AMD Instinct MI325X vs AMD Radeon 890M Comparison

The Verdict

The AMD Instinct MI325X and AMD Radeon 890M occupy entirely separate segments of the hardware spectrum, and the recorded data leaves no ambiguity about their intended roles. The MI325X is a data center accelerator with 256 GB of HBM3e memory, an 8192-bit bus, and 81.72 TFLOPS of FP32 compute. The 890M is a 15 W integrated graphics processor embedded in a mobile chip, sharing system memory and delivering 5.939 TFLOPS. The MI325X has no benchmark scores in the database, while the 890M carries a full suite of results averaging 9210 points, placing it at the 45th percentile among all GPUs. Anyone needing massive memory bandwidth and compute throughput for AI or HPC workloads should select the MI325X. Anyone building a thin-and-light laptop that needs acceptable integrated graphics for everyday use and light gaming should select the 890M. The two products do not compete; they answer different questions.

Architecture Differences

The MI325X uses the CDNA 3.0 architecture on a 5 nm TSMC process, with the Aqua Vanjaram chip. The die measures 1017 mm² and contains 153,000 million transistors, yielding a density of 150.4 million transistors per square millimeter. It packs 19,456 shading units and 1,216 texture mapping units. Notably, it has zero ROPs and a pixel rate of 0 MPixel/s, indicating that it is not designed for rasterized display output. Its texture rate reaches 2,553.6 GTexel/s. The memory subsystem uses 256 GB of HBM3e across an 8192-bit bus, providing 6.14 TB/s of bandwidth. The memory clock is 1500 MHz, which translates to 6 Gbps effective. The boost clock is 2100 MHz with a base of 1000 MHz.

The 890M uses RDNA 3.5 on a 4 nm TSMC process, with the Strix Point chip. The die is 233 mm² and contains 34,000 million transistors, with a density of 145.9 million transistors per square millimeter. It has 1,024 shading units, 64 TMUs, and 32 ROPs. It also includes 16 ray tracing cores. The pixel rate is 92.80 GPixel/s, and the texture rate is 185.6 GTexel/s. Its memory is system shared, with no dedicated VRAM, and bandwidth is system dependent. The base clock is 400 MHz, boosting to 2900 MHz. The 890M supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, whereas the MI325X reports N/A for all graphics APIs, reinforcing its compute-only orientation. The 890M is a PCIe 4.0 x8 device; the MI325X uses PCIe 5.0 x16. The MI325X is an OAM module with no display outputs and no power connectors, requiring a 1400 W suggested PSU. The 890M is an IGP with portable-device-dependent outputs and a 15 W TDP.

Head-to-Head Benchmarks

The database contains no direct head-to-head benchmark comparisons between the MI325X and the 890M. The MI325X has an empty benchmark array, an average score of 0, and a 50th percentile ranking. The 890M has ten recorded benchmark results. Its highest score comes from Geekbench Vulkan at 40,808 points, followed by Geekbench OpenCL at 37,254 points. In PassMark tests, the 890M scores 8,076 in G3D, 4,157 in GPU Compute, 979 in G2D, 97 in DirectX 9, 73 in DirectX 11, 37 in DirectX 12, and 33 in DirectX 10. In 3DMark Steel Nomad DX12, it scores 590.

The 890M's nearest rivals in the database provide context. The AMD Radeon Vega 8 averages 9,221 points, which is 0.1% higher than the 890M's 9,210 average. The NVIDIA GeForce GTX 960 averages 9,273 points, 0.7% higher. The NVIDIA GeForce GTX 465 averages 9,294 points, 0.9% higher. The NVIDIA GeForce GTX 850M averages 9,302 points, 1% higher. The 890M trails all four nearest rivals by margins under 1.1 points in percentage terms. This places it in a tight cluster of older discrete GPUs and integrated parts, confirming that its performance sits near the lower end of dedicated graphics from prior generations.

The MI325X, by contrast, has no recorded scores in the database, so no direct numeric comparison is possible. The only quantitative comparison available is architectural: the MI325X's FP32 throughput of 81.72 TFLOPS is roughly 13.8 times the 890M's 5.939 TFLOPS. The MI325X's texture rate of 2,553.6 GTexel/s is about 13.8 times the 890M's 185.6 GTexel/s. The MI325X's memory bandwidth of 6.14 TB/s is not directly comparable to the 890M's system-dependent shared memory, but the magnitude difference is stark. The MI325X has 19,456 shading units versus 1,024 on the 890M, a 19-fold difference in raw shader count.

