AMD Instinct MI300X vs AMD Radeon 680M Comparison

Head-to-Head Benchmarks

The only directly comparable benchmark in the database is Geekbench OpenCL, and the margin is extreme. The AMD Instinct MI300X scores 317,994, while the AMD Radeon 680M scores 23,468. That is a delta of 1,255%, meaning the MI300X delivers roughly 13.5 times the raw compute throughput in this workload. The MI300X takes the sole win in the head-to-head comparison, with 1 win versus 0 for the 680M.

Context from the database's nearest rival data makes the scale clearer. The MI300X sits at the 100th percentile of all GPUs, meaning no recorded part scores higher. Its nearest rivals are all NVIDIA data center accelerators: the B200 at 345,482 (8% ahead), the H200 NVL at 334,891 (5% ahead), the L40S at 295,763 (7.5% behind), and the RTX 6000 Ada Generation at 287,237 (10.7% behind). The 680M, by contrast, sits at the 57th percentile, and its nearest rivals are a mix of older and modern consumer parts: the GeForce GTX 580 at 15,283 (0.1% behind), the GeForce RTX 2060 at 15,290 (0.1% behind), the GeForce RTX 3050 OEM at 15,199 (0.5% ahead), and the Radeon RX 7600 at 15,171 (0.7% ahead). The 680M's average benchmark score of 15,270 is less than one twentieth of the MI300X's single recorded score.

The MI300X's advantage in OpenCL is not incremental; it is an order of magnitude. The 680M's score of 23,468 would need to be multiplied more than thirteen times to match. This reflects the fundamental positioning of the two parts. The MI300X is a data center accelerator designed for massive parallel workloads, while the 680M is an integrated GPU aimed at portable devices. The benchmark gap is a direct consequence of that design intent.

Where Each One Wins

The MI300X wins on raw compute, and it wins everywhere that raw compute matters. Its OpenCL score of 317,994 places it at the absolute top of the database's percentile ranking. For workloads such as large-scale inference, scientific simulation, or any compute task that can saturate thousands of shaders, the MI300X is in a different class. Its nearest rivals are all NVIDIA accelerators, and it beats two of them outright: the L40S by 7.5% and the RTX 6000 Ada Generation by 10.7%. It trails only the B200 and H200 NVL, and even then by single-digit margins. The data shows a part that competes at the very top of the accelerator market.

The 680M wins on integration and portability. It is an IGP with a 50 W TDP, no external power connectors, and display outputs that are portable device dependent. It supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, making it a functional graphics solution for a laptop or handheld. Its Vulkan score of 21,965 and OpenCL score of 23,468 are modest but usable for light gaming and general GPU acceleration. The 680M also has 12 ray tracing cores, a feature the MI300X lacks entirely. For a device that needs graphics output and basic compute in a low-power envelope, the 680M is the only viable choice of the two. The MI300X has no display outputs at all.

The split is clean. The MI300X is for compute without display. The 680M is for display with modest compute. Neither part is a substitute for the other.

Architecture Differences

The two GPUs come from different architecture families and different process nodes. The MI300X uses CDNA 3.0, AMD's compute-optimized architecture, built on a 5 nm TSMC process. The 680M uses RDNA 2.0, a graphics-oriented architecture, built on a 6 nm TSMC process. The MI300X's chip is codenamed Aqua Vanjaram, while the 680M uses Rembrandt+.

Transistor counts differ by more than an order of magnitude. The MI300X packs 153,000 million transistors on a 1017 mm² die, for a density of 150.4 million transistors per square millimeter. The 680M has 13,100 million transistors on a 208 mm² die, for a density of 63.0 million per square millimeter. The MI300X is a massive chip; the 680M is a modest integrated slice of a mobile APU.

Compute resources follow the same pattern. The MI300X has 19,456 shading units, 1,216 texture mapping units, and 0 ROPs. The 680M has 768 shading units, 48 TMUs, and 32 ROPs. The MI300X has no ray tracing cores; the 680M has 12. Pixel rate for the MI300X is listed as 0 MPixel/s, since it has no ROPs and no display path. The 680M delivers 70.40 GPixel/s. Texture rate tells the opposite story: the MI300X hits 2,553.6 GTexel/s versus 105.6 GTexel/s for the 680M.

Memory architecture is entirely different. The MI300X uses 192 GB of HBM3 on an 8192-bit bus, with 5.32 TB/s of bandwidth and a memory clock of 1300 MHz (5.2 Gbps effective). The 680M uses system shared memory, with system dependent bandwidth and no dedicated VRAM. The MI300X's memory subsystem is built for massive data movement; the 680M relies on whatever the host system provides.

Floating point performance reflects the design goals. The MI300X delivers 81.72 TFLOPS FP32 and 81.72 TFLOPS FP16 at a 1:1 ratio, indicating a compute part that does not rely on packed math for acceleration. The 680M delivers 3.379 TFLOPS FP32 and 6.758 TFLOPS FP16 at a 2:1 ratio, a more conventional graphics-oriented arrangement where FP16 is used to double throughput.

