Check your Snowflake and Databricks bill
We put Teradata, Snowflake, and Databricks through a rigorous, real-world workload comparison, measuring cost per query and query performance to show exactly how each platform holds up under the demands of modern data and AI. Before your next Snowflake invoice arrives, see how Teradata helps you do more with faster queries, smarter analytics, and production-ready AI, without paying more for it.
| Teradata | Snowflake | Databricks | |
|---|---|---|---|
| Cost per query |
$0.0009
Teradata
|
$0.0686
Snowflake
|
$0.0108
Databricks
|
| Queries processed |
197,366
Teradata
|
3,144
Snowflake
|
23,825
Databricks
|
Lowest average per-query cost
Cost per query comparison across Teradata, Snowflake, and Databricks using a $70 per compute-hour baseline.
The operational cost per query for each platform was analyzed using a base unit of $70 per compute hour. That represented the hourly cost of running each cloud-based platform under the following conditions:
- Generally available software versions
- Mid-tier platform offering
- One-year service pricing commitment
Each system was configured to maximize value within this budget constraint, utilizing the best publicly available knowledge for configuration and setup. The analysis demonstrated a stark contrast:
- Teradata: $0.0009 per query
- Snowflake: $0.0686 per query
- Databricks: $0.0108 per query
Teradata was found to be more than 20x cheaper than Snowflake and 12x cheaper than Databricks.
Highest query performance throughput
Query throughput comparison across Teradata, Snowflake, and Databricks using 50 queries representing mixed analytical workloads.
The analysis included 50 queries spanning multiple query types to simulate a modern, data-driven workload. We tested them over a two-hour period designed to gauge each platform's ability to handle complex analytical queries, operational reporting, and tactical queries that reflect the demands of data-driven organizations.
The analysis showed a massive difference between platforms:
- Teradata: 197,366 queries
- Snowflake: 3,144 queries
- Databricks: 23,825 queries
That means Teradata was able to handle 62 times more queries than Snowflake and 8 times more queries than Databricks in the same amount of time, under similar circumstances.
Claims above are based on real-world mixed analytical workloads on competitive systems of comparable configurations. For more information on the Teradata competitive workload comparison methodology and process, contact John.Myers@Teradata.com.
Workload comparison methodology: Teradata vs. Snowflake vs. Databricks
Study conducted by Teradata Market Intelligence (author: John Myers, Senior Director, Market Intelligence). Published July 2024 as “Cloud-Based Analytics Platforms: A Workload Comparison Study” (Version 1.0). Testing was performed by Teradata on generally available software versions as of that date. The methodology was designed to ensure fair and meaningful comparisons across platforms with different architectures. It has three components:
A real-world workload mix
The workload combines queries from an industry-standard decision support benchmark (TPC-H) with queries that simulate diverse, concurrent, operational analytics. It consists of 50 queries and a throughput test that runs continuously for two hours, stressing every layer of the platform, from query optimizer to workload management to storage, rather than running through a set of queries once.
Benchmark foundation: the 22 business-oriented queries from the Transaction Processing Performance Council’s TPC-H specification. Added queries: 28 queries simulating real-world, data-driven scenarios, including complex analytical queries, reporting, and operational analytics. Caching control: queries are parameterized so the same data is not requested in each round of execution, which limits caching. Note: this is a modified, Teradata-designed workload built on TPC-H. It is not an official, audited TPC-H result.
Cost-equivalent environments
The target was approximately $70 per compute hour for a base unit of processing, using generally available software versions, a mid-tier platform offering, and a one-year service pricing commitment. Exact parity was not possible because vendor pricing and node specifications differ. Each system was configured to maximize value within that budget using best practices and publicly available configuration guidance. During the test, each system could scale up to four units of processing, so actual costs diverged from the base unit cost depending on how each platform scaled.
Minimized tuning
No workload-specific tuning was performed. Performance features that DBAs would commonly enable and that were generally available were deployed. This included materialized views, deployed on select larger tables in line with DBA best practices on every platform that offered them. Teradata calls this feature a “single table join index,” and Snowflake calls it a “materialized view.” Databricks did not offer it as a generally available feature when the test was executed, so it was not included in the Databricks configuration.
