AMD Instinct MI300X vs Intel Arc B370 Comparison

Where Each One Wins

The recorded data presents two devices with fundamentally different roles. The AMD Instinct MI300X is an accelerator oriented toward compute workloads, while the Intel Arc B370 is an integrated graphics processor designed for portable devices. Each product wins in its own domain, and the benchmark results reflect that separation.

The MI300X delivers a Geekbench OpenCL score of 317,994. That places it at the 100th percentile among all GPUs in the database. Its nearest rivals in that test are the NVIDIA B200 at 345,482, the NVIDIA H200 NVL at 334,891, the NVIDIA L40S at 295,763, and the NVIDIA RTX 6000 Ada Generation at 287,237. Against those, the MI300X trails the B200 by 8% and the H200 NVL by 5%, while it leads the L40S by 7.5% and the RTX 6000 Ada by 10.7%. That result shows a compute part competing directly with the top data center accelerators from NVIDIA, sitting in the same performance neighborhood as those parts.

The Arc B370, in contrast, posts a 3DMark Steel Nomad DX12 score of 1,184. That lands at the 5th percentile among all GPUs. Its nearest rivals are older discrete parts: the ATI Mobility Radeon HD 5570 at 1,186, the ATI Radeon HD 5770 at 1,190, the AMD Radeon HD 7650M at 1,192, and the AMD FirePro M2000 at 1,168. The Arc B370 trails the HD 5570 by 0.2%, the HD 5770 by 0.5%, and the HD 7650M by 0.7%, while it leads the FirePro M2000 by 1.4%. Those margins are tiny, which indicates the Arc B370 delivers integrated-level graphics performance comparable to modest discrete GPUs from over a decade ago.

The wins are therefore domain-specific. The MI300X wins decisively in raw compute throughput, as its OpenCL result demonstrates. The Arc B370 wins in the integrated graphics space, where its low power envelope and portable-device focus matter more than absolute performance. Neither part is a substitute for the other, and the benchmark gap is not a measure of quality but of intended use.

Architecture Differences

The two chips come from different design philosophies and manufacturing processes. The MI300X uses the Aqua Vanjaram chip with the CDNA 3.0 architecture, built on a 5 nm process at TSMC. It packs 153,000 million transistors onto a 1017 mm² die, giving a transistor density of 150.4 million per square millimeter. The Arc B370 uses the Panther Lake chip with the Xe3-LPG architecture, built on a 3 nm process at Intel. Its transistor count and die size are not recorded in the database, so no direct density comparison is possible.

The MI300X is a discrete accelerator with a PCIe 5.0 x16 bus interface and an OAM module slot width. It has no display outputs and no power connectors, requiring an external power delivery system. The suggested PSU is 1150 W, and the thermal design power is 750 W. The Arc B370 is an integrated graphics processor with an IGP bus interface and slot width. It has no power connectors either, but its thermal design power is just 25 W, and it relies on the host system's power delivery. Display outputs are listed as portable device dependent.

Memory configurations differ sharply. The MI300X uses 192 GB of HBM3 memory on an 8192-bit bus, delivering 5.32 TB/s of bandwidth. Its memory clock is 1300 MHz, which translates to 5.2 Gbps effective. The Arc B370 uses system shared memory, with size, type, bus width, and bandwidth all listed as system dependent. That means the Arc B370's memory performance varies with the host platform, while the MI300X has dedicated, fixed memory resources.

Compute resources also diverge. The MI300X has 19,456 shading units, 1,216 texture mapping units, and no ROPs, reflecting its compute-first design. Its texture rate is 2,553.6 GTexel/s, and its pixel rate is 0 MPixel/s because it has no ROPs. The Arc B370 has 1,280 shading units, 40 TMUs, 20 ROPs, and 10 ray tracing cores. Its pixel rate is 48.00 GPixel/s, and its texture rate is 96.00 GTexel/s. The Arc B370 also supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, while the MI300X lists N/A for all graphics APIs, confirming its lack of a graphics pipeline.

Clock behavior differs as well. The MI300X has a base clock of 1000 MHz and a boost clock of 2100 MHz. The Arc B370 has a base clock of 300 MHz and a boost clock of 2400 MHz. The Arc B370's higher boost clock does not compensate for its far fewer shading units, but it does indicate a design tuned for burst workloads in power-constrained portable devices.

FAQ

Q: Which device has the higher raw compute performance?

A: The AMD Instinct MI300X delivers 81.72 TFLOPS of FP32 performance, while the Intel Arc B370 delivers 6.144 TFLOPS of FP32. The MI300X also delivers 81.72 TFLOPS of FP16 at a 1:1 ratio, whereas the Arc B370 delivers 12.29 TFLOPS of FP16 at a 2:1 ratio.

Q: How do their benchmark scores compare?

A: The MI300X scores 317,994 in Geekbench OpenCL, placing at the 100th percentile. The Arc B370 scores 1,184 in 3DMark Steel Nomad DX12, placing at the 5th percentile. These are different tests for different workloads, so direct comparison is not meaningful.

Q: What memory configurations do they use?