The API support also differs. The MI300X lists DirectX, OpenGL, and Vulkan as N/A. The 680M supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The MI300X is not a graphics card in the conventional sense; the 680M is.

FAQ

Q: Which GPU has the higher OpenCL benchmark score?

A: The AMD Instinct MI300X scores 317,994 in Geekbench OpenCL, compared to 23,468 for the AMD Radeon 680M, a 1,255% difference.

Q: Does the AMD Radeon 680M support ray tracing?

A: Yes, the 680M includes 12 ray tracing cores. The MI300X has no ray tracing cores listed in the database.

Q: What memory does each GPU use?

A: The MI300X uses 192 GB of HBM3 on an 8192-bit bus with 5.32 TB/s bandwidth. The 680M uses system shared memory with system dependent bandwidth.

Q: Can the MI300X output video to a display?

A: No. The MI300X has no display outputs and lists DirectX, OpenGL, and Vulkan as N/A. The 680M has display outputs described as portable device dependent.

Q: How do the two compare in terms of power consumption?

A: The MI300X has a TDP of 750 W and requires a 1150 W suggested PSU. The 680M has a TDP of 50 W and no suggested PSU listed.

Q: What is the release date for each part?

A: The MI300X was released on December 5, 2023. The 680M was released on January 2, 2023.

Specification Differences

| Specification | AMD Instinct MI300X | AMD Radeon 680M |

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

| Architecture | CDNA 3.0 | RDNA 2.0 |

| Chip | Aqua Vanjaram | Rembrandt+ |

| Process Node | 5 nm | 6 nm |

| Transistors | 153,000 million | 13,100 million |

| Die Size | 1017 mm² | 208 mm² |

| Transistor Density | 150.4M / mm² | 63.0M / mm² |

| Base Clock | 1000 MHz | 2000 MHz |

| Boost Clock | 2100 MHz | 2200 MHz |

| Memory Size | 192 GB | System Shared |

| Memory Type | HBM3 | System Shared |

| Memory Bus Width | 8192 bit | System Shared |

| Memory Bandwidth | 5.32 TB/s | System Dependent |

| Shading Units | 19456 | 768 |

| TMUs | 1216 | 48 |

| ROPs | 0 | 32 |

| Ray Tracing Cores | None | 12 |

| Pixel Rate | 0 MPixel/s | 70.40 GPixel/s |

| Texture Rate | 2,553.6 GTexel/s | 105.6 GTexel/s |

| FP32 | 81.72 TFLOPS | 3.379 TFLOPS |

| FP16 | 81.72 TFLOPS (1:1) | 6.758 TFLOPS (2:1) |

| TDP | 750 W | 50 W |

| Slot Width | OAM Module | IGP |

| Power Connectors | None | None |

| Suggested PSU | 1150 W | None |

| Bus Interface | PCIe 5.0 x16 | PCIe 4.0 x8 |

| Display Outputs | No outputs | Portable Device Dependent |

| DirectX | N/A | 12 Ultimate (12_2) |

| OpenGL | N/A | 4.6 |

| Vulkan | N/A | 1.4 |

| Production Status | Not listed | Active |

| Release Date | 2023-12-05 | 2023-01-02 |

| Predecessor | Radeon Instinct | Vega II IGP |

| Successor | None | Navi III IGP |

The Verdict

The data supports only one choice for raw compute: the AMD Instinct MI300X. It scores 317,994 in OpenCL, sits at the 100th percentile of all GPUs, and beats the NVIDIA L40S by 7.5% and the RTX 6000 Ada Generation by 10.7%. It has 192 GB of HBM3 memory, 19,456 shading units, and 81.72 TFLOPS of FP32 throughput. For anyone building a compute node for AI inference, scientific workloads, or any heavily parallel task, the MI300X is the part the benchmark results point to. Its only recorded rivals above it are the NVIDIA B200 and H200 NVL, and the margins are 8% and 5% respectively.

The AMD Radeon 680M is the only sensible choice for a portable device. It is an IGP with a 50 W TDP, display outputs, DirectX 12 Ultimate support, and 12 ray tracing cores. Its OpenCL score of 23,468 and Vulkan score of 21,965 are low in absolute terms, but they place it at the 57th percentile, squarely among capable consumer parts like the GeForce RTX 2060 and Radeon RX 7600. It can render graphics and run accelerated workloads in a laptop or handheld. The MI300X cannot do any of that, as it has no display path and no consumer API support.

There is no scenario where these two parts compete for the same socket. The MI300X is a 750 W accelerator module with no video outputs. The 680M is a 50 W integrated GPU built into a mobile processor. The benchmark data confirms the obvious: the MI300X is more than an order of magnitude faster in compute, and the 680M is the only one of the two that can drive a screen. Pick the MI300X for a server. Pick the 680M for a portable device. The recorded measurements do not support any other conclusion.