AMD Radeon Instinct MI60 vs NVIDIA Quadro P6000 Comparison

The data is unambiguous: the AMD Radeon Instinct MI60 outperforms the NVIDIA Quadro P6000 in both recorded benchmarks, securing a clear win in compute workloads. However, the two cards are built for fundamentally different purposes, and the choice between them hinges on whether raw compute throughput or established ecosystem features matter more for the intended use case.

Where Each One Wins

The AMD Radeon Instinct MI60 wins every benchmark recorded in the data, making it the outright performance leader in this comparison. Its average benchmark score of 92,466 places it in the 93rd percentile of all GPUs, while the NVIDIA Quadro P6000 sits at 69,986 in the 90th percentile. The MI60’s advantage is not marginal—it leads by 39.3% in Geekbench OpenCL and 25.6% in Geekbench Vulkan. This makes the MI60 the definitive choice for compute-heavy tasks such as machine learning training, scientific simulation, or any workload that leverages OpenCL or Vulkan APIs.

The NVIDIA Quadro P6000, despite losing both head-to-head benchmarks, still holds relevance in its own right. It achieves a Geekbench Vulkan score of 73,590, which is notably closer to the MI60’s 92,444 than the OpenCL gap suggests. This indicates that the P6000’s Vulkan performance is comparatively stronger relative to its OpenCL showing, making it a more balanced option for applications that rely on Vulkan. Additionally, the P6000’s nearest rivals in the data—such as the AMD Radeon Pro WX 8200 (69,870, deltaPct 0.2%) and the NVIDIA RTX A3000 Mobile (70,140, deltaPct -0.2%)—show that it is very much in line with contemporary workstation GPUs, meaning it is not obsolete by any stretch.

The MI60’s wins are decisive in raw compute, but the P6000’s strengths lie elsewhere: it offers a more mature display output configuration with 1x DVI and 4x DisplayPort 1.4a, compared to the MI60’s single mini-DisplayPort 1.4a. For multi-display professional visualization setups, the P6000’s connectivity is a practical advantage that no benchmark score can capture.

Head-to-Head Benchmarks

The Geekbench OpenCL result delivers the largest margin of victory. The MI60 scores 92,488 against the P6000’s 66,382, a delta of 39.3% in favor of AMD. This is a substantial lead, driven by the MI60’s superior raw compute specifications: 14.75 TFLOPS FP32 versus 12.63 TFLOPS, and a massive 1.02 TB/s memory bandwidth compared to 432.8 GB/s. The MI60 also has more shading units (4,096 vs 3,840) and TMUs (256 vs 240), though the P6000 counters with more ROPs (96 vs 64).

In Geekbench Vulkan, the MI60 wins again with 92,444 against 73,590, a 25.6% advantage. While this is a smaller gap than OpenCL, it is still a commanding lead. Interestingly, the P6000’s Vulkan score of 73,590 is actually higher than its OpenCL score of 66,382, suggesting that NVIDIA’s Pascal architecture handles Vulkan more efficiently relative to OpenCL. Meanwhile, the MI60’s scores are nearly identical across both APIs (92,488 vs 92,444), indicating exceptional API-agnostic consistency. For developers targeting Vulkan specifically, the P6000’s relative improvement in that API narrows the gap, but it does not eliminate the MI60’s overall superiority.

The data also shows the MI60’s performance context: it sits just 0.9% above the NVIDIA RTX A4500 (91,671) and 1.5% above the RTX A4500 Mobile (91,134), while trailing the AMD Radeon Pro VII by 4.8% and the Radeon RX 7900M by 5.2%. The P6000, by contrast, is virtually tied with the Radeon Pro WX 8200 (0.2% ahead) and the RTX A3000 Mobile (0.2% behind), showing it operates in a different performance tier altogether.

Architecture Differences

The MI60 is built on AMD’s GCN 5.1 architecture using the Vega 20 chip, fabricated on TSMC’s 7 nm process node. This is a massive process advantage over the P6000’s 16 nm node, allowing AMD to pack 13,230 million transistors into a 331 mm² die, yielding a transistor density of 40.0M per mm². The P6000, using NVIDIA’s Pascal architecture with the GP102 chip, manages 11,800 million transistors on a much larger 471 mm² die, resulting in just 25.1M transistors per mm². The MI60’s density advantage is nearly 60% higher, which explains its ability to deliver superior performance while maintaining similar physical dimensions (both are 267 mm in length and 111 mm in height, dual-slot cards).

Memory architecture is another major differentiator. The MI60 uses 32 GB of HBM2 with a 4096-bit bus, delivering 1.02 TB/s of bandwidth—more than double the P6000’s 432.8 GB/s from 24 GB of GDDR5X on a 384-bit bus. This bandwidth advantage is critical for memory-bound compute workloads. The MI60 also supports PCIe 4.0 x16, while the P6000 is limited to PCIe 3.0 x16, doubling the theoretical host interface bandwidth.

Compute capabilities diverge sharply in FP16. The MI60 delivers 29.49 TFLOPS FP16 (2:1 ratio), making it highly efficient for mixed-precision workloads. The P6000’s FP16 performance is a paltry 197.4 GFLOPS (1:64 ratio), effectively negligible for practical purposes. This makes the MI60 vastly more capable for AI inference or training that uses half-precision arithmetic.

