AMD Instinct MI350P vs Intel Arc Graphics 128EU Mobile Comparison

Where Each One Wins

The recorded data splits these two processors into entirely separate operating domains. The AMD Instinct MI350P is a dual-slot accelerator with no display outputs, built for compute workloads that demand massive memory capacity and throughput. The Intel Arc Graphics 128EU Mobile is an integrated graphics processor (IGP) designed for portable devices, with display outputs dependent on the host laptop. The benchmark wins are not close: the MI350P dominates in raw throughput metrics, while the Intel part wins in the practical realities of mobile integration.

The MI350P shows overwhelming advantages in every throughput category that matters for data-center compute. Its FP32 output is 36.04 TFLOPS against 4.608 TFLOPS for the Intel part, a 7.8x difference. FP16 performance is identical on the AMD side at 36.04 TFLOPS with a 1:1 ratio, while Intel delivers 9.216 TFLOPS with a 2:1 ratio. Texture rate favors the AMD accelerator at 1,126.4 GTexel/s versus 144.0 GTexel/s. The Intel part does claim one win: pixel rate. It delivers 72.00 GPixel/s, while the MI350P is recorded at 0 MPixel/s, which makes sense for an accelerator with no raster output pipeline and no display outputs.

The architecture split is clear. The MI350P is a CDNA 4.0 compute accelerator with 8,192 shading units, 512 TMUs, and zero ROPs. The Intel Arc Graphics 128EU Mobile uses Xe-LPG with 1,024 shading units, 64 TMUs, and 32 ROPs. The AMD chip has no rasterization hardware, no DirectX support, no OpenGL support, no Vulkan support. The Intel part supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4. One is a server compute card, the other is a mobile graphics solution.

Architecture Differences

The manufacturing and packaging differences are substantial. The MI350P uses a 3 nm process at TSMC, with 73,000 million transistors on a 1,190 mm² die. That works out to a transistor density of 61.3M per mm². The Intel Meteor Lake part uses a 10 nm process at Intel, with transistor count and die size not recorded in the database. The MI350P is a massive dual-slot card measuring 267 mm in length, 111 mm in height, and 40 mm in width. The Intel part is an IGP with no physical dimensions listed, integrated directly into the host processor package.

Memory architecture could not be more different. The MI350P carries 144 GB of HBM3e memory on an 8,192-bit bus, delivering 8.19 TB/s of bandwidth. The Intel Arc Graphics 128EU Mobile uses system shared memory, with the bus width described as system shared and bandwidth as system dependent. The AMD part has a dedicated memory clock of 2000 MHz, which translates to 8 Gbps effective. The Intel part has no dedicated memory clock because it relies on the host system's memory.

Clock behavior also diverges. The MI350P has a base clock of 1000 MHz and a boost of 2200 MHz. The Intel part has a base of 300 MHz and a boost of 2250 MHz. Despite the similar boost figures, the AMD chip sustains much higher throughput because of its 8x shading unit count and 512 TMUs. Power envelopes are in different universes: the MI350P draws 600 W and requires a 1000 W suggested PSU with a single 16-pin connector. The Intel part consumes 28 W and has no power connector requirements because it is an IGP.

The API support gap is total. The MI350P lists N/A for DirectX, OpenGL, and Vulkan. The Intel part supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4. The MI350P uses a PCIe 5.0 x16 interface. The Intel part uses a Ring Bus. The AMD accelerator has no display outputs, while the Intel IGP's display outputs are portable device dependent. Release dates are also far apart: the Intel part arrived on 2023-12-13, while the MI350P is dated 2026-05-06. The AMD predecessor is Radeon Instinct, while the Intel predecessor is HD Graphics-M.

The Verdict

The data indicates these products should never be compared as alternatives. The AMD Instinct MI350P is a compute accelerator for server environments where 144 GB of HBM3e and 8.19 TB/s bandwidth matter more than any graphics output. The Intel Arc Graphics 128EU Mobile is a mobile integrated GPU for laptops where 28 W power consumption and system shared memory are the defining constraints. Neither part can substitute for the other.

The MI350P wins on every compute throughput metric in the database. Its FP32 performance of 36.04 TFLOPS is roughly 7.8x the Intel part's 4.608 TFLOPS. Its FP16 performance is also 36.04 TFLOPS with a 1:1 ratio, versus 9.216 TFLOPS at 2:1 for the Intel part. Texture rate is 1,126.4 GTexel/s versus 144.0 GTexel/s. The MI350P has no pixel rate because it lacks ROPs entirely, while the Intel part delivers 72.00 GPixel/s. The AMD card requires 600 W, a 1000 W suggested PSU, and a dual-slot chassis. The Intel IGP requires 28 W and fits inside a mobile processor.

The verdict from the recorded data is straightforward. The MI350P is for compute farms, AI training, and scientific workloads that can use 144 GB of HBM3e. The Intel Arc Graphics 128EU Mobile is for laptops that need DirectX 12 support, OpenGL 4.6, and Vulkan 1.4 with minimal power draw. The MI350P cannot render graphics or output video. The Intel part cannot approach the MI350P's memory bandwidth or compute throughput. The choice depends entirely on the workload, not on performance preference.

FAQ

Q: Which GPU has higher FP32 performance?

A: The AMD Instinct MI350P delivers 36.04 TFLOPS FP32, while the Intel Arc Graphics 128EU Mobile delivers 4.608 TFLOPS. The AMD part is approximately 7.8x faster in FP32.

