vulkan workloads. Its 4608 shading units, 288 texture mapping units, and 128 raster operation pipelines give it a structural advantage over the MI60’s 4096 shaders, 256 TMUs, and 64 ROPs. The V620’s 20.28 TFLOPS of FP32 performance and 40.55 TFLOPS of FP16 performance (2:1 ratio) dwarf the MI60’s 14.75 TFLOPS FP32 and 29.49 TFLOPS FP16. In any application that scales with shader count or raw FLOPs—rendering, simulation, machine learning inference—the V620 is the clear winner. Its 72 ray-tracing cores also give it a hardware feature that the MI60 lacks entirely, making it the better choice for any ray-traced or path-traced workload, even if those tasks are more common in professional visualization than in data-center compute.

The AMD Radeon Instinct MI60 wins in memory bandwidth. Its 1.02 TB/s of HBM2 bandwidth on a 4096-bit bus is exactly double the V620’s 512.0 GB/s on a 256-bit bus. For workloads that are bottlenecked by memory access—large matrix operations, data shuffling, certain database or analytics tasks—the MI60’s bandwidth advantage could offset its lower compute throughput. The MI60 also has a display output (1x mini-DisplayPort 1.4a), which the V620 completely lacks. If a deployment requires a GPU that can also drive a monitor or diagnostic display, the MI60 is the only choice. The MI60’s lower transistor count (13,230 million vs. 26,800 million) and smaller die size (331 mm² vs. 520 mm²) also suggest it is a simpler, potentially more power-efficient part per unit of memory bandwidth, though both cards carry a 300 W TDP.

FAQ

Q: Which card has a higher average benchmark score?

A: The AMD Radeon PRO V620 has an average benchmark score of 136,472, while the AMD Radeon Instinct MI60 scores 92,466.

Q: How much faster is the V620 in Vulkan?

A: The V620 scores 144,364 in Geekbench Vulkan, which is 56.2% higher than the MI60’s 92,444.

Q: Does the MI60 have any advantage in memory bandwidth?

A: Yes, the MI60 offers 1.02 TB/s of bandwidth via HBM2 on a 4096-bit bus, versus the V620’s 512.0 GB/s via GDDR6 on a 256-bit bus.

Q: Can either card output video to a display?

A: The MI60 has 1x mini-DisplayPort 1.4a, while the V620 has no display outputs.

Q: Which card has ray-tracing cores?

A: Only the V620 has ray-tracing cores (72 of them); the MI60 has none.

Q: What are the FP32 performance figures for each?

A: The V620 delivers 20.28 TFLOPS FP32, while the MI60 delivers 14.75 TFLOPS FP32.

Architecture Differences

The two cards come from entirely different architectural lineages. The AMD Radeon PRO V620 is built on the Navi 21 chip using RDNA 2.0 architecture, manufactured on a 7 nm process at TSMC. It packs 26,800 million transistors on a 520 mm² die, giving a transistor density of 51.5M per mm². The AMD Radeon Instinct MI60, by contrast, uses the Vega 20 chip with GCN 5.1 architecture, also on a 7 nm process at TSMC, but with only 13,230 million transistors on a 331 mm² die, for a density of 40.0M per mm². The V620 is the denser, more modern design.

The memory subsystems are fundamentally different. The V620 uses 32 GB of GDDR6 on a 256-bit bus, with a 2000 MHz memory clock (16 Gbps effective) delivering 512.0 GB/s. The MI60 uses 32 GB of HBM2 on a 4096-bit bus, with a 1000 MHz memory clock (2 Gbps effective) delivering 1.02 TB/s. The MI60’s bus width is 16 times wider, but its effective clock is eight times slower, resulting in exactly double the bandwidth. This is a classic trade-off between wide-and-slow (HBM2) versus narrow-and-fast (GDDR6) memory designs.

Compute resources differ substantially. The V620 has 4608 shading units, 288 TMUs, and 128 ROPs, with a pixel rate of 281.6 GPixel/s and a texture rate of 633.6 GTexel/s. The MI60 has 4096 shaders, 256 TMUs, and 64 ROPs, with a pixel rate of 115.2 GPixel/s and a texture rate of 460.8 GTexel/s. The V620’s ROP count is double the MI60’s, which is why its pixel rate is more than double. The V620 also has 72 ray-tracing cores, a feature completely absent from the MI60. Neither card has tensor cores.

Clock speeds favor the V620 as well. The V620 has a base clock of 1825 MHz and a boost clock of 2200 MHz, while the MI60 has a base of 1200 MHz and a boost of 1800 MHz. The V620’s higher clocks, combined with its larger shader count, explain its commanding lead in FP32 (20.28 TFLOPS vs. 14.75 TFLOPS) and FP16 (40.55 TFLOPS vs. 29.49 TFLOPS) performance. Both cards have a 300 W TDP and are dual-slot designs, but the V620 requires 2x 8-pin power connectors while the MI60 uses 1x 6-pin + 1x 8-pin. Both are PCIe 4.0 x16.

API support differs slightly. The V620 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The MI60 supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.3. The V620’s higher Vulkan version and DirectX 12 Ultimate support align with its RDNA 2.0 design, which includes hardware ray tracing and other modern features. The MI60’s GCN 5.1 architecture is older and lacks these capabilities. The V620 was released on 2021-11-03, while the MI60 came earlier on 2018-11-17, a three-year gap that explains the architectural divergence. Both are end-of-life, with no successors listed.