ByteMLPerf

Benchmark focuses on AI Accelerators

From a practical production perspective.

logo
🚀

Open Source & Reproducibility

ByteMLPerf offers an open-source AI accelerator benchmarking tool, ensuring easy access and utilization for companies and research institutions alike.

AI Production-Centric Design

ByteMLPerf continually updates its benchmarks to reflect current business scenarios and the state-of-the-art (SOTA).

💻

Business & SOTA Aligned

The tool evaluates beyond just performance and accuracy, taking into account factors such as compiler usability and the applicability of models in real business environments.

🔍

Comprehensive Assessment Standards

ByteMLPerf focuses on a holistic evaluation, including power consumption, cost-effectiveness, and cross-platform compatibility.

🌐

Community-Driven Innovation

Encourages contributions from developers and researchers worldwide, making the benchmarking tool not just an assessment tool but also a platform for innovation.

🎯

Transparency and Openness

Maintains a high level of transparency and openness, with all testing methodologies, datasets, and evaluation criteria being publicly accessible.

Reliable Benchmarking Results

Emphasizing the reproducibility of results ensures that the evaluations are not only accurate but also trustworthy, allowing all participants to compare on a fair and transparent basis.

Inference General Benchmark

Vendor Chip Infomation

Provide accurate chip data for user reference and comparison.

SKUBoard DesignMemory SpecificationsComputing Performance SpecificationsNetworking Parameters
VendorNamePurposePictureProcess Size(NM)Board SizeBus InterfaceTDP(W)Memory Hierarchy GraphMemory一级缓存二级缓存PE层次架构图PEScalar ParametersVector ParametersTensor Parameters通信方式端口数量RDMA协议下行带宽上行带宽
Memory TypeMemroy Size(GB)Memory Bandwidth(GB/s)缓存类型缓存容量缓存带宽缓存类型缓存容量缓存带宽算力架构Parallelism Mode通信带宽Scalar PrecisionINT8向量算力FP16向量算力FP32向量算力Vector PrecisionINT8向量算力FP16向量算力FP32向量算力Tensor PrecisionINT8向量算力FP16向量算力FP32向量算力
HabanaGaudi2Training/InferenceDeviceOAM 1.1600Memory HierarchyHBM2e962450United Buffer4811.2Memory Hierarchy异构多核
AWSTrainiumTraining/Inference-FHFL, Dual Slot CardPCIe 5.0x16Memory Hierarchy32820-
AWSInferentiaInference---
AWSInferentia2Inference-FHFL, Dual Slot CardPCIe 5.0x16Memory Hierarchy32820-
QUALCOMMAIC100InferenceDevice7HHHL, Single Slot CardPCIe 4.0x875Memory HierarchyLPDDR4x32137Memory Hierarchy
Stream ComputingSTC920InferenceDevice12FHFL, Dual Slot CardPCIe 4.0x16150-LPDDR4x16119.4-
MoffettS30InferenceDevice12FHFL, Dual Slot CardPCIe 4.0x16250Memory HierarchyLPDDR4x60246Distributed Buffer(x12)Memory Hierarchy异构多核SIMD/MIMT
MoffettS4InferenceDevice12FHFL, Single Slot CardPCIe 3.0x1670Memory HierarchyLPDDR4x2082Distributed Buffer(x4)1.88211.7Memory Hierarchy异构多核SIMD/MIMT204.8
MoffettS10InferenceDevice12FHFL, Single Slot CardPCIe 4.0x16165Memory HierarchyLPDDR4x40164Distributed Buffer(x8)3.7-异构多核SIMD/MIMT409.6
GraphcoreIPU C600InferenceDevice7FHFL, Dual Slot CardPCIe 4.0x16180Memory Hierarchy无片上DDRDistributed Buffer90065-同构众核MIMD
NVIDIAT4Training/InferenceDevice12HHHL, Single Slot CardPCIe 3.0x1670-GDDR616320Cache(x40)2.564-同构众核SIMT
NVIDIAA100 PCIeTraining/InferenceDevice7FHFL, Dual Slot CardPCIe 4.0x16300-HBM2e801935Cache(x108)20.73640-同构众核SIMT
NVIDIAH100 PCIeTraining/InferenceDevice4FHFL, Dual Slot CardPCIe 5.0x16350-HBM3802039Cache(x114)29.18450-同构众核SIMT
NVIDIAA30 PCIeTraining/InferenceDevice7FHFL, Dual Slot CardPCIe 4.0x16165-HBM2e241223Cache(x56)10.75224-同构众核SIMT
NVIDIAA100 SXM4Training/InferenceDevice7N/ASXM400-HBM2e802039Cache(x108)20.73640-同构众核SIMT
NVIDIAA10 PCIeTraining/InferenceDevice8FHFL, Single Slot CardPCIe 4.0x16150-GDDR624600.2Cache(x72)9.1266-同构众核SIMT
NVIDIAH100 SXM5Training/InferenceDevice4N/ASXM700-HBM3803350Cache(x132)33.79250-同构众核SIMT

Chip Comparison

Compare details between different chips.

FeatureA100-SXM4Gaudi2
厂商NVIDIAHabana
型号A100 SXM4Gaudi2
用途Training/InferenceTraining/Inference
接口SXMOAM 1.1
内存类型HBM2eHBM2e
缓存类型Cache(x108)United Buffer
缓存容量(MB)20.73648
算力架构同构众核异构多核
通信方式NV-LinkRoCE-v2

Supporting Vendors

Vendor 1Vendor 2Vendor 3Vendor 4

Contributors

© 2024 Bytedance Inc. All Rights Reserved.