Enterprise price performance formula
Enterprise price performance = Hourly testing cost ÷ Queries completed per hour (reported as cost per query).
Results
Teradata vs. Snowflake vs. Databricks: 2-hour throughput test
| Teradata Vantage Cloud | Snowflake | Databricks | |
|---|---|---|---|
| Base unit cost (per compute hour) |
$69.31
Teradata Vantage Cloud
|
$72.00
Snowflake
|
$72.80
Databricks
|
| Actual two-hour testing cost |
$358.25
Teradata Vantage Cloud
|
$431.25
Snowflake
|
$513.05
Databricks
|
| Queries completed per hour |
197,366
Teradata Vantage Cloud
|
3,144
Snowflake
|
23,825
Databricks
|
| Cost per query |
$0.0009
Teradata Vantage Cloud
|
$0.0686
Snowflake
|
$0.0108
Databricks
|
| Cost relative to Teradata |
1x
Teradata Vantage Cloud
|
~75x
Snowflake
|
~12x
Databricks
|
In this test, Teradata completed the most queries per hour at the lowest cost per query. The study attributes the difference in actual cost to scaling behavior: Snowflake and Databricks provisioned additional compute more aggressively as demand rose, while Teradata used its available compute more fully before scaling, relying on its query optimization and workload management.
These results reflect a Teradata-conducted test of a specific mixed analytical workload on cost-equivalent, not identical, configurations. Results will vary with workload, configuration, and pricing. For more on the Teradata competitive workload comparison methodology, contact John.Myers@Teradata.com.
Related resources
Workload comparison FAQs
What is the difference between Teradata, Snowflake, and Databricks?
What is the difference between Teradata, Snowflake, and Databricks?
Teradata, Snowflake, and Databricks all run in the cloud, but each describes itself differently. Teradata offers its Autonomous Knowledge Platform with a choice of deployment options, including Teradata Cloud, as well as on-premises and hybrid environments. Snowflake describes its platform as a unified data cloud platform for building, using, and sharing data and applications. Databricks describes itself as a lakehouse-based platform that combines data lake and data warehouse capabilities for data engineering and analytics.
How does Teradata compare with Snowflake and Databricks on cost per query?
How does Teradata compare with Snowflake and Databricks on cost per query?
In a Teradata-conducted workload comparison published in July 2024, Teradata had the lowest cost per query: $0.0009, compared with $0.0686 for Snowflake and $0.0108 for Databricks. Calculated from these figures, Snowflake’s cost per query was about 75x Teradata’s, and Databricks’ was about 12x.
Cost per query is the hourly testing cost divided by average queries completed per hour. All three environments started cost-equivalent at approximately $70 per compute hour and could scale up to four units of processing, so actual costs differed. Results reflect a Teradata-conducted test of a mixed workload on cost-equivalent, not identical, configurations at the time of testing.
Source:
Teradata: Cloud-Based Analytics Platforms: A Workload Comparison Study (Version 1.0), July 2024.
How does Teradata compare with Snowflake and Databricks on query performance?
How does Teradata compare with Snowflake and Databricks on query performance?
In a Teradata-conducted workload comparison published in July 2024, Teradata averaged 197,366 queries completed per hour over a two-hour throughput test, compared with 23,825 for Databricks and 3,144 for Snowflake. That is about 63x Snowflake’s throughput and 8x Databricks’.
The workload was a 50-query mix of analytical, reporting, and operational analytics (tactical) queries, run continuously for two hours. All three environments started cost-equivalent at approximately $70 per compute hour and could scale up to four units of processing. Results reflect cost-equivalent, not identical, configurations at the time of testing.
Source:
Teradata: Cloud-Based Analytics Platforms: A Workload Comparison Study (Version 1.0), July 2024.
What is a workload comparison?
What is a workload comparison?
In Teradata’s methodology, a workload comparison test evaluates cloud analytics platforms by running a mixed set of queries continuously over a set period and measuring how each platform handles diverse, concurrent workloads, rather than measuring individual queries in isolation. It measures performance, cost efficiency, and scalability.
