AMD Instinct MI350P vs Intel Arc B390 Comparison

The Verdict

The AMD Instinct MI350P and Intel Arc B390 occupy entirely separate segments of the hardware spectrum, and the recorded data confirms they are not direct competitors. The MI350P is a dual-slot, 600 W accelerator with no display outputs, built for compute workloads, while the Arc B390 is an integrated processor graphics solution with a 80 W TDP and active production status. The database shows zero head-to-head benchmark entries between them, and their benchmark availability differs entirely: the MI350P has no recorded benchmark scores, while the Arc B390 has a single 3DMark Steel Nomad DX12 score of 1482.

The MI350P sits at the 50th percentile among all GPUs in the database, with an average benchmark score of zero due to the absence of recorded tests. The Arc B390 sits at the 9th percentile with an average score of 1482, placing it near the bottom of the performance distribution. The Arc B390's nearest rivals are all low-end NVIDIA parts: the GeForce GT 520MX (1463, 1.3% slower), GeForce 800M (1460, 1.5% slower), GeForce GT 625 OEM (1446, 2.5% slower), and GeForce GT 710 (1443, 2.7% slower). The MI350P has no nearest rivals listed.

Data indicates the MI350P is for server and datacenter compute roles, while the Arc B390 is for portable devices. The MI350P uses 8192 shading units, 512 TMUs, and zero ROPs, confirming its compute-focused design. The Arc B390 uses 1536 shading units, 48 TMUs, and 24 ROPs, with 12 ray tracing cores. The MI350P delivers 36.04 TFLOPS FP32 and 36.04 TFLOPS FP16 at a 1:1 ratio; the Arc B390 delivers 7.680 TFLOPS FP32 and 15.36 TFLOPS FP16 at a 2:1 ratio. The MI350P's texture rate is 1,126.4 GTexel/s versus 120.0 GTexel/s for the Arc B390, a 9.4x difference.

Where Each One Wins

The MI350P wins decisively in raw compute throughput. Its FP32 output of 36.04 TFLOPS is 4.7x higher than the Arc B390's 7.680 TFLOPS. The FP16 comparison shows the MI350P at 36.04 TFLOPS versus the Arc B390's 15.36 TFLOPS, a 2.3x advantage. The texture rate of 1,126.4 GTexel/s versus 120.0 GTexel/s represents a 9.4x gap in fill-rate capability. The MI350P's memory configuration is entirely different: 144 GB of HBM3e on an 8192-bit bus with 8.19 TB/s bandwidth, whereas the Arc B390 uses system shared memory with system dependent bandwidth.

The Arc B390 wins in features relevant to consumer and portable use. It supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, while the MI350P lists N/A for all three APIs. The Arc B390 has display outputs (portable device dependent), while the MI350P has no outputs. The Arc B390 includes 12 ray tracing cores and 24 ROPs, both absent from the MI350P which has zero ROPs. The Arc B390's pixel rate is 60.00 GPixel/s; the MI350P's pixel rate is 0 MPixel/s.

The Arc B390 also wins on power efficiency and integration. Its 80 W TDP is 7.5x lower than the MI350P's 600 W TDP. The Arc B390 requires no power connectors and no suggested PSU rating, while the MI350P needs a single 16-pin connector and a 1000 W suggested PSU. The Arc B390 is an IGP with no slot width, while the MI350P is dual-slot and measures 267 mm in length, 111 mm in height, and 40 mm in width.

Architecture Differences

The MI350P uses the CDNA 4.0 architecture with the MI350 128CU chip, manufactured on a 3 nm process at TSMC. The Arc B390 uses the Xe3-LPG architecture with the Panther Lake chip, also on a 3 nm process but fabricated by Intel. The MI350P belongs to the Instinct (MIx) generation, while the Arc B390 belongs to Arc Graphics-M (Panther Lake).

The MI350P integrates 73,000 million transistors on a 1190 mm² die, yielding a transistor density of 61.3M per mm². The Arc B390's transistor count, die size, and density are all listed as unknown in the database. The MI350P's clock configuration shows a 1000 MHz base and 2200 MHz boost, with memory at 2000 MHz (8 Gbps effective). The Arc B390 has a 300 MHz base and 2500 MHz boost, with system shared memory.

The shading unit counts differ by a factor of 5.3: 8192 for the MI350P versus 1536 for the Arc B390. TMU counts differ by 10.7x: 512 versus 48. The MI350P has zero ROPs, while the Arc B390 has 24. The MI350P has no listed ray tracing cores or tensor cores; the Arc B390 lists 12 ray tracing cores but no tensor cores. The MI350P's FP16 performance is 1:1 with FP32, indicating a compute-oriented design that does not accelerate half-precision separately; the Arc B390's FP16 is 2:1, doubling throughput over FP32.

The memory subsystems are fundamentally different. The MI350P uses dedicated HBM3e with 144 GB capacity, an 8192-bit bus, and 8.19 TB/s bandwidth. The Arc B390 uses system shared memory, meaning capacity, bus width, and bandwidth are system dependent. The MI350P connects via PCIe 5.0 x16; the Arc B390 uses an IGP bus interface. The MI350P has no API support listed, while the Arc B390 supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4.

The release dates differ by about three months: the Arc B390 was released on 2026-01-26, and the MI350P on 2026-05-06. The MI350P's predecessor is listed as Radeon Instinct; the Arc B390 has no predecessor listed. Neither has a successor listed.

FAQ

Q: How do the FP32 performance figures compare between the two cards?

