AMD Instinct MI308X vs NVIDIA GeForce RTX 5090 SE Comparison

Head-to-Head Benchmarks

The recorded data contains no benchmark scores for either the AMD Instinct MI308X or the NVIDIA GeForce RTX 5090 SE. Both entries show an average benchmark score of zero, with no individual test results listed in the database. Consequently, the head-to-head comparison relies entirely on the theoretical specifications and architectural capabilities recorded for each accelerator.

The AMD Instinct MI308X delivers a peak FP32 throughput of 81.72 TFLOPS, which places it ahead of the RTX 5090 SE's 66.94 TFLOPS by roughly 22 percent in raw single-precision compute. The same ratio applies to FP16 performance, where both cards operate at a 1:1 ratio with their FP32 figures. This indicates the AMD part offers a higher ceiling for dense compute workloads that scale with raw floating-point throughput.

The NVIDIA card counters with a significantly higher pixel rate of 380.3 GPixel/s, while the AMD accelerator records 0 MPixel/s, reflecting its lack of traditional raster output processors. The RTX 5090 SE also posts a texture rate of 1,045.9 GTexel/s, whereas the MI308X reaches 2,553.6 GTexel/s, more than double the NVIDIA figure. These numbers point to divergent design priorities: the AMD part is built for massive texture and compute throughput, while the NVIDIA card retains conventional graphics pipeline functionality.

Neither component has a recorded percentile rank above 50, and both sit at the median of the database's GPU population. The absence of actual benchmark submissions means the percentile values are placeholders rather than derived from measured performance. The wins counter shows zero for both sides, confirming no test data exists to establish a definitive performance winner in any workload category.

Architecture Differences

The two accelerators stem from different architectural lineages. AMD's Instinct MI308X uses the CDNA 3.0 architecture with the Aqua Vanjaram chip, while NVIDIA's GeForce RTX 5090 SE employs Blackwell 2.0 with the GB202 die. Both are fabricated on a 5 nm process at TSMC, but the transistor counts diverge sharply. The AMD chip integrates 153,000 million transistors on a 1017 mm² die, yielding a density of 150.4 million transistors per square millimeter. The NVIDIA die holds 92,200 million transistors on 750 mm², for a density of 122.9 million per square millimeter. The MI308X therefore packs roughly 66 percent more transistors into a die that is about 36 percent larger by area.

Memory architecture presents the most pronounced structural difference. The MI308X carries 192 GB of HBM3 across an 8192-bit bus, producing 5.32 TB/s of bandwidth. The RTX 5090 SE uses 24 GB of GDDR7 on a 384-bit bus, delivering 1.34 TB/s. The AMD card offers eight times the capacity and just under four times the bandwidth, a configuration suited to holding very large datasets on-device. The NVIDIA card's 1.34 TB/s is still a high-bandwidth design, but it relies on far smaller capacity and a narrower interface.

Compute unit organization also differs. The MI308X records 19,456 shading units and 1,216 texture mapping units, with no raster output units listed. The RTX 5090 SE has 14,080 shading units, 440 texture units, and 160 ROPs. The NVIDIA part additionally includes 110 ray tracing cores and 440 tensor cores, while the AMD card lists no dedicated RT or tensor core counts in the database. Clock behavior shows the NVIDIA chip running at a 1740 MHz base and 2377 MHz boost, compared to 1000 MHz base and 2100 MHz boost for the AMD part. Despite the lower clocks, the MI308X achieves higher aggregate FP32 throughput due to its larger execution resource pool.

Power and physical design also diverge. The MI308X is rated at 750 W TDP with a suggested PSU of 1150 W, mounted as an OAM module with no power connectors and no display outputs. The RTX 5090 SE consumes 500 W, suggests a 900 W PSU, fits a dual-slot form factor with a single 16-pin connector, and provides one HDMI 2.1b output plus three DisplayPort 2.1b outputs. The AMD part targets server racks with no video output, while the NVIDIA card is a conventional add-in board with full display support. API support differs completely: the MI308X lists no DirectX, OpenGL, or Vulkan support, whereas the RTX 5090 SE supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

FAQ

Q: Which accelerator has higher peak FP32 performance?

A: The AMD Instinct MI308X records 81.72 TFLOPS FP32, while the NVIDIA GeForce RTX 5090 SE records 66.94 TFLOPS. The AMD part leads by approximately 22 percent.

Q: How much memory does each card provide?

A: The MI308X includes 192 GB of HBM3 on an 8192-bit bus with 5.32 TB/s bandwidth. The RTX 5090 SE includes 24 GB of GDDR7 on a 384-bit bus with 1.34 TB/s bandwidth.

Q: Does the AMD card support graphics APIs like DirectX or Vulkan?

