AMD Radeon Instinct MI60 vs NVIDIA GeForce RTX 5090 D Comparison

AMD Radeon Instinct MI60 and NVIDIA GeForce RTX 5090 D represent two distinct eras of GPU design, separated by over six years of architectural evolution. The benchmark data shows a stark performance chasm, with the RTX 5090 D winning both head-to-head tests by overwhelming margins, yet the MI60 still holds relevance in specific compute contexts. This analysis walks through the raw numbers, architectural foundations, and practical implications for each card.

Head-to-Head Benchmarks

The two available head-to-head comparisons paint an unambiguous picture of generational dominance. In Geekbench OpenCL, the NVIDIA GeForce RTX 5090 D scores 310,674 against the AMD Radeon Instinct MI60’s 92,488, a delta of -70.2% for the AMD part. This means the RTX 5090 D delivers roughly 3.36 times the OpenCL compute throughput of the MI60. The Vulkan results are even more lopsided: the RTX 5090 D posts 376,915 versus the MI60’s 92,444, a -75.5% delta, representing a 4.08x advantage.

These results align with the broader benchmark averages. The MI60’s average benchmark score sits at 92,466, placing it in the 93rd percentile of all GPUs. The RTX 5090 D, despite its individual test dominance, has a lower average score of 77,712 due to the inclusion of several Passmark tests (DirectX 9, 10, 11, 12, and 2D) that drag down its aggregate. Its percentile rank is 92nd, just one point below the MI60. This discrepancy highlights how benchmark selection dramatically influences perceived performance — the MI60 excels in the two Geekbench tests it appears in, while the RTX 5090 D’s broader test suite reveals less impressive relative performance in legacy DirectX workloads.

Looking at nearest rivals provides additional context. The MI60’s closest competitor is the NVIDIA RTX A4500, with a delta of only 0.9% in the MI60’s favor, indicating that the MI60 remains competitive with professional GPUs from several generations later. Conversely, the RTX 5090 D’s nearest rivals include the AMD Radeon RX 6650M XT (1.1% faster than the 5090 D) and the RX 6850M XT (-1.6% slower), showing that its average score is heavily influenced by less demanding tests. The RTX 5090 D’s Geekbench scores, however, are in a completely different league — its Vulkan score of 376,915 is 4x higher than the MI60’s, making it one of the highest single-test results in the database.

Architecture Differences

The architectural gulf between these two cards is vast. The MI60 uses the Vega 20 chip based on GCN 5.1 architecture, manufactured on TSMC’s 7 nm process. It contains 13,230 million transistors on a 331 mm² die, yielding a transistor density of 40.0 million per mm². In contrast, the RTX 5090 D uses the GB202 chip based on Blackwell 2.0, built on a 5 nm process, with 92,200 million transistors packed into a 750 mm² die — a density of 122.9 million per mm², over three times denser.

Core configurations differ dramatically. The MI60 has 4,096 shading units, 256 TMUs, and 64 ROPs, with no dedicated ray tracing or tensor cores. The RTX 5090 D has 21,760 shading units, 680 TMUs, and 176 ROPs, plus 170 RT cores and 680 tensor cores. This results in a 5.3x increase in shading units and a 7.1x increase in texture mapping units. Clock speeds tell a similar story: the MI60 runs at 1200 MHz base and 1800 MHz boost, while the RTX 5090 D runs at 2017 MHz base and 2407 MHz boost — a 34% higher boost clock.