FAQ

Q: Which GPU has higher FP32 compute performance?

A: The AMD Instinct MI325X delivers 81.72 TFLOPS of FP32, while the AMD Radeon 890M delivers 5.939 TFLOPS. The MI325X is approximately 13.8 times faster in this metric.

Q: Does the AMD Radeon 890M support ray tracing?

A: Yes. The 890M includes 16 ray tracing cores and supports DirectX 12 Ultimate (12_2). The MI325X has no listed ray tracing cores and reports N/A for DirectX support.

Q: What is the memory configuration of the MI325X?

A: The MI325X has 256 GB of HBM3e memory on an 8192-bit bus, with 6.14 TB/s of bandwidth. The memory clock is 1500 MHz, effective at 6 Gbps. The 890M uses system shared memory with system-dependent bandwidth.

Q: How does the 890M compare to its nearest rivals in average benchmark score?

A: The 890M averages 9,210 points. The AMD Radeon Vega 8 averages 9,221, the NVIDIA GeForce GTX 960 averages 9,273, the NVIDIA GeForce GTX 465 averages 9,294, and the NVIDIA GeForce GTX 850M averages 9,302. The 890M trails each by 0.1% to 1%.

Q: What is the power requirement for each product?

A: The MI325X has a TDP of 1000 W and a suggested PSU of 1400 W. The 890M has a TDP of 15 W and no suggested PSU listed.

Q: What process nodes do the two chips use?

A: The MI325X uses TSMC's 5 nm process with the Aqua Vanjaram chip. The 890M uses TSMC's 4 nm process with the Strix Point chip.

Where Each One Wins

The MI325X wins decisively in every metric related to raw compute and memory capacity. Its 256 GB HBM3e pool exceeds any integrated or consumer GPU memory configuration. The 8192-bit bus and 6.14 TB/s bandwidth support data-intensive workloads that would starve a system-shared memory part. The 19,456 shading units and 81.72 TFLOPS FP32 throughput handle large matrix operations and dense compute kernels. The 1,216 TMUs and 2,553.6 GTexel/s texture rate serve workloads requiring massive texture fetch throughput. The PCIe 5.0 x16 interface doubles the bus bandwidth available to the 890M's PCIe 4.0 x8 link. The 1000 W TDP and 1400 W suggested PSU indicate sustained, high-power operation in a server chassis. The 2024-10-09 release date places it after the 890M's 2024-07-14 launch, positioning it as a newer data center product.

The 890M wins in power efficiency, portability, and graphics feature support. Its 15 W TDP is a fraction of the MI325X's 1000 W, enabling deployment in battery-powered laptops. It has 32 ROPs and a pixel rate of 92.80 GPixel/s, allowing actual display output, while the MI325X has zero ROPs and no display outputs. The 890M includes 16 ray tracing cores, which the MI325X lacks. It supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4, while the MI325X reports N/A for all three APIs. The 890M's 2900 MHz boost clock is higher than the MI325X's 2100 MHz, reflecting its focus on latency-sensitive, interactive graphics rather than sustained compute. The 4 nm process gives it a smaller die at 233 mm² versus 1017 mm², and fewer transistors at 34,000 million versus 153,000 million.

The benchmark data shows the 890M performing near the level of older discrete GPUs. Its 8,076 PassMark G3D score and 40,808 Geekbench Vulkan score indicate usable integrated graphics performance. The 590 3DMark Steel Nomad DX12 score shows modest capability in modern DirectX 12 titles. The 890M's average score of 9,210 places it at the 45th percentile of all GPUs, just below the 50th percentile ranking of the MI325X, though the MI325X's percentile comes from an empty benchmark array and an average score of 0. The 890M's nearest rivals sit within 1% of its average score, meaning real-world differences among these parts are negligible. The MI325X has no such comparisons because no benchmark results are recorded for it.

For workloads requiring graphics output, rasterization, ray tracing, or DirectX support, the 890M is the only viable option of the two. For workloads requiring maximum memory bandwidth, FP32 throughput, or texture processing, the MI325X is the clear choice. The data does not present a scenario where both serve the same user. The MI325X is a compute accelerator without display capability, and the 890M is an integrated GPU with full graphics stack support. The 890M's production status is Active, while the MI325X's is not listed. The MI325X succeeds the Radeon Instinct line, and the 890M succeeds the Navi II IGP line. Neither product has a successor listed in the database.