AMD Instinct MI300A vs Intel Arc Pro B390 Comparison

The Verdict

The database records two fundamentally different compute devices, and the verdict follows directly from their physical and architectural profiles. The AMD Instinct MI300A is an accelerator-class module with a 750 W power envelope, 128 GB of HBM3 memory, and 61.29 TFLOPS of FP32 throughput. The Intel Arc Pro B390 is an integrated graphics processor with an 80 W power envelope, system-shared memory, and 7.680 TFLOPS of FP32 performance. The MI300A targets high-throughput compute environments where massive memory bandwidth and raw parallel throughput are the priority. The B390 targets lightweight, power-constrained mobile or embedded systems where graphics API support and display output matter more than raw compute density.

Benchmark results in the database show no head-to-head wins for either part, and both sit at the 50th percentile among all GPUs with an average benchmark score of zero. The absence of recorded benchmarks means the analysis must rely on specification differences rather than measured performance deltas. However, the specification gap is so wide that the intended use cases are unambiguous. The MI300A is a server or workstation accelerator for dense numerical workloads. The B390 is an integrated solution for portable devices that need modern graphics features without a discrete card.

Architecture Differences

The MI300A uses the CDNA 3.0 architecture on a 5 nm TSMC process, with the chip carrying 153,000 million transistors on a 1017 mm² die. That yields a transistor density of 150.4 million per square millimeter. The B390 uses the Xe3-LPG architecture on Intel's 3 nm process, with transistor count and die size recorded as unknown. The MI300A is built for compute, not display: it has no display outputs, no DirectX, OpenGL, or Vulkan API support, and its pixel rate is listed as zero. The B390, in contrast, supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, making it a fully featured graphics processor.

The MI300A's memory subsystem is a dedicated 128 GB pool of HBM3 across an 8192-bit bus, delivering 5.32 TB/s of bandwidth. The B390 uses system-shared memory with a system-dependent bandwidth, which means its performance scales with the host platform's memory configuration. The MI300A has 14,592 shading units, 912 texture mapping units, and no ray tracing cores. The B390 has 1,536 shading units, 48 TMUs, 24 ROPs, and 12 ray tracing cores. The architectural priorities are clear: the MI300A maximizes parallel compute resources, while the B390 includes fixed-function graphics hardware for rendering workloads.

Clock behavior also differs sharply. The MI300A runs at a 1000 MHz base and 2100 MHz boost, with memory clocked at 1300 MHz (5.2 Gbps effective). The B390 has a 300 MHz base and a 2500 MHz boost. The B390's higher boost clock reflects its integrated nature and power budget, while the MI300A's lower boost clock still produces far higher throughput because of its massive shader count.

Head-to-Head Benchmarks

The database contains no recorded head-to-head benchmarks between the MI300A and the B390. Wins for each part are zero, and the benchmark arrays are empty. Consequently, the comparison must be drawn from the recorded specification data, which is extensive and decisive in several categories.

In raw FP32 compute, the MI300A delivers 61.29 TFLOPS, which is approximately 8 times the B390's 7.680 TFLOPS. That ratio is the single largest performance gap between the two parts. The MI300A's texture rate of 1,915.2 GTexel/s is roughly 16 times the B390's 120.0 GTexel/s, reflecting the 912 TMUs versus 48 TMUs. The MI300A's pixel rate is recorded as zero because it is not a rasterization device; the B390 produces 60.00 GPixel/s from its 24 ROPs.

Memory bandwidth shows an even starker contrast. The MI300A's 5.32 TB/s is a fixed hardware capability, whereas the B390's bandwidth is system-dependent, meaning no fixed number can be compared directly. The MI300A's 128 GB of dedicated HBM3 memory is a monolithic pool for compute kernels, while the B390 must share system memory with the host CPU and other components. For workloads that fit within the MI300A's memory, the bandwidth advantage alone can dominate execution time.

The B390 counters with features the MI300A lacks entirely: ray tracing cores and a full graphics API stack. The 12 RT cores and DirectX 12 Ultimate support make the B390 suitable for hardware-accelerated ray tracing in supported applications. The MI300A has no RT cores and no graphics APIs, so it cannot perform these tasks at all. The B390 also has a much higher boost clock at 2500 MHz versus 2100 MHz, though this does not compensate for the shader count disparity in compute workloads.

FAQ

Q: Which device has higher FP32 compute throughput?

A: The AMD Instinct MI300A delivers 61.29 TFLOPS of FP32 performance, which is approximately 8 times the Intel Arc Pro B390's 7.680 TFLOPS.

Q: Does either device support ray tracing?