Memory subsystems diverge completely. The MI60 uses 32 GB of HBM2 on a 4096-bit bus, achieving 1.02 TB/s bandwidth at 1000 MHz (2 Gbps effective). The RTX 5090 D uses 32 GB of GDDR7 on a 512-bit bus, reaching 1.79 TB/s bandwidth at 1750 MHz (28 Gbps effective). While both have identical capacity, the RTX 5090 D delivers 75% more bandwidth. Pixel and texture rates reflect the core count differences: the MI60 produces 115.2 GPixel/s and 460.8 GTexel/s, versus the RTX 5090 D’s 423.6 GPixel/s and 1,636.8 GTexel/s. FP32 compute is 14.75 TFLOPS for the MI60 versus 104.8 TFLOPS for the RTX 5090 D — a 7.1x difference. FP16 performance shows a notable architectural choice: the MI60 achieves 29.49 TFLOPS via 2:1 ratio, while the RTX 5090 D matches its FP32 at 104.8 TFLOPS with 1:1 ratio.

Where Each One Wins

Despite the RTX 5090 D’s overwhelming compute advantage, the MI60 has clear strengths in specific areas. The MI60’s HBM2 memory, with its 4096-bit bus, offers lower latency characteristics that can benefit certain scientific computing workloads where memory access patterns are less predictable. Its 32 GB capacity equals the RTX 5090 D, making it viable for large datasets, but at 1.02 TB/s bandwidth it cannot match the GDDR7’s throughput. The MI60’s 300 W TDP is substantially lower than the RTX 5090 D’s 575 W, making it a more manageable option for dense server deployments where power density is a constraint. Its 267 mm length (10.5 inches) also fits in more chassis than the RTX 5090 D’s 304 mm (12 inches).

The RTX 5090 D wins decisively in every raw performance metric. Its 4x Vulkan advantage and 3.36x OpenCL advantage make it the clear choice for any compute task that scales with raw throughput. The presence of 170 RT cores and 680 tensor cores enables ray tracing and AI acceleration that the MI60 lacks entirely. The RTX 5090 D’s PCIe 5.0 x16 interface doubles the bandwidth of the MI60’s PCIe 4.0 x16, reducing data transfer bottlenecks. Its 1.79 TB/s memory bandwidth is critical for bandwidth-bound workloads like large language model inference or high-resolution rendering.

For gaming and consumer workloads, the RTX 5090 D is the only viable option — the MI60 has just a single mini-DisplayPort 1.4a output, while the RTX 5090 D provides 1x HDMI 2.1b and 3x DisplayPort 2.1b. The RTX 5090 D also supports DirectX 12 Ultimate (12_2) and Vulkan 1.4, versus the MI60’s DirectX 12 (12_1) and Vulkan 1.3. The MI60’s production status is end-of-life, while the RTX 5090 D is active with a successor planned (GeForce 60).

The Verdict

The data unequivocally favors the RTX 5090 D for anyone seeking maximum performance. Its 104.8 TFLOPS FP32 compute, 1.79 TB/s bandwidth, and 4x Vulkan benchmark advantage make it the superior choice for compute-intensive applications, AI workloads, and modern gaming. The RTX 5090 D’s 170 RT cores and 680 tensor cores provide hardware acceleration for ray tracing and neural network operations that the MI60 cannot perform at all. Its active production status and PCIe 5.0 interface ensure forward compatibility with current and future platforms.

The MI60 remains relevant only in narrow, specific circumstances. Its 300 W TDP and compact 267 mm length make it easier to deploy in power-constrained or space-constrained environments. Its 93rd percentile ranking versus the RTX 5090 D’s 92nd percentile suggests that in some benchmark suites, the MI60’s consistent performance across its tested workloads gives it a higher aggregate standing. For organizations with existing GCN 5.1 codebases optimized for the MI60’s architecture, the migration cost to Blackwell might outweigh the 3-4x performance gain. However, these cases are diminishing as the MI60 is end-of-life with no successor.

The practical recommendation is straightforward: choose the RTX 5090 D for new deployments, performance-critical tasks, or any workload requiring modern features like ray tracing, tensor operations, or DirectX 12 Ultimate. Choose the MI60 only if you have legacy software that requires GCN 5.1-specific optimizations, or if your power and space envelopes cannot accommodate a 575 W, 304 mm dual-slot card. The 7.1x FP32 compute advantage and 75% memory bandwidth increase of the RTX 5090 D are too substantial to ignore for most users.

