AMD Radeon Instinct MI60 vs NVIDIA GB10 Comparison

Head-to-Head Benchmarks

The benchmark data presents a decisive picture: the NVIDIA GB10 outperforms the AMD Radeon Instinct MI60 in both recorded tests. In Geekbench OpenCL, the GB10 scores 120,137 against the MI60’s 92,488, a 29.9% advantage. The Vulkan test shows a similar margin, with the GB10 at 114,648 versus 92,444 for the MI60, a 24% lead. Across both workloads, the GB10 wins two out of two comparisons, and its average benchmark score of 117,393 stands 27% above the MI60’s 92,466.

The GB10’s dominance is further contextualized by its position among rivals. Its average score places it 0.3% ahead of the NVIDIA RTX 4000 SFF Ada Generation (117,088) and 1.3% behind the AMD Radeon PRO W7700 (118,976), but 2.6% ahead of the NVIDIA Tesla V100 SXM2 16 GB (114,395) and 3% ahead of the NVIDIA RTX A5500 Mobile (113,944). The MI60, by contrast, sits 0.9% above the NVIDIA RTX A4500 (91,671) and 1.5% above the RTX A4500 Mobile (91,134), yet trails the AMD Radeon Pro VII (97,131) by 4.8% and the AMD Radeon RX 7900M (97,487) by 5.2%. This places the GB10 in a higher performance tier altogether, with its nearest competitors being modern workstation parts, whereas the MI60 competes with previous-generation mid-range offerings.

The raw compute figures reinforce this gap. The GB10 delivers 29.71 TFLOPS of FP32 throughput, exactly double the MI60’s 14.75 TFLOPS. In FP16, the GB10 again reaches 29.71 TFLOPS with a 1:1 ratio, while the MI60 achieves 29.49 TFLOPS but only with a 2:1 rate, meaning its FP16 performance comes with reduced precision or requires specialized instructions. The GB10’s pixel rate of 116.1 GPixel/s is nearly identical to the MI60’s 115.2 GPixel/s, but its texture rate of 928.5 GTexel/s more than doubles the MI60’s 460.8 GTexel/s. These numbers indicate that the GB10 is not merely faster in aggregate scores but also in raw throughput metrics that matter for compute-heavy workloads.

The Verdict

From the data alone, the NVIDIA GB10 is the clear choice for any application prioritizing raw compute performance. Its 29.9% lead in OpenCL and 24% lead in Vulkan over the MI60 are substantial, and its average score places it in the 95th percentile of all GPUs, versus the 93rd percentile for the MI60. The GB10’s FP32 throughput is exactly double, and its texture rate is 101.5% higher, making it the superior option for tasks that stress shader or texture operations. The GB10 also holds a significant memory capacity advantage at 128 GB versus 32 GB, which matters for large datasets that exceed the MI60’s frame buffer.

However, the MI60 is not without merit in specific contexts. Its memory bandwidth of 1.02 TB/s is nearly four times the GB10’s 273.2 GB/s, thanks to a 4096-bit bus versus 256-bit. For workloads that are memory-bandwidth bound rather than compute bound, this could offset some of the GB10’s raw compute lead. The MI60 also supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.3, whereas the GB10 lists no API support, suggesting the MI60 may be more suitable for graphics or compute stacks that require these interfaces. The MI60’s 32 GB of HBM2 memory, while smaller in capacity, offers far higher bandwidth per byte, which can be critical for certain scientific simulations.

The production statuses differ as well: the GB10 is active, while the MI60 is end-of-life. For new deployments, the GB10 is the only forward-looking option. The GB10’s launch MSRP is 3,999 USD, while the MI60 has no listed launch MSRP. The GB10’s 140 W TDP and 300 W suggested PSU also make it far more power-efficient than the MI60’s 300 W TDP and 700 W suggested PSU, though the MI60’s dual-slot design with 1x 6-pin + 1x 8-pin connectors contrasts with the GB10’s integrated form factor and no power connectors.

Strictly from the data, the GB10 wins on compute, capacity, and efficiency, while the MI60 wins on bandwidth and API compatibility. Buyers needing maximum raw FP32 or FP16 performance should pick the GB10; those with memory-bandwidth-critical workloads or requiring legacy API support may find the MI60 adequate, but its end-of-life status argues against new adoption.

Architecture Differences

The two accelerators stem from fundamentally different architectures. The NVIDIA GB10 uses the GB20B chip based on Blackwell 2.0 architecture, manufactured on a 5 nm process at TSMC. The AMD Radeon Instinct MI60 uses the Vega 20 chip based on GCN 5.1 architecture, also built by TSMC but on a 7 nm process. This node difference is reflected in transistor density: the MI60 packs 13,230 million transistors into a 331 mm² die, yielding 40.0M transistors per mm², while the GB10’s transistor count is unknown but its die size is larger at 382 mm².

