AMD Instinct MI308X vs AMD Radeon 860M Comparison

Head-to-Head Benchmarks

The AMD Instinct MI308X and AMD Radeon 860M occupy opposite ends of the accelerator spectrum, and the recorded data confirms a complete separation in performance class. The MI308X delivers 81.72 TFLOPS of FP32 compute, while the 860M delivers 3.072 TFLOPS. This represents a 26.6x gap in raw floating-point throughput, a margin that dwarfs any comparison found in the nearest rival data for the 860M. The texture rate tells a similar story: the MI308X achieves 2,553.6 GTexel/s against 96.00 GTexel/s for the 860M, a 26.6x difference that scales directly from the shading unit count.

Pixel throughput is the one area where the conventional relationship inverts. The MI308X records 0 MPixel/s because it has no ROPs, so its pixel rate is effectively zero in the database. The 860M, by contrast, produces 48.00 GPixel/s from its 16 ROPs. This is not a competitive advantage in any practical sense, but it is a structural difference: the MI308X is a compute accelerator with no display or rasterization pipeline, whereas the 860M is a full graphics processor.

The 860M has two recorded benchmark scores in the database. Its Geekbench OpenCL score is 22,759, and its Geekbench Vulkan score is 30,043, giving an average benchmark score of 26,401. That average places it at the 72nd percentile among all GPUs. The nearest rivals bracket it tightly: the NVIDIA GeForce MX550 scores 26,421 (0.1% lower), the NVIDIA GeForce RTX 5060 scores 26,331 (0.3% higher), the AMD Radeon RX 5700 XT 50th Anniversary scores 26,553 (0.6% lower), and the NVIDIA RTX A4000 scores 26,683 (1.1% lower). The 860M sits within a 1.4% band across these four rivals, indicating that its compute performance is closely clustered with mid-range discrete graphics solutions from the previous generation. The MI308X has no benchmark scores and no nearest rivals recorded in the database, which reflects its positioning as a specialized accelerator rather than a general-purpose graphics card.

Memory bandwidth reinforces the class divide. The MI308X accesses 192 GB of HBM3 through an 8192-bit bus, yielding 5.32 TB/s of bandwidth. The 860M relies on System Shared memory, with bandwidth listed as System Dependent. No comparable figure exists for the integrated part, but the architectural difference is clear: the MI308X is built to feed 19,456 shading units, while the 860M's 512 shading units are served by the host memory subsystem.

Architecture Differences

The two processors come from different design lineages within AMD. The MI308X uses the Aqua Vanjaram chip with CDNA 3.0 architecture, part of the Instinct (MIx) generation. It is fabricated by TSMC on a 5 nm process, integrating 153,000 million transistors on a 1017 mm² die, for a transistor density of 150.4M per mm². The 860M uses the Krackan Point chip with RDNA 3.5 architecture, part of the Navi III IGP (Strix Point Mobile) generation. It is fabricated by TSMC on a 4 nm process, though its transistor count and die size are not recorded in the database.

Execution resources differ by an order of magnitude. The MI308X carries 19,456 shading units, 1,216 TMUs, and no ROPs. The 860M carries 512 shading units, 32 TMUs, 16 ROPs, and 8 ray tracing cores. The MI308X has no ray tracing cores listed, and neither part lists tensor cores. Clock behavior also diverges sharply. The MI308X runs at a base clock of 1000 MHz and a boost clock of 2100 MHz. The 860M runs at a base clock of 600 MHz and a boost clock of 3000 MHz. Despite the 860M's higher boost frequency, its much smaller execution width limits its throughput.

Memory architecture is fundamentally different. The MI308X uses 192 GB of HBM3 with an 8192-bit bus and 5.32 TB/s bandwidth, with memory clocked at 1300 MHz (5.2 Gbps effective). The 860M uses System Shared memory with a System Shared bus width and System Dependent bandwidth. The MI308X lists no display outputs, while the 860M's outputs are Portable Device Dependent. The MI308X supports no graphics APIs (DirectX, OpenGL, and Vulkan are all listed as N/A), while the 860M supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Power and integration also separate the two. The MI308X is an OAM Module with a 750 W TDP and no power connectors, requiring a suggested PSU of 1150 W. The 860M is an IGP with a 15 W TDP and no power connectors. Bus interfaces differ: the MI308X uses PCIe 5.0 x16, while the 860M uses PCIe 4.0 x8. The MI308X was released on 2023-12-05, while the 860M was released on 2025-02-28. The MI308X's predecessor is Radeon Instinct, and the 860M's predecessor is Navi II IGP. The 860M has an Active production status; the MI308X's production status is not recorded.

The Verdict

The data indicates that these two products are not competitors. The MI308X is a 750 W accelerator with 81.72 TFLOPS of FP32 performance, 192 GB of HBM3, and no display or graphics API support. The 860M is a 15 W integrated processor with 3.072 TFLOPS, ray tracing cores, full graphics API support, and a 72nd percentile benchmark standing. Any selection between them depends entirely on workload class.

