Unlocking Optimal Performance: Certifying L40S and H100 GPUs with SQream

By SQream

11.16.2023 twitter linkedin facebook

In the quest for enhanced data processing and analytics performance, selecting hardware that seamlessly integrates and performs optimally is crucial. The choice of GPUs plays a pivotal role in not only accelerating data tasks but also in ensuring the scalability and efficiency of data operations. With the aim to provide a clearer roadmap towards achieving superior performance, we at SQream undertook the initiative to test and certify the L40S and H100 GPUs. The insights derived from this exercise are poised to significantly benefit organizations striving to elevate their data processing capabilities to the next level.

Certification of L40S and H100 GPUs

We are thrilled to announce that the L40S and H100 GPUs have been certified for use with SQream. This certification is a testament to SQream’s commitment to keeping pace with evolving hardware technologies and ensuring our customers have the best tools at their disposal.

Dive into the Benchmark: TPC-H Test

Our benchmark of choice was the TPC-H test, conducted with 1TB of dataset. While a larger dataset would have provided more concrete insights, the 1TB test still yielded significant findings.

The server specifications for the tests were as follows:

– 4th Gen Intel Xeon Scalable LGA4677, 36 cores, 4TB RAM

– One server with L40S 48GB GPU and another with H100 80GB GPU

– Operating System: REHL 8.6 for the H100 machine, and REHL 8.8 for the L40S machine, with Cuda version 12.2.1 on both.

Performance Insights: Worker Configurations Matter

The tests were structured to assess the performance of these GPUs under different worker configurations: 1, 4, and 8 workers per GPU. Here’s a snapshot of the performance differentials:

Data Loading Efficiency: L40S showcased excellent performance across all worker configurations.

Running TPC-H Queries: H100 outperformed with a single worker per GPU configuration, being 24% faster. 

Splitting Resources: A Path to Enhanced Performance

Our exploration didn’t stop at mere performance comparison. We delved deeper to understand the impact of splitting GPU resources on data querying and loading:

– Splitting both GPUs to 4 workers yielded better results for data queries, while an 8-worker split significantly accelerated data loading tasks.

Tailoring Configurations: Every Use Case is Unique

The takeaways from our tests underline the importance of tailoring GPU configurations to the specifics of each use case. The optimal configuration can vary, potentially influenced by server hardware, the desired analytical use case, and other factors (remember that SQream can be adapted to accommodate different use cases on the same system using services).

Unlocking New Opportunities for Large Enterprises

Large enterprises across industries stand to gain numerous benefits from harnessing the power of SQream’s certified L40S and H100 GPUs. Here are just a few compelling use cases that highlight the transformative capabilities of our high-performance data analytics solution:

  1. Accelerated Data Science and AI: Large enterprises heavily rely on data science and AI to gain valuable insights, drive innovation, and make informed business decisions. By leveraging SQream’s certified GPUs, these organizations can significantly accelerate the processing and analysis of massive data sets, enabling data scientists and AI practitioners to extract deeper insights, uncover hidden patterns, and develop cutting-edge AI models in record time.
  2. Faster Analytics: The ability to process and analyze immense volumes of data is paramount for large enterprises operating in fast-paced environments. SQream’s GPUs, combined with our advanced analytics platform, empower these organizations to perform rapid analysis on massive data sets, facilitating faster decision-making, proactive problem-solving, and optimizing operational efficiency.
  3. Improved Customer Experience: Large enterprises often grapple with the challenge of meeting customer expectations in today’s hyper-connected world. With SQream’s high-performance analytics solution, organizations can harness the power of GPU-accelerated data processing to deliver personalized, real-time experiences to customers. By leveraging comprehensive, near-instantaneous insights, businesses can better understand customer preferences, anticipate needs, and tailor offerings to provide a superior customer experience.
  4. Optimized Supply Chain Management: Efficient supply chain management is crucial for large enterprises operating in global markets. SQream’s GPU-accelerated analytics solution empowers these organizations to process and analyze massive amounts of supply chain data, enabling them to optimize inventory levels, reduce costs, streamline logistics, and enhance overall supply chain operations. Real-time visibility into the supply chain allows for agile decision-making and the ability to quickly adapt to changing market conditions.
  5. Enhanced Fraud Detection and Security: Large enterprises face constant threats from fraudulent activities and security breaches. With SQream’s GPU-accelerated analytics, organizations can rapidly analyze vast amounts of data to detect anomalies, patterns, and potential security breaches in real-time. By leveraging the power of certified GPUs, large enterprises can strengthen their fraud detection capabilities, mitigate risks, and proactively protect their valuable assets and sensitive customer information.

These use cases represent a snapshot of the potential benefits that large enterprises can unlock by harnessing the power of SQream’s certified L40S and H100 GPUs. By embracing cutting-edge hardware technologies and leveraging our high-performance analytics platform, organizations can transcend the limitations of traditional data processing, uncover valuable insights, and drive innovation, ultimately gaining a competitive edge in their respective industries.


The certification of L40S and H100 GPUs and the insights garnered from the tests are steps forward in our relentless pursuit of performance optimization. 

Our testing results emphasize the necessity of customizing GPU setups according to the unique requirements of each application. The ideal configuration may differ based on several elements, including server hardware, the specific analytical application, and additional variables. It’s important to note that SQream is designed to be flexible, allowing adaptation to various applications within the same system through the use of services

Check out this blog to see how H100 compared to A100 on the TPC-H benchmark.