A: Only the Intel Arc Pro B390 includes ray tracing hardware, with 12 RT cores. The AMD Instinct MI300A has no ray tracing cores recorded.

Q: How do the memory systems differ?

A: The MI300A has 128 GB of dedicated HBM3 memory on an 8192-bit bus with 5.32 TB/s bandwidth. The B390 uses system-shared memory with system-dependent bandwidth and bus width.

Q: What graphics APIs does each device support?

A: The B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The MI300A has no graphics API support recorded, with DirectX, OpenGL, and Vulkan all listed as N/A.

Q: Which device has a higher boost clock?

A: The Intel Arc Pro B390 boosts to 2500 MHz, while the AMD Instinct MI300A boosts to 2100 MHz.

Q: What is the power envelope difference?

A: The MI300A has a TDP of 750 W and a suggested PSU of 1150 W. The B390 has a TDP of 80 W and no suggested PSU recorded.

Where Each One Wins

The AMD Instinct MI300A wins in every category that measures raw parallel compute throughput. Its 61.29 TFLOPS FP32 figure, 1,915.2 GTexel/s texture rate, and 5.32 TB/s memory bandwidth place it in a different performance class than the B390. The 14,592 shading units versus 1,536 means the MI300A can process far more data per clock cycle. Its 128 GB HBM3 pool allows large models and datasets to reside entirely in fast memory, avoiding the bandwidth penalties of system-shared memory. The MI300A is the clear choice for dense compute workloads such as scientific simulation, machine learning training, and data analytics that require massive memory capacity and bandwidth. Its PCIe 5.0 x16 interface and OAM module form factor indicate a server-oriented design that assumes a host system with substantial power delivery.

The Intel Arc Pro B390 wins in portability, power efficiency, and graphics feature support. Its 80 W TDP is less than 11% of the MI300A's 750 W, and its IGP slot width means it requires no additional power connectors and no discrete card installation. The B390's 12 RT cores and full DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4 support make it functional for graphics rendering, ray tracing, and general-purpose GPU compute within a mobile or embedded context. The 60.00 GPixel/s pixel rate confirms it can drive display output, though the actual outputs are portable-device dependent. The B390's 2500 MHz boost clock is higher than the MI300A's 2100 MHz, which helps in latency-sensitive tasks that rely on single-core throughput.

The two devices do not compete for the same workloads. The MI300A is a compute accelerator with no display path, while the B390 is an integrated graphics processor with no dedicated memory. The B390's 15.36 TFLOPS FP16 (2:1) performance shows it can handle half-precision workloads, but even that figure is lower than the MI300A's FP32 output. The MI300A has no recorded FP16 figure, so no direct comparison is possible there.

Specification Differences

The two devices differ across nearly every recorded specification field.

Process and Foundry: The MI300A uses a 5 nm process at TSMC, while the B390 uses a 3 nm process at Intel. The MI300A has 153,000 million transistors on a 1017 mm² die; the B390's transistor count and die size are unknown.

Clocks: The MI300A has a 1000 MHz base and 2100 MHz boost, with memory at 1300 MHz (5.2 Gbps effective). The B390 has a 300 MHz base and 2500 MHz boost, with system-shared memory.

Memory: The MI300A has 128 GB HBM3, 8192-bit bus, 5.32 TB/s bandwidth. The B390 uses system-shared memory with system-dependent bandwidth and bus width.

Compute Units: The MI300A has 14,592 shading units, 912 TMUs, and 0 ROPs. The B390 has 1,536 shading units, 48 TMUs, and 24 ROPs. The MI300A has no RT cores; the B390 has 12.

Throughput: The MI300A produces 61.29 TFLOPS FP32 and 1,915.2 GTexel/s with 0 MPixel/s pixel rate. The B390 produces 7.680 TFLOPS FP32, 15.36 TFLOPS FP16 (2:1), 120.0 GTexel/s, and 60.00 GPixel/s.

Power and Form Factor: The MI300A has a 750 W TDP, OAM Module slot width, no power connectors, and a 1150 W suggested PSU. The B390 has an 80 W TDP, IGP slot width, no power connectors, and no suggested PSU.

Interface: The MI300A uses PCIe 5.0 x16; the B390 uses IGP.

Display and APIs: The MI300A has no display outputs and no DirectX, OpenGL, or Vulkan support. The B390 has portable-device-dependent display outputs and supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Production and Release: The MI300A has no recorded production status and was released on 2023-12-05. The B390 is marked as active production and was released on 2026-01-26. The MI300A's predecessor is Radeon Instinct; the B390's is HD Graphics-WM. Neither has a recorded successor or launch MSRP.