Intel Arc Graphics 48EU Mobile

The Intel Arc Graphics 48EU Mobile is an integrated graphics processor built on the Xe-LPG architecture and part of the Meteor Lake chip family, belonging to the Arc Graphics-M generation. Intel fabricates the chip on a 10 nm process, and the part is listed as Active in production with a release date of December 13, 2023. It succeeds the HD Graphics-M line. The GPU contains 384 shading units, 24 texture mapping units, and 8 raster output units, with a base clock of 300 MHz and a boost clock of 1800 MHz. The memory subsystem is entirely System Shared, and the bus interface is a Ring Bus. The part's standing among all GPUs is the 50th percentile, with an average benchmark score of 0.

Who Should Consider It

The data positions this part at the 50th percentile of all GPUs, a median placement. That means it is neither a high-end part nor a bottom-tier one. The compute figures support this: FP32 throughput is 1,382.4 GFLOPS, texture fill is 43.20 GTexel/s, and pixel fill is 14.40 GPixel/s. These are modest numbers, and the 8 ROPs are a small count. Given that the memory is System Shared and the slot width is IGP, this is a part for portable devices where dedicated graphics memory is not an option.

For resolution and settings guidance, the data does not include game-specific benchmark scores, so exact settings recommendations cannot be derived. However, the pixel rate of 14.40 GPixel/s and the 8 ROPs suggest that the part will handle lower resolutions more comfortably than higher ones. The texture rate of 43.20 GTexel/s, combined with 24 TMUs, provides a moderate texture throughput. Users who run everyday graphics workloads at modest resolutions are the likely audience. The 50th percentile standing indicates that, in a broad comparison, this GPU performs at the median, meaning it can manage standard display output and light 3D tasks without being designed for demanding high-resolution scenarios. The boost clock of 1800 MHz is the maximum achievable under load, while the base clock of 300 MHz is the idle floor. Because the memory bandwidth is System Dependent, the actual experience will vary with the host system's memory configuration, so a device with faster system memory will extract more from this GPU than one with slower memory. The 384 shading units are the primary compute resource, and their throughput of 1,382.4 GFLOPS is the ceiling for general-purpose shader work.

Ray Tracing and Feature Set

The FACT PACK lists no dedicated ray tracing cores and no tensor cores for this part; both fields are empty. Consequently, the data does not indicate any hardware-accelerated ray tracing capability. Any ray tracing workload would have to be processed by the general-purpose shading units, of which there are 384. Similarly, there are no tensor cores listed, so AI-accelerated features that rely on dedicated tensor hardware are not indicated by the data. The architecture is Xe-LPG, which is the graphics architecture for the Meteor Lake generation, and the generation field identifies this as Arc Graphics-M (Meteor Lake).

On the API front, the part supports DirectX 12 with feature level 12_1, OpenGL 4.6, and Vulkan 1.4. The DirectX 12_1 feature level is a specific capability tier that includes features like conservative rasterization and rasterizer-ordered views. Vulkan 1.4 support is a current-generation API level, and OpenGL 4.6 is a mature specification. These API levels mean the hardware is capable of running modern graphics workloads at the driver level, even though the raw compute resources are modest. The display outputs are listed as Portable Device Dependent, meaning the available display connections depend on the portable device in which the GPU is integrated, the data does not specify which connectors are present. The absence of RT and tensor core counts is notable: the part relies entirely on its 384 shading units for all compute work, including any effects that might otherwise be accelerated by dedicated hardware.

Power and Cooling

The TDP is 28 W, which is the only power figure provided in the data. The slot width is listed as IGP, meaning this is an integrated graphics processor. No power connectors are listed, and no suggested PSU is provided, which is consistent with an integrated part that draws power through the host system rather than a dedicated connector. Because the data does not list a cooler or thermal solution, cooling is system-dependent. The 28 W TDP is modest; the data does not specify any cooling hardware, so the host device's existing thermal solution is what manages this GPU's heat output. The absence of a suggested PSU and power connector data means the power delivery requirements are handled at the system level. For a portable device, a 28 W TDP is a manageable figure, and the integrated nature of the part means no discrete power cabling is required. The base clock of 300 MHz and boost clock of 1800 MHz define the operating range, and the power draw will scale between those states, but the data does not provide a breakdown of power consumption at each clock state.