A: The MI350P delivers 36.04 TFLOPS FP32, which is 4.7x higher than the Arc B390's 7.680 TFLOPS FP32.

Q: Which card has more memory bandwidth?

A: The MI350P has 8.19 TB/s bandwidth from 144 GB of HBM3e on an 8192-bit bus. The Arc B390 uses system shared memory with system dependent bandwidth.

Q: Does the Arc B390 support ray tracing?

A: Yes, the Arc B390 includes 12 ray tracing cores. The MI350P has no ray tracing cores listed and zero ROPs.

Q: What are the power requirements for each card?

A: The MI350P has a 600 W TDP and requires a single 16-pin power connector with a 1000 W suggested PSU. The Arc B390 has an 80 W TDP, requires no power connectors, and has no suggested PSU rating.

Q: Which card supports DirectX 12 Ultimate?

A: The Arc B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The MI350P lists N/A for all three APIs.

Q: How does the Arc B390 compare to its nearest rivals?

A: The Arc B390 scores 1482 in 3DMark Steel Nomad DX12, which is 1.3% higher than the GeForce GT 520MX (1463), 1.5% higher than the GeForce 800M (1460), 2.5% higher than the GeForce GT 625 OEM (1446), and 2.7% higher than the GeForce GT 710 (1443).

Head-to-Head Benchmarks

The database contains no head-to-head benchmark entries between the MI350P and the Arc B390. Wins are recorded as zero for both sides. This absence of direct comparison data reflects the fundamental positioning of each product. The MI350P has no benchmark scores at all, while the Arc B390 has exactly one recorded test.

The Arc B390's single benchmark result is 3DMark Steel Nomad DX12 with a score of 1482. This places it at the 9th percentile among all GPUs in the database. Its nearest rivals, all NVIDIA parts, are within a narrow band: the GeForce GT 520MX trails by 1.3%, the GeForce 800M by 1.5%, the GeForce GT 625 OEM by 2.5%, and the GeForce GT 710 by 2.7%. The data shows the Arc B390 edges out these low-end discrete GPUs by small margins.

The MI350P's percentile ranking of 50th is based on an average benchmark score of zero, indicating the database has not yet recorded any performance tests for this accelerator. Without benchmark data, the percentile reflects positional information only, not measured performance.

The compute capability comparison can be made from recorded specifications. The MI350P's texture rate of 1,126.4 GTexel/s is 9.4x the Arc B390's 120.0 GTexel/s. The MI350P's FP32 throughput of 36.04 TFLOPS is 4.7x the Arc B390's 7.680 TFLOPS. The MI350P's FP16 throughput of 36.04 TFLOPS is 2.3x the Arc B390's 15.36 TFLOPS. The MI350P's memory bandwidth of 8.19 TB/s is not directly comparable to the Arc B390's system dependent figure, but the dedicated HBM3e implementation with an 8192-bit bus represents a substantially larger memory subsystem.

The Arc B390's pixel rate of 60.00 GPixel/s contrasts with the MI350P's 0 MPixel/s, reflecting the MI350P's lack of ROPs and display-oriented output. The Arc B390's 12 ray tracing cores and 24 ROPs enable graphics features that the MI350P does not implement. The MI350P's zero ROP count and N/A API listings confirm it is not designed for rasterized graphics workloads.

Clock behavior differs as well. The MI350P has a 1000 MHz base and 2200 MHz boost. The Arc B390 has a 300 MHz base and 2500 MHz boost, a 2.2x higher boost clock but a 3.3x lower base clock. The Arc B390's boost clock exceeds the MI350P's by 300 MHz, though the MI350P's massively larger shading unit count dominates throughput calculations.

The MI350P's physical design is a dual-slot card at 267 mm length, 111 mm height, and 40 mm width. The Arc B390 is an IGP with no dimensions listed. The MI350P uses a PCIe 5.0 x16 interface; the Arc B390 uses an IGP bus interface. The MI350P has no display outputs; the Arc B390's display outputs are portable device dependent.

Manufacturing details show the MI350P uses 73,000 million transistors on a 1190 mm² die with a 61.3M per mm² density. The Arc B390's transistor count, die size, and density are unknown. Both use 3 nm process nodes, but the MI350P is fabricated at TSMC while the Arc B390 is fabricated at Intel. The MI350P's CDNA 4.0 architecture is compute-oriented; the Arc B390's Xe3-LPG architecture is graphics-oriented for integrated use in Panther Lake mobile processors.

The MI350P's FP16 ratio of 1:1 indicates no dedicated half-precision path, while the Arc B390's 2:1 ratio doubles FP16 throughput relative to FP32. This makes the Arc B390 relatively stronger in workloads that use FP16, despite its lower absolute numbers. However, the MI350P's absolute FP16 output of 36.04 TFLOPS remains 2.3x higher than the Arc B390's 15.36 TFLOPS.

The Arc B390's nearest rival comparison shows it performs at a level consistent with older low-end NVIDIA mobile and OEM parts. The 1.3% to 2.7% margins over the GT 520MX, GeForce 800M, GT 625 OEM, and GT 710 are small but consistent, placing the Arc B390 marginally ahead of each. The MI350P has no such comparisons in the database.

The data supports a clear separation: the MI350P is a high-throughput compute accelerator with massive memory capacity and bandwidth, while the Arc B390 is a low-power integrated graphics solution with modern API support and ray tracing capability. Their respective release dates, with the Arc B390 in January 2026 and the MI350P in May 2026, further establish their distinct market roles.