FAQ

Q: Which card has higher Geekbench Vulkan performance?

A: The NVIDIA GeForce RTX 5090 D scores 376,915 in Geekbench Vulkan, compared to the AMD Radeon Instinct MI60’s 92,444, making the RTX 5090 D 75.5% faster (a 4.08x advantage).

Q: Do both cards have the same memory capacity?

A: Yes, both the AMD Radeon Instinct MI60 and NVIDIA GeForce RTX 5090 D have 32 GB of memory. However, the MI60 uses HBM2 with a 4096-bit bus, while the RTX 5090 D uses GDDR7 with a 512-bit bus.

Q: What is the power consumption difference?

A: The AMD Radeon Instinct MI60 has a 300 W TDP, while the NVIDIA GeForce RTX 5090 D has a 575 W TDP. The suggested PSU is 700 W for the MI60 and 950 W for the RTX 5090 D.

Q: Which card has ray tracing cores?

A: Only the NVIDIA GeForce RTX 5090 D has ray tracing cores, with 170 RT cores. The AMD Radeon Instinct MI60 has no RT cores or tensor cores.

Q: What is the FP32 compute capability of each card?

A: The AMD Radeon Instinct MI60 delivers 14.75 TFLOPS FP32, while the NVIDIA GeForce RTX 5090 D delivers 104.8 TFLOPS FP32 — a 7.1x difference.

Q: Which card has a higher transistor density?

A: The NVIDIA GeForce RTX 5090 D has a transistor density of 122.9M per mm², versus the AMD Radeon Instinct MI60’s 40.0M per mm², due to the newer 5 nm process versus the MI60’s 7 nm process.

Specification Differences

| Specification | AMD Radeon Instinct MI60 | NVIDIA GeForce RTX 5090 D |

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

| Architecture | GCN 5.1 | Blackwell 2.0 |

| Process Node | 7 nm | 5 nm |

| Transistors | 13,230 million | 92,200 million |

| Die Size | 331 mm² | 750 mm² |

| Transistor Density | 40.0M / mm² | 122.9M / mm² |

| Base Clock | 1200 MHz | 2017 MHz |

| Boost Clock | 1800 MHz | 2407 MHz |

| Memory Type | HBM2 | GDDR7 |

| Memory Bus Width | 4096 bit | 512 bit |

| Memory Bandwidth | 1.02 TB/s | 1.79 TB/s |

| Shading Units | 4096 | 21760 |

| TMUs | 256 | 680 |

| ROPs | 64 | 176 |

| RT Cores | N/A | 170 |

| Tensor Cores | N/A | 680 |

| FP32 Performance | 14.75 TFLOPS | 104.8 TFLOPS |

| FP16 Performance | 29.49 TFLOPS (2:1) | 104.8 TFLOPS (1:1) |

| TDP | 300 W | 575 W |

| Power Connectors | 1x 6-pin + 1x 8-pin | 1x 16-pin |

| Suggested PSU | 700 W | 950 W |

| Bus Interface | PCIe 4.0 x16 | PCIe 5.0 x16 |

| Display Outputs | 1x mini-DisplayPort 1.4a | 1x HDMI 2.1b3x DisplayPort 2.1b |

| DirectX Support | 12 (12_1) | 12 Ultimate (12_2) |

| Vulkan Support | 1.3 | 1.4 |

| Length | 267 mm (10.5 inches) | 304 mm (12 inches) |

| Height | 111 mm (4.4 inches) | 137 mm (5.4 inches) |

| Width | N/A | 48 mm (1.9 inches) |

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

| Release Date | 2018-11-17 | 2025-01-29 |

| Predecessor | FirePro Data Center | GeForce 40 |

| Successor | N/A | GeForce 60 |

| Launch MSRP | N/A | 2,299 USD |