The GB10’s Blackwell architecture introduces dedicated RT cores (48) and tensor cores (384), neither of which the MI60 possesses. The GB10 also has 6,144 shading units, 384 TMUs, and 48 ROPs, whereas the MI60 has 4,096 shading units, 256 TMUs, and 64 ROPs. The higher shading unit count and TMU count explain the GB10’s superior texture rate, while the MI60’s higher ROP count gives it a slightly higher pixel rate per clock, though the overall pixel rates are nearly identical due to clock differences.

The GB10’s memory subsystem uses LPDDR5X on a 256-bit bus, while the MI60 uses HBM2 on a 4096-bit bus. This is a fundamental architectural choice: the GB10 opts for higher capacity (128 GB) with moderate bandwidth, while the MI60 opts for lower capacity (32 GB) with extremely high bandwidth. The GB10’s clock speeds are higher (base 1665 MHz, boost 2418 MHz) versus the MI60 (base 1200 MHz, boost 1800 MHz), which contributes to its compute advantage. The GB10 also features a PCIe 5.0 x16 interface versus the MI60’s PCIe 4.0 x16.

The GB10 has no API support listed for DirectX, OpenGL, or Vulkan, indicating it is purely a compute device, whereas the MI60 supports all three. The GB10’s display output is limited to 1x HDMI, while the MI60 has 1x mini-DisplayPort 1.4a. The GB10 is an IGP (integrated graphics processor) with 150 mm length, 51 mm height, and 150 mm width, while the MI60 is a dual-slot card measuring 267 mm in length and 111 mm in height.

Specification Differences

The following fields differ between the two accelerators:

| Field | NVIDIA GB10 | AMD Radeon Instinct MI60 |

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

| Chip | GB20B | Vega 20 |

| Architecture | Blackwell 2.0 | GCN 5.1 |

| Process Node | 5 nm | 7 nm |

| Transistors | unknown | 13,230 million |

| Die Size | 382 mm² | 331 mm² |

| Transistor Density | null | 40.0M / mm² |

| Base Clock | 1665 MHz | 1200 MHz |

| Boost Clock | 2418 MHz | 1800 MHz |

| Memory Clock | 1067 MHz 8.5 Gbps effective | 1000 MHz 2 Gbps effective |

| Memory Size | 128 GB | 32 GB |

| Memory Type | LPDDR5X | HBM2 |

| Memory Bus Width | 256 bit | 4096 bit |

| Memory Bandwidth | 273.2 GB/s | 1.02 TB/s |

| Shading Units | 6144 | 4096 |

| TMUs | 384 | 256 |

| ROPs | 48 | 64 |

| RT Cores | 48 | null |

| Tensor Cores | 384 | null |

| Pixel Rate | 116.1 GPixel/s | 115.2 GPixel/s |

| Texture Rate | 928.5 GTexel/s | 460.8 GTexel/s |

| FP32 | 29.71 TFLOPS | 14.75 TFLOPS |

| FP16 | 29.71 TFLOPS (1:1) | 29.49 TFLOPS (2:1) |

| TDP | 140 W | 300 W |

| Slot Width | IGP | Dual-slot |

| Power Connectors | None | 1x 6-pin + 1x 8-pin |

| Suggested PSU | 300 W | 700 W |

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

| Display Outputs | 1x HDMI | 1x mini-DisplayPort 1.4a |

| DirectX | N/A | 12 (12_1) |

| OpenGL | N/A | 4.6 |

| Vulkan | N/A | 1.3 |

| Dimensions | 150 mm x 51 mm x 150 mm | 267 mm x 111 mm |

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

| Release Date | 2025-10-14 | 2018-11-17 |

| Predecessor | Server Hopper | FirePro Data Center |

| Successor | Server Rubin | null |

| Launch MSRP | 3,999 USD | null |

FAQ

Q: Which GPU has higher FP32 performance?

A: The NVIDIA GB10 delivers 29.71 TFLOPS of FP32, exactly double the AMD Radeon Instinct MI60’s 14.75 TFLOPS.

Q: How much faster is the GB10 in OpenCL benchmarks?

A: The GB10 scores 120,137 in Geekbench OpenCL versus 92,488 for the MI60, a 29.9% advantage.

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

A: Yes, the MI60’s HBM2 memory provides 1.02 TB/s of bandwidth on a 4096-bit bus, compared to the GB10’s 273.2 GB/s on a 256-bit bus.

Q: What is the memory capacity difference?

A: The GB10 has 128 GB of LPDDR5X, while the MI60 has 32 GB of HBM2.

Q: Which GPU supports more graphics APIs?

A: The MI60 supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.3, while the GB10 lists no API support.

Q: What is the power consumption difference?

A: The GB10 has a 140 W TDP with a 300 W suggested PSU, while the MI60 has a 300 W TDP with a 700 W suggested PSU.