Clock speeds tell a different story: the P6000 has a higher base clock (1506 MHz vs 1200 MHz) and a lower boost clock (1645 MHz vs 1800 MHz). The MI60’s higher boost clock, combined with its architectural efficiency, contributes to its lead in texture rate (460.8 GTexel/s vs 394.8 GTexel/s) and FP32 throughput. However, the P6000 wins in pixel rate (157.9 GPixel/s vs 115.2 GPixel/s), thanks to its 96 ROPs versus the MI60’s 64. This suggests the P6000 may be slightly better suited for traditional rasterization tasks.

Power consumption also favors the P6000: it draws 250 W TDP versus the MI60’s 300 W, and recommends a 600 W PSU versus 700 W. The P6000 requires a single 8-pin power connector, while the MI60 needs a 6-pin and an 8-pin. The MI60 also supports Vulkan 1.3, while the P6000 supports Vulkan 1.4, a minor software advantage for NVIDIA.

The Verdict

The AMD Radeon Instinct MI60 is the clear winner for compute-intensive workloads. It leads by 39.3% in OpenCL and 25.6% in Vulkan, offers double the memory bandwidth (1.02 TB/s vs 432.8 GB/s), and provides 32 GB of HBM2 versus 24 GB of GDDR5X. Its FP16 performance (29.49 TFLOPS) is an order of magnitude beyond the P6000’s (197.4 GFLOPS), making it the only sensible choice for machine learning or scientific computing that leverages half-precision. The MI60’s 7 nm process and PCIe 4.0 support also future-proof it better than the P6000’s 16 nm and PCIe 3.0.

The NVIDIA Quadro P6000, however, is not without merit. It wins on pixel rate (157.9 GPixel/s vs 115.2 GPixel/s), has a lower TDP (250 W vs 300 W), and offers far superior display connectivity with 1x DVI and 4x DisplayPort 1.4a versus the MI60’s lone mini-DisplayPort. Its Vulkan score of 73,590, while lower, is closer to the MI60’s 92,444 than its OpenCL score is, indicating better API efficiency for Vulkan workloads. The P6000 also has a higher base clock (1506 MHz vs 1200 MHz), which could translate to better performance in lightly-threaded or latency-sensitive tasks.

Pick the MI60 if your priority is raw compute throughput, memory bandwidth, or FP16 acceleration. Pick the P6000 if you need multi-display output, lower power draw, or are targeting Vulkan-specific workloads where its relative efficiency narrows the gap. The data does not support the P6000 as a compute leader, but it remains a competent workstation card for visualization and single-GPU tasks.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The AMD Radeon Instinct MI60 has an average benchmark score of 92,466, while the NVIDIA Quadro P6000 scores 69,986.

Q: How large is the OpenCL performance gap?

A: The MI60 wins Geekbench OpenCL with 92,488 against the P6000’s 66,382, a delta of 39.3%.

Q: Does the P6000 win any benchmarks?

A: No, the P6000 loses both recorded benchmarks. The MI60 wins Geekbench OpenCL and Geekbench Vulkan, giving AMD 2 wins and NVIDIA 0.

Q: What is the memory bandwidth difference?

A: The MI60 offers 1.02 TB/s from 32 GB HBM2 on a 4096-bit bus, while the P6000 provides 432.8 GB/s from 24 GB GDDR5X on a 384-bit bus.

Q: How do their FP16 capabilities compare?

A: The MI60 delivers 29.49 TFLOPS FP16 (2:1 ratio), while the P6000 manages only 197.4 GFLOPS (1:64 ratio)—a difference of roughly 150x.

Q: Which card supports more display outputs?

A: The P6000 has 1x DVI and 4x DisplayPort 1.4a, while the MI60 has just 1x mini-DisplayPort 1.4a.

Specification Differences

| Specification | AMD Radeon Instinct MI60 | NVIDIA Quadro P6000 |

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

| Architecture | GCN 5.1 | Pascal |

| Process Node | 7 nm | 16 nm |

| Transistors | 13,230 million | 11,800 million |

| Die Size | 331 mm² | 471 mm² |

| Transistor Density | 40.0M / mm² | 25.1M / mm² |

| Base Clock | 1200 MHz | 1506 MHz |

| Boost Clock | 1800 MHz | 1645 MHz |

| Memory Clock | 1000 MHz (2 Gbps effective) | 1127 MHz (9 Gbps effective) |

| Memory Size | 32 GB | 24 GB |

| Memory Type | HBM2 | GDDR5X |

| Memory Bus Width | 4096 bit | 384 bit |

| Memory Bandwidth | 1.02 TB/s | 432.8 GB/s |

| Shading Units | 4096 | 3840 |

| TMUs | 256 | 240 |

| ROPs | 64 | 96 |

| Pixel Rate | 115.2 GPixel/s | 157.9 GPixel/s |

| Texture Rate | 460.8 GTexel/s | 394.8 GTexel/s |

| FP32 | 14.75 TFLOPS | 12.63 TFLOPS |

| FP16 | 29.49 TFLOPS (2:1) | 197.4 GFLOPS (1:64) |

| TDP | 300 W | 250 W |

| Power Connectors | 1x 6-pin + 1x 8-pin | 1x 8-pin |

| Suggested PSU | 700 W | 600 W |

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

| Display Outputs | 1x mini-DisplayPort 1.4a | 1x DVI, 4x DisplayPort 1.4a |

| Vulkan Version | 1.3 | 1.4 |

| Release Date | 2018-11-17 | 2016-09-30 |

| Production Status | End-of-life | End-of-life |

| Launch MSRP | N/A | 5,999 USD |