Q: Does the AMD Instinct MI350P support DirectX or Vulkan?

A: No. The database lists N/A for DirectX, OpenGL, and Vulkan on the MI350P. The Intel Arc Graphics 128EU Mobile supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4.

Q: What memory configuration does each part use?

A: The MI350P uses 144 GB of HBM3e on an 8,192-bit bus with 8.19 TB/s bandwidth. The Intel Arc Graphics 128EU Mobile uses system shared memory with system dependent bandwidth.

Q: How do the power requirements compare?

A: The MI350P has a 600 W TDP and requires a 1000 W suggested PSU with a single 16-pin power connector. The Intel part has a 28 W TDP and needs no power connectors because it is an IGP.

Q: Which part has display outputs?

A: The MI350P has no display outputs. The Intel Arc Graphics 128EU Mobile has display outputs that are portable device dependent.

Q: When was each product released?

A: The Intel Arc Graphics 128EU Mobile has a release date of 2023-12-13. The AMD Instinct MI350P is dated 2026-05-06.

Head-to-Head Benchmarks

The head-to-head benchmark list is empty in the database, so direct measured comparisons are unavailable. However, the specification-level data provides clear performance deltas. The FP32 gap is the most telling: 36.04 TFLOPS versus 4.608 TFLOPS. That is a 7.82x advantage for the MI350P. FP16 shows a different ratio structure: the MI350P achieves 36.04 TFLOPS at 1:1, while the Intel part achieves 9.216 TFLOPS at 2:1. The AMD part's 1:1 ratio means it does not halve throughput for FP16, whereas the Intel part doubles its FP32 rate for FP16.

Texture rate favors the MI350P by a wide margin. The AMD accelerator reaches 1,126.4 GTexel/s with 512 TMUs. The Intel part reaches 144.0 GTexel/s with 64 TMUs. That is a 7.82x difference, the same ratio as the shading unit count: 8,192 versus 1,024. The pixel rate is the only category where Intel wins outright. The Intel part delivers 72.00 GPixel/s from 32 ROPs, while the MI350P is recorded at 0 MPixel/s because it has 0 ROPs. This is not a performance failure on the AMD side; it is a design choice for a compute-only accelerator.

Memory bandwidth is another dominant MI350P advantage. The AMD card offers 8.19 TB/s from HBM3e on an 8,192-bit bus. The Intel part's bandwidth is system dependent, with no fixed figure recorded. The MI350P's 144 GB capacity also dwarfs any system shared allocation a mobile IGP would typically access, though the database does not specify the Intel part's maximum shared memory. Clock speeds are similar at boost: 2200 MHz for the AMD part, 2250 MHz for the Intel part. Base clocks differ more: 1000 MHz versus 300 MHz. The AMD part sustains a higher floor, which matters for consistent compute throughput.

The process node gap is significant. The MI350P uses 3 nm at TSMC with 73,000 million transistors. The Intel part uses 10 nm at Intel with no recorded transistor count. The AMD die is enormous at 1,190 mm², consistent with a data-center accelerator. The Intel part has no listed die size because it is integrated into a larger mobile processor package. The MI350P's transistor density of 61.3M per mm² reflects the dense 3 nm process, though direct density comparison with the Intel part is impossible without its die size.

Specification Differences

The two parts differ in nearly every recorded specification field. Process node: the MI350P uses 3 nm at TSMC, while the Intel part uses 10 nm at Intel. Transistor count: 73,000 million for AMD, not recorded for Intel. Die size: 1,190 mm² for AMD, not recorded for Intel. Transistor density: 61.3M per mm² for AMD, not recorded for Intel.

Clock speeds: the MI350P has a 1000 MHz base and 2200 MHz boost. The Intel part has a 300 MHz base and 2250 MHz boost. Memory clock: 2000 MHz (8 Gbps effective) for AMD, system shared for Intel. Memory configuration: 144 GB HBM3e on 8,192-bit with 8.19 TB/s for AMD, system shared with system dependent bandwidth for Intel.

Compute units: 8,192 shading units, 512 TMUs, 0 ROPs for AMD. The Intel part has 1,024 shading units, 64 TMUs, and 32 ROPs. Pixel rate: 0 MPixel/s for AMD, 72.00 GPixel/s for Intel. Texture rate: 1,126.4 GTexel/s for AMD, 144.0 GTexel/s for Intel. FP32: 36.04 TFLOPS for AMD, 4.608 TFLOPS for Intel. FP16: 36.04 TFLOPS (1:1) for AMD, 9.216 TFLOPS (2:1) for Intel.

Power and cooling: 600 W TDP, dual-slot, 1x 16-pin power connector, 1000 W suggested PSU for AMD. The Intel part has a 28 W TDP, is an IGP, has no power connectors, and no suggested PSU. Bus interface: PCIe 5.0 x16 for AMD, Ring Bus for Intel. Display outputs: none for AMD, portable device dependent for Intel. API support: N/A for DirectX, OpenGL, and Vulkan on AMD; DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4 on Intel.

Dimensions: the MI350P measures 267 mm by 111 mm by 40 mm. The Intel part has no dimensions because it is an IGP. Release dates: 2026-05-06 for AMD, 2023-12-13 for Intel. Production status: not recorded for AMD, active for Intel. Predecessors: Radeon Instinct for AMD, HD Graphics-M for Intel. The MI350P uses the MI350 128CU chip, while the Intel part uses the Meteor Lake chip. The architectures are CDNA 4.0 for AMD and Xe-LPG for Intel.