AMD Instinct MI308X vs AMD Radeon 890M Comparison

The AMD Instinct MI308X and the AMD Radeon 890M occupy opposite ends of the GPU spectrum, one a massive accelerator for compute workloads and the other an integrated graphics processor for mobile devices. The recorded data shows no direct head-to-head benchmark comparisons between the two, as the MI308X has no benchmark scores in the database, while the 890M has a full suite of recorded measurements. The MI308X holds a higher percentile ranking at 50 compared to the 890M’s 45, but the MI308X’s average benchmark score is recorded as zero, reflecting the absence of standardized tests for this class of hardware. The 890M, by contrast, has an average benchmark score of 9210 across ten tests. This analysis relies entirely on the specification differences and the available performance data for the 890M, using the MI308X’s architectural details as the basis for comparison.

The Verdict

The data indicates that these two products serve entirely different purposes, and the choice between them depends on the workload and platform constraints. For any task involving standard graphics APIs, the 890M is the only option with recorded functionality. The MI308X lists no APIs for DirectX, OpenGL, or Vulkan, and has no display outputs, making it unsuitable for conventional rendering or desktop use. The 890M supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, and its display outputs are described as portable device dependent, confirming its role in mobile systems.

For compute-heavy workloads, the MI308X presents a vastly larger resource pool. It has 19,456 shading units, 192 GB of HBM3 memory, and 5.32 TB/s of memory bandwidth. The 890M has 1,024 shading units, system shared memory, and bandwidth that is system dependent. The MI308X’s compute capabilities, such as 81.72 TFLOPS for both FP32 and FP16, dwarf the 890M’s 5.939 TFLOPS for both precision levels. The verdict from the specifications alone is clear: the MI308X is designed for high-throughput server or accelerator workloads, while the 890M is an integrated solution for portable devices where power draw and physical size are critical constraints.

The 890M’s benchmark results show it competes closely with older discrete graphics cards. Its average score of 9210 places it within 1% of the AMD Radeon Vega 8, which scores 9221, and within 1% of the NVIDIA GeForce GTX 960, which scores 9273. These comparisons suggest that the 890M delivers performance in line with a modest discrete GPU from previous generations, despite being an integrated processor. The MI308X has no such comparisons in the database, meaning its performance relative to other accelerators cannot be quantified here.

Architecture Differences

The MI308X uses the CDNA 3.0 architecture, built on a 5 nm process at TSMC, and features the Aqua Vanjaram chip. The 890M uses the RDNA 3.5 architecture, built on a 4 nm process also at TSMC, with the Strix Point chip. The transistor counts differ substantially: the MI308X has 153,000 million transistors on a 1017 mm² die, while the 890M has 34,000 million transistors on a 233 mm² die. Transistor density is similar, with the MI308X at 150.4 million transistors per mm² and the 890M at 145.9 million per mm², indicating that the process nodes yield comparable packing efficiency despite the size difference.

Clock speeds reveal different design philosophies. The MI308X has a base clock of 1000 MHz and a boost clock of 2100 MHz, while the 890M starts at 400 MHz and boosts to 2900 MHz. The 890M’s higher boost clock reflects its focus on burst performance in power-constrained environments, whereas the MI308X prioritizes sustained throughput across a massive array of compute units. The MI308X has no render output units, a pixel rate of 0 MPixel/s, and no ray tracing cores, reinforcing its role as a compute accelerator rather than a graphics renderer. The 890M has 32 render output units, a pixel rate of 92.80 GPixel/s, and 16 ray tracing cores, making it a full-featured graphics processor.

Memory architecture is another major divergence. The MI308X uses 192 GB of HBM3 memory on an 8192-bit bus, achieving 5.32 TB/s of bandwidth. The 890M uses system shared memory with a system dependent bandwidth, which ties its performance to the host system’s memory configuration. The MI308X’s memory clock is 1300 MHz with 5.2 Gbps effective speed, while the 890M’s memory clock is listed as system shared. The MI308X’s power consumption is recorded at 750 W with a suggested power supply of 1150 W, whereas the 890M draws 15 W and has no suggested power supply listed.

Where Each One Wins

The MI308X wins decisively in raw compute throughput. Its FP32 performance of 81.72 TFLOPS is over 13 times higher than the 890M’s 5.939 TFLOPS. Texture rate also favors the MI308X at 2,553.6 GTexel/s compared to 185.6 GTexel/s for the 890M. The MI308X’s memory bandwidth of 5.32 TB/s provides a massive advantage for workloads that stream large datasets, such as scientific simulations or machine learning training, where the 890M’s system dependent bandwidth would be a limiting factor. The MI308X’s 192 GB of HBM3 memory far exceeds the 890M’s system shared allocation, allowing larger datasets to reside on the accelerator.

The 890M wins in areas related to graphics output and API support. It has a full set of modern graphics APIs, including DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4, while the MI308X has no API support recorded. The 890M’s pixel rate of 92.80 GPixel/s and its 32 render output units enable actual display output, which the MI308X cannot provide. The 890M also has ray tracing capabilities with 16 ray tracing cores, a feature absent from the MI308X. The 890M’s power draw of 15 W makes it suitable for battery-powered devices, while the MI308X’s 750 W requirement and OAM module form factor restrict it to data center environments.

Benchmark data for the 890M shows its strengths in specific tests. The highest recorded score is 40,808 in the Geekbench Vulkan test, followed by 37,254 in Geekbench OpenCL. The Passmark G3D score is 8,076, while the Passmark GPU compute score is 4,157. These results indicate that the 890M performs best in compute-oriented API tests rather than legacy DirectX tests. Its Passmark DirectX 9 score is 97, the DirectX 11 score is 73, the DirectX 12 score is 37, and the DirectX 10 score is 33. The 3DMark Steel Nomad DX12 test yields a score of 590. The MI308X has no comparable test results, so its real-world performance must be inferred from its specifications.

