| [Real-time scoring CLR extension](real-time-scoring.md) | [sp_rxPredict](https://docs.microsoft.com/sql/relational-databases/system-stored-procedures/sp-rxpredict-transact-sql) on a serialized model | R: RevoScaleR, MicrosoftML <br/>Python: revoscalepy, microsoftml | Tens of milliseconds, on average. |

| [Native scoring C++ extension](native-scoring-predict-transact-sql.md) | [PREDICT T-SQL function](https://docs.microsoft.com/sql/t-sql/queries/predict-transact-sql) on a serialized model | R: RevoScaleR <br/>Python: revoscalepy | Less than 20 milliseconds, on average. |

[**Real-time scoring**](real-time-scoring.md) was the first solution for high-performance scoring. Introduced in early versions of SQL Server 2017 and later updates to SQL Server 2016, real-time scoring relies on CLR libraries that stand in for R and Python processing over Microsoft-controlled functions in RevoScaleR, MicrosoftML (R), revoscalepy, and microsoftml (Python). CLR libraries are invoked using the **sp_rxPredict** stored procedure to generates scores from any supported model type, without calling the R or Python runtime.

[**Native scoring**](native-scoring-predict-transact-sql.md) is a SQL Server 2017 feature, implemented as a native C++ library, but only for RevoScaleR and revoscalepy models. It is the fastest and more secure approach, but supports a smaller set of functions relative to other methodologies.

+ [Native scoring](../predictions/native-scoring-predict-transact-sql.md), for fast batch prediction from SQL Server without calling R

For all these reasons, there is no single solution for mitigating this initial loading time, as the performance impact varies significantly depending on the task. For example, caching is performed for single-row scoring in batches; hence, successive scoring operations are much faster and neither the model nor the R runtime is reloaded. You can also use [native scoring](../predictions/native-scoring-predict-transact-sql.md) to avoid loading the R runtime entirely.

You can use the T-SQL [PREDICT](../t-sql/queries/predict-transact-sql.md) function to perform [native scoring](predictions/native-scoring-predict-transact-sql.md) on a pre-trained model that has been previously saved in the required binary format.

| Initial release | [**R integration for in-database analytics**](r/sql-server-r-services.md). <br/><br/> R packages for calling R functions in T-SQL, and vice versa. RevoScaleR functions provide R analytics at scale by chunking data into component parts, coordinating and managing distributed processing, and aggregating results. In SQL Server 2016 R Services (In-Database), the RevoScaleR engine is integrated with a database engine instance, brining data and analytics together in the same processing context. <br/><br/>T-SQL and R integration through [sp_execute_external_script](https://docs.microsoft.com/sql/relational-databases/system-stored-procedures/sp-execute-external-script-transact-sql). You can call any R code using this stored procedure. This secure infrastructure enables enterprise-grade deployment of Rn models and scripts that can be called from an application using a simple stored procedure. Additional performance gains are achieved by streaming data from SQL to R processes and MPI ring parallelization. <br/><br/>You can use the T-SQL [PREDICT](../t-sql/queries/predict-transact-sql.md) function to perform [native scoring](predictions/native-scoring-predict-transact-sql.md) on a pre-trained model that has been previously saved in the required binary format.|

On Linux, SQL Server 2017 does not have R or Python integration, but you can use [Native scoring](predictions/native-scoring-predict-transact-sql.md) on Linux because that functionality is available through T-SQL [PREDICT](../t-sql/queries/predict-transact-sql.md), which runs on Linux. Native scoring enables high-performance scoring from a pretrained model, without calling or even requiring an R runtime.