A: The MI300X uses 192 GB of HBM3 on an 8192-bit bus with 5.32 TB/s bandwidth. The Arc B370 uses system shared memory, with its size, type, bus width, and bandwidth all dependent on the host system.

Q: Are these devices intended for the same market?

A: No. The MI300X is an OAM module accelerator with no display outputs and a 750 W TDP. The Arc B370 is an integrated graphics processor with a 25 W TDP and portable device dependent display outputs. They target entirely different platforms.

Q: What is the production status of each?

A: The Arc B370 is listed as active production. The MI300X has no production status recorded in the database.

Q: Which device supports modern graphics APIs?

A: The Arc B370 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The MI300X lists N/A for all graphics APIs, consistent with its compute-only design.

Specification Differences

The two devices differ across nearly every recorded specification field. The MI300X uses a 5 nm TSMC process, while the Arc B370 uses a 3 nm Intel process. The MI300X has 153,000 million transistors on a 1017 mm² die, while the Arc B370's transistor count and die size are unknown. The MI300X's transistor density is 150.4 million per square millimeter, while the Arc B370's is not recorded.

Clock speeds differ: the MI300X has a 1000 MHz base and 2100 MHz boost, while the Arc B370 has a 300 MHz base and 2400 MHz boost. Memory also differs: the MI300X has 192 GB of HBM3 at 1300 MHz with 5.2 Gbps effective speed, an 8192-bit bus, and 5.32 TB/s bandwidth, while the Arc B370 has system shared memory with system dependent specifications.

Compute resources: the MI300X has 19,456 shading units, 1,216 TMUs, and 0 ROPs, while the Arc B370 has 1,280 shading units, 40 TMUs, and 20 ROPs. The Arc B370 has 10 ray tracing cores, while the MI300X has none recorded. Pixel rates: the MI300X is 0 MPixel/s, the Arc B370 is 48.00 GPixel/s. Texture rates: the MI300X is 2,553.6 GTexel/s, the Arc B370 is 96.00 GTexel/s. FP32 performance: the MI300X is 81.72 TFLOPS, the Arc B370 is 6.144 TFLOPS. FP16 performance: the MI300X is 81.72 TFLOPS at 1:1, the Arc B370 is 12.29 TFLOPS at 2:1.

Power and form factor: the MI300X has a 750 W TDP, an OAM module slot, no power connectors, and a suggested PSU of 1150 W. The Arc B370 has a 25 W TDP, an IGP slot, no power connectors, and no suggested PSU. Bus interfaces: the MI300X uses PCIe 5.0 x16, the Arc B370 uses IGP. Display outputs: the MI300X has none, the Arc B370 has portable device dependent outputs. APIs: the MI300X lists N/A for DirectX, OpenGL, and Vulkan, while the Arc B370 lists 12 Ultimate (12_2), 4.6, and 1.4 respectively.

Release dates: the MI300X launched on 2023-12-05, the Arc B370 on 2026-01-26. The MI300X's predecessor is Radeon Instinct, while the Arc B370 has no recorded predecessor. The Arc B370's production status is active, while the MI300X's is not recorded.

Head-to-Head Benchmarks

The database contains no direct head-to-head benchmark entries between the MI300X and the Arc B370. Each device was tested in a different benchmark suited to its intended workload, and the recorded scores reflect that separation.

The MI300X's Geekbench OpenCL score of 317,994 is its only recorded benchmark. That score positions it at the 100th percentile among all GPUs, meaning no other GPU in the database scores higher in that test. Its nearest rival, the NVIDIA B200, scores 345,482 and leads by 8%, while the H200 NVL at 334,891 leads by 5%. The MI300X leads the L40S by 7.5% and the RTX 6000 Ada by 10.7%. Those deltas show a compute accelerator clustered with the top NVIDIA data center parts, trailing the newest B200 but beating the L40S and RTX 6000 Ada.

The Arc B370's 3DMark Steel Nomad DX12 score of 1,184 places it at the 5th percentile. Its nearest rivals are all older AMD and ATI parts, with scores ranging from 1,168 to 1,192. The Arc B370 trails the HD 7650M by 0.7%, the HD 5770 by 0.5%, and the HD 5570 by 0.2%, while it leads the FirePro M2000 by 1.4%. Those margins are within a couple of percentage points, indicating the Arc B370 performs near the level of those legacy discrete GPUs in this particular DirectX 12 test.

The biggest win for the MI300X is its absolute compute score, which is orders of magnitude above the Arc B370's result. The biggest win for the Arc B370 is its efficiency profile, with a 25 W TDP versus 750 W, and its integrated form factor that requires no external power connectors. The MI300X's FP32 throughput of 81.72 TFLOPS is more than 13 times the Arc B370's 6.144 TFLOPS. The MI300X's memory bandwidth of 5.32 TB/s is a dedicated resource, while the Arc B370's is system dependent.

The data shows two products optimized for different ends of the hardware spectrum. The MI300X delivers compute density at the 100th percentile, while the Arc B370 delivers integrated graphics at the 5th percentile. Neither benchmark result invalidates the other, because the workloads and platforms are not comparable. The MI300X wins on raw compute and memory capacity, while the Arc B370 wins on power consumption and portability.