FAQ

Q: Which GPU has more shading units?

A: The AMD Instinct MI308X has 19,456 shading units, while the AMD Radeon 890M has 1,024 shading units.

Q: Does the AMD Instinct MI308X support DirectX?

A: No, the MI308X lists no DirectX support in the database. The AMD Radeon 890M supports DirectX 12 Ultimate (12_2).

Q: What is the memory configuration of the AMD Radeon 890M?

A: The 890M uses system shared memory with a system dependent bandwidth and bus width, meaning its memory performance depends on the host system.

Q: How does the AMD Radeon 890M compare to the NVIDIA GeForce GTX 960 in average benchmark score?

A: The 890M has an average benchmark score of 9210, which is 0.7% lower than the GTX 960’s score of 9273.

Q: What is the power consumption difference between the two GPUs?

A: The MI308X has a thermal design power of 750 W, while the 890M has a thermal design power of 15 W.

Q: Does the MI308X have ray tracing cores?

A: No, the MI308X has no ray tracing cores listed. The 890M has 16 ray tracing cores.

Head-to-Head Benchmarks

There are no recorded head-to-head benchmark comparisons between the MI308X and the 890M in the database, and the MI308X has no individual benchmark scores. The comparison must therefore rely on the 890M’s measured results and the MI308X’s technical specifications. The MI308X’s FP32 output of 81.72 TFLOPS is 13.8 times the 890M’s 5.939 TFLOPS. The texture rate difference is similarly large: the MI308X delivers 2,553.6 GTexel/s, which is 13.8 times the 890M’s 185.6 GTexel/s. Memory bandwidth is where the gap becomes extreme, with the MI308X at 5.32 TB/s versus the 890M’s system dependent figure, which cannot be quantified from the available data but is inherently limited by shared system memory.

The 890M’s recorded benchmarks provide context for its performance tier. Its 3DMark Steel Nomad DX12 score of 590 reflects modern API performance. The Geekbench scores, 37,254 for OpenCL and 40,808 for Vulkan, show that the 890M handles compute-style workloads reasonably well for an integrated part. Passmark results vary by DirectX version, with DirectX 9 at 97, DirectX 11 at 73, DirectX 12 at 37, and DirectX 10 at 33. The Passmark G2D score of 979 and G3D score of 8,076 indicate that 2D performance is comparatively stronger than 3D performance. The GPU compute score of 4,157 sits between the DirectX scores and the Geekbench results, suggesting that different testing methodologies produce widely varying results for this integrated GPU.

The MI308X’s specifications indicate it should dominate in any compute workload, but the absence of benchmark data means no recorded measurement confirms this. The 890M’s nearest rivals in the database are all discrete or older integrated parts, with the AMD Radeon Vega 8 scoring 9221, 0.1% higher than the 890M, and the NVIDIA GeForce GTX 960 scoring 9273, 0.7% higher. The NVIDIA GeForce GTX 465 scores 9294, 0.9% higher, and the NVIDIA GeForce GTX 850M scores 9302, 1% higher. These narrow margins show that the 890M is competitive with a range of older graphics solutions, but it does not exceed any of them in average score.

Specification Differences

The two GPUs differ in nearly every measurable specification. The MI308X uses the CDNA 3.0 architecture with the Aqua Vanjaram chip, while the 890M uses RDNA 3.5 with the Strix Point chip. The process nodes are 5 nm for the MI308X and 4 nm for the 890M, both fabricated by TSMC. The MI308X has 153,000 million transistors on a 1017 mm² die, while the 890M has 34,000 million transistors on a 233 mm² die. Transistor density is 150.4 million per mm² for the MI308X and 145.9 million per mm² for the 890M.

Clock speeds differ, with the MI308X at 1000 MHz base and 2100 MHz boost, and the 890M at 400 MHz base and 2900 MHz boost. The MI308X has 19,456 shading units, 1,216 texture mapping units, and no render output units, while the 890M has 1,024 shading units, 64 texture mapping units, and 32 render output units. The MI308X has no ray tracing cores, while the 890M has 16. Pixel rate is 0 MPixel/s for the MI308X and 92.80 GPixel/s for the 890M. Texture rate is 2,553.6 GTexel/s for the MI308X and 185.6 GTexel/s for the 890M. FP32 and FP16 performance are both 81.72 TFLOPS for the MI308X and 5.939 TFLOPS for the 890M.

Memory configurations are fundamentally different. The MI308X has 192 GB of HBM3 memory on an 8192-bit bus with 5.32 TB/s bandwidth, while the 890M uses system shared memory with system dependent bandwidth and bus width. The MI308X’s memory clock is 1300 MHz with 5.2 Gbps effective speed, while the 890M’s memory clock is system shared. Power consumption is 750 W for the MI308X with a 1150 W suggested power supply, and 15 W for the 890M with no suggested power supply. The MI308X uses an OAM module slot width with no power connectors, while the 890M is an IGP with no power connectors. The bus interface is PCIe 5.0 x16 for the MI308X and PCIe 4.0 x8 for the 890M. Display outputs are absent on the MI308X and portable device dependent on the 890M. API support is absent for the MI308X and includes DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4 for the 890M. The MI308X was released on December 5, 2023, and the 890M on July 14, 2024. The 890M’s production status is active, while the MI308X’s status is not recorded.