FAQ

Q: What is the TDP of the Intel Arc Graphics 48EU Mobile?

A: The TDP is 28 W.

Q: Does this GPU have dedicated ray tracing cores?

A: The data does not list any dedicated ray tracing cores; the rtCores field is empty. No tensor cores are listed either.

Q: Which graphics APIs are supported?

A: The part supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4.

Q: What type of memory does it use?

A: The memory size, type, and bus width are all listed as System Shared, and the bandwidth is described as System Dependent.

Q: When was the part released?

A: The release date is December 13, 2023.

Q: What is the clock speed range?

A: The base clock is 300 MHz and the boost clock is 1800 MHz.

Benchmark Performance

The benchmark data for this part is sparse. The benchmarks array is empty, and the average benchmark score is 0. The only comparative metric is the percentile standing of 50 among all GPUs, which places it at the median. No nearest rivals are listed, so no direct percentage deltas can be computed against other parts. In the absence of run scores, the compute throughput figures serve as the quantitative basis for assessment.

FP32 performance is 1,382.4 GFLOPS. FP16 performance is 2.765 TFLOPS, achieved at a 2:1 ratio relative to FP32, meaning the FP16 rate is exactly double the FP32 rate. This 2:1 relationship indicates the hardware can execute FP16 operations at twice the throughput of FP32 operations, which is a common pattern for packed math paths. The texture rate is 43.20 GTexel/s, corresponding to the 24 TMUs, and the pixel rate is 14.40 GPixel/s, corresponding to the 8 ROPs. These fill rates are internally consistent with the unit counts. The 8 ROPs are a low number, which limits pixel output and directly impacts how many pixels can be written per second, a constraint that becomes more relevant at higher resolutions where more pixels must be processed per frame.

Given the 50th percentile standing, the data suggests this part performs in the middle of the field among all GPUs. The average benchmark score of 0 indicates that no aggregate score has been recorded, so the percentile is the primary ranking signal. Without rival delta values, the analysis cannot state how far ahead or behind any specific competitor this part is. The compute throughput figures, however, give a clear sense of the part's capability ceiling: 1,382.4 GFLOPS of FP32 work, 43.20 GTexel/s of texturing, and 14.40 GPixel/s of rasterization. These are the numbers that define what the GPU can do per second, and they are the only quantitative performance indicators available in the data.

Memory Subsystem

The memory subsystem is fully System Shared. The size, type, and bus width are all listed as System Shared, and the bandwidth is listed as System Dependent. This means the GPU does not have a dedicated VRAM pool; it uses the host system's memory, and the available bandwidth depends on the system's memory configuration. The data does not specify a dedicated bus width or a bandwidth figure, so the memory performance cannot be quantified independently of the host system.

For high-resolution workloads, this has implications. Because the memory bandwidth is system dependent, the achievable bandwidth will vary from one host device to another. A system with a wider memory bus or faster memory will provide more bandwidth to this GPU, while a system with slower memory will bottleneck it. The pixel rate of 14.40 GPixel/s and the 8 ROPs limit how many pixels can be written per second, which is a constraint at higher resolutions where the pixel count per frame increases. The system-shared memory also means the CPU and GPU draw from the same memory pool, which can create contention in memory-intensive scenarios. The bus interface is a Ring Bus, which is the interconnect used by this integrated part to communicate with the rest of the system.

The absence of a dedicated VRAM figure means that any resolution-based assessment must account for the system-dependent nature of the memory. The data does not provide a fixed bandwidth number, so the memory performance cannot be quantified independently of the host system. What the data does show is that the memory subsystem is not a fixed specification, it scales with the host platform. This makes the GPU's real-world performance inherently variable across different portable devices, and it is a key factor to weigh when considering this part for any workload that is sensitive to memory bandwidth.