A: No. The database lists DirectX, OpenGL, and Vulkan support as N/A for the MI308X. The RTX 5090 SE supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Q: What are the power requirements for each?

A: The MI308X has a 750 W TDP and suggests a 1150 W PSU. The RTX 5090 SE has a 500 W TDP and suggests a 900 W PSU.

Q: Are there any benchmark results available for these two cards?

A: No. Both entries show an average benchmark score of zero and no individual benchmark records, so no measured performance comparison can be made from the database.

Q: Which card has ray tracing hardware?

A: The RTX 5090 SE lists 110 ray tracing cores. The MI308X does not list any RT core count in the database.

Specification Differences

The following fields differ between the two accelerators according to the database records.

The MI308X uses 153,000 million transistors on a 1017 mm² die with a density of 150.4 million per square millimeter. The RTX 5090 SE uses 92,200 million transistors on a 750 mm² die with a density of 122.9 million per square millimeter.

Base clocks are 1000 MHz for the AMD part and 1740 MHz for the NVIDIA part. Boost clocks are 2100 MHz and 2377 MHz, respectively. Memory clocks differ as 1300 MHz (5.2 Gbps effective) for the MI308X and 1750 MHz (28 Gbps effective) for the RTX 5090 SE.

Memory capacity, type, bus width, and bandwidth all differ: 192 GB HBM3, 8192-bit, 5.32 TB/s versus 24 GB GDDR7, 384-bit, 1.34 TB/s.

Shading units are 19,456 versus 14,080. Texture units are 1,216 versus 440. The MI308X has no ROPs, while the RTX 5090 SE has 160. The NVIDIA card has 110 RT cores and 440 tensor cores; the AMD card has no such entries.

Pixel rate is 0 MPixel/s for the MI308X and 380.3 GPixel/s for the RTX 5090 SE. Texture rate is 2,553.6 GTexel/s versus 1,045.9 GTexel/s. FP32 and FP16 are 81.72 TFLOPS versus 66.94 TFLOPS for both precision modes.

TDP is 750 W versus 500 W. The MI308X is an OAM Module with no power connectors, while the RTX 5090 SE is dual-slot with one 16-pin connector. Suggested PSU is 1150 W versus 900 W.

Display outputs: none for the AMD part, while the NVIDIA card has 1x HDMI 2.1b and 3x DisplayPort 2.1b. API support is entirely absent for the MI308X, while the RTX 5090 SE supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Physical dimensions are recorded only for the NVIDIA card: 267 mm length, 111 mm height, 40 mm width. The AMD card has no recorded dimensions.

Release dates differ: the MI308X was released on 2023-12-05, while the RTX 5090 SE is dated 2025-12-31. The NVIDIA card has an active production status, while the AMD card's production status is not recorded. The RTX 5090 SE has a launch MSRP of 1,499 USD. The AMD card has no launch MSRP recorded.

Where Each One Wins

The AMD Instinct MI308X is the stronger choice for compute-heavy, data-intensive workloads that fit within a server context. Its 192 GB HBM3 pool and 5.32 TB/s bandwidth allow it to hold and feed far larger working sets than the 24 GB GDDR7 configuration of the RTX 5090 SE. The 81.72 TFLOPS FP32 and FP16 figures exceed the NVIDIA card's 66.94 TFLOPS in both precisions, giving the AMD part a measurable edge in dense linear algebra, scientific simulation, and machine learning training tasks that rely on raw floating-point throughput. The 2,553.6 GTexel/s texture rate also points to strong performance in texture-bound compute kernels, even though the card lacks conventional graphics output.

The NVIDIA GeForce RTX 5090 SE wins in every scenario that requires a traditional graphics pipeline. It is the only one of the two with ROPs, ray tracing cores, tensor cores, and full DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4 support. Its 380.3 GPixel/s pixel rate confirms it can drive displays and handle rasterization work, which the MI308X cannot do at all given its lack of display outputs and zero pixel rate. The 24 GB GDDR7 memory is smaller but still substantial for consumer or workstation graphics, and the 500 W TDP with a dual-slot design makes it far easier to install in a standard PC chassis. The card's display outputs, including HDMI 2.1b and three DisplayPort 2.1b connections, make it suitable for multi-monitor setups.

The use-case split is therefore clean. The MI308X belongs in accelerator racks where compute density and memory capacity matter more than graphics output, power efficiency, or software ecosystem compatibility with consumer APIs. The RTX 5090 SE belongs in systems that need both strong compute and full graphics functionality, including ray tracing and tensor workloads, with the flexibility of conventional PCIe installation and display connectivity. The lack of benchmark data means these conclusions rest on specification analysis rather than measured performance, but the architectural records point to two fundamentally different products aimed at different segments of the market.