By contrast, a serial test runs queries one after another and shows how each performs on its own. Teradata’s white paper likens a serial test to drag racing and a throughput test to an endurance race. Teradata’s approach also sets cost-equivalent environments and minimizes system tuning.
Source:
Teradata: Cloud-Based Analytics Platforms: A Workload Comparison Study (Version 1.0), July 2024.
How was the Teradata, Snowflake, and Databricks comparison conducted?
How was the Teradata, Snowflake, and Databricks comparison conducted?
Teradata published the comparison in a July 2024 white paper authored by John Myers, Senior Director, Market Intelligence. It tested Teradata, Snowflake, and Databricks using three principles: a real-world workload mix, cost-equivalent environments, and minimal system tuning. Each platform was configured at approximately $70 per compute hour using generally available software, a mid-tier offering, and a one-year pricing commitment, and could scale up to four units of processing during the test. No workload-specific tuning was performed beyond features database administrators commonly enable, including materialized views where generally available. Databricks was not configured with this feature because it was not generally available at the time of the comparison. The test ran continuously for two hours on a 50-query workload.
Source:
Teradata: Cloud-Based Analytics Platforms: A Workload Comparison Study (Version 1.0), July 2024.
What workloads were included in the comparison?
What workloads were included in the comparison?
The workload had 50 queries: 22 queries derived from the Transaction Processing Performance Council’s TPC-H decision support benchmark, and 28 added queries simulating real-world, data-driven scenarios, including analytical, reporting, and operational analytics (tactical) queries. Queries were parameterized each round to limit data caching, and the test ran continuously for two hours, stressing the query optimizer, workload management, and storage. Because it adds queries and runs as a mixed throughput test, this workload is derived from TPC-H and is not comparable to published TPC-H results. Cost per query is not a direct TPC metric.
Sources:
Teradata: Cloud-Based Analytics Platforms: A Workload Comparison Study (Version 1.0), July 2024.
TPC: Fair Use of TPC Benchmarks (Quick Reference v1.0.0), June 5, 2019.
How does Teradata perform on mixed analytical workloads?
How does Teradata perform on mixed analytical workloads?
In a Teradata-conducted workload comparison published in July 2024, Teradata averaged 197,366 queries completed per hour on a mixed workload of analytical, reporting, and operational analytics (tactical) queries, compared with 23,825 for Databricks and 3,144 for Snowflake. The white paper attributes the result to Teradata’s query optimization and workload management.
According to the white paper, Teradata used available compute resources more fully before requesting more, while Snowflake and Databricks provisioned additional compute more quickly as demand rose. This is Teradata’s own explanation, and the paper does not isolate a single cause. Teradata describes these query optimization and workload management capabilities as patented. Results reflect cost-equivalent, not identical, configurations at the time of testing.
Source:
Teradata: Cloud-Based Analytics Platforms: A Workload Comparison Study (Version 1.0), July 2024.
How does workload management affect cloud analytics performance and cost?
How does workload management affect cloud analytics performance and cost?
Workload management controls how an analytical platform prioritizes, queues, and allocates compute resources across multiple concurrent queries. Teradata’s white paper argues it affects both throughput and cost, because a platform that uses its existing compute more fully before scaling can complete more queries for a similar spend.
In the study, actual two-hour testing costs were $358.25 for Teradata. Costs were $431.25 for Snowflake and $513.05 for Databricks—about 20% and 43% higher than Teradata’s, respectively. Cost per query uses the hourly cost, which is half of each two-hour total. The much larger gap in cost per query ($0.0009, $0.0686, and $0.0108) came mainly from differences in overall throughput. The paper credits query optimization and workload management but does not isolate either as the sole cause.
Source:
Teradata: Cloud-Based Analytics Platforms: A Workload Comparison Study (Version 1.0), July 2024.
Get started with your migration to Teradata today
Fill out the form to engage with Teradata specialists to discuss your specific requirements and objectives and develop a customized migration plan tailored to your needs.