For compute workloads that require massive FP32 throughput, large memory capacity, and high bandwidth, the MI308X is the only viable option in this comparison. Its 5.32 TB/s memory bandwidth and 192 GB capacity are orders of magnitude beyond what the 860M can access. For graphics workloads, gaming, or any application requiring DirectX, OpenGL, or Vulkan, the 860M is the only option, because the MI308X has no graphics API support at all.

The 860M's benchmark data shows it performs within 1.1% of the NVIDIA RTX A4000 and within 0.3% of the NVIDIA GeForce RTX 5060, making it a capable integrated solution relative to those discrete parts. The MI308X has no recorded benchmarks or rivals, so its relative standing cannot be quantified from the database, only its absolute specifications.

FAQ

Q: Which GPU is faster in FP32 compute?

A: The AMD Instinct MI308X delivers 81.72 TFLOPS of FP32, which is 26.6 times the 3.072 TFLOPS of the AMD Radeon 860M.

Q: Does the AMD Radeon 860M support ray tracing?

A: Yes, the 860M has 8 ray tracing cores, while the MI308X lists no ray tracing cores.

Q: Which GPU has more memory bandwidth?

A: The MI308X has 5.32 TB/s of bandwidth from 192 GB of HBM3 on an 8192-bit bus. The 860M uses System Shared memory with System Dependent bandwidth.

Q: What is the 860M's average benchmark score and how does it compare to its rivals?

A: The 860M's average benchmark score is 26,401. It is 0.1% behind the NVIDIA GeForce MX550 (26,421), 0.3% ahead of the NVIDIA GeForce RTX 5060 (26,331), 0.6% behind the AMD Radeon RX 5700 XT 50th Anniversary (26,553), and 1.1% behind the NVIDIA RTX A4000 (26,683).

Q: Can the MI308X output video to a display?

A: No, the MI308X lists "No outputs" for display outputs and has no graphics API support. The 860M lists Portable Device Dependent display outputs and supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4.

Q: What are the power requirements of each GPU?

A: The MI308X has a 750 W TDP and requires a suggested PSU of 1150 W. The 860M has a 15 W TDP and no suggested PSU is listed.

Where Each One Wins

The MI308X wins decisively in compute throughput. Its 81.72 TFLOPS FP32 and FP16 (1:1) performance, 2,553.6 GTexel/s texture rate, and 5.32 TB/s memory bandwidth are built for data-center scale computation. The 192 GB HBM3 pool allows working sets that the 860M cannot approach with System Shared memory. The PCIe 5.0 x16 interface provides twice the lane width and a newer generation than the 860M's PCIe 4.0 x8. The MI308X also wins on transistor integration: 153,000 million transistors on a 1017 mm² die, compared to the 860M's unrecorded transistor count and die size.

The 860M wins in graphics capability. It is the only part of the two with ROPs (16), pixel rate (48.00 GPixel/s), and ray tracing cores (8). It supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4, while the MI308X supports none. The 860M wins on power efficiency per the recorded TDP: 15 W versus 750 W, a 50x difference. It also wins on clock speed, with a 3000 MHz boost versus 2100 MHz, and on production status, listed as Active while the MI308X has no recorded status. The 860M's benchmark presence gives it a measurable standing (72nd percentile) that the MI308X lacks entirely.

Specification Differences

The two parts differ in every major specification category recorded in the database.

Process node: 5 nm for the MI308X, 4 nm for the 860M. Foundry is TSMC for both.

Transistors: 153,000 million for the MI308X, unknown for the 860M. Die size: 1017 mm² for the MI308X, unknown for the 860M. Transistor density: 150.4M per mm² for the MI308X, not recorded for the 860M.

Clocks: the MI308X has a 1000 MHz base and 2100 MHz boost. The 860M has a 600 MHz base and 3000 MHz boost. Memory clock is 1300 MHz (5.2 Gbps effective) for the MI308X; the 860M's memory clock is System Shared.

Memory: 192 GB HBM3 on an 8192-bit bus with 5.32 TB/s bandwidth for the MI308X. System Shared size, type, bus width, and System Dependent bandwidth for the 860M.

Execution units: 19,456 shading units, 1,216 TMUs, 0 ROPs for the MI308X. 512 shading units, 32 TMUs, 16 ROPs, and 8 ray tracing cores for the 860M.

Rates: 0 MPixel/s and 2,553.6 GTexel/s for the MI308X. 48.00 GPixel/s and 96.00 GTexel/s for the 860M.

Compute: 81.72 TFLOPS FP32 and 81.72 TFLOPS FP16 (1:1) for the MI308X. 3.072 TFLOPS FP32 and 3.072 TFLOPS FP16 (1:1) for the 860M.

Power: 750 W TDP, OAM Module slot width, no power connectors, suggested PSU 1150 W for the MI308X. 15 W TDP, IGP slot width, no power connectors, no suggested PSU for the 860M.

Interface and outputs: PCIe 5.0 x16 and no display outputs for the MI308X. PCIe 4.0 x8 and Portable Device Dependent display outputs for the 860M.

APIs: none for the MI308X. DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4 for the 860M.

Release dates: 2023-12-05 for the MI308X, 2025-02-28 for the 860M. Predecessors: Radeon Instinct for the MI308X, Navi II IGP for the 860M. Production status: not recorded for the MI308X, Active for the 860M.