StarZero All shows Try the podcast clipper →

Apr 6, 2017 · 21m · mad

The Power of GPU Analytics // Todd Mostak, MapD (FirstMark's Data Driven)

Todd Mostak · 15m spoken Matt Turck · 42s spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

In this presentation from FirstMark's Data Driven NYC, MapD CEO Todd Mostak demonstrates how GPU-powered database and rendering technologies overcome traditional CPU bottlenecks to execute sub-second analytics and interactive visualizations on billion-row datasets.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 3.9% of the talking time here. How this is scored →

Matt as informed peer 0.8 Guest teaching 2.4 Guest disagreement 0.2 Matt pushing back 0.4
05100:0010:0020:000:10–2:19 · Matt as informed peer 0/10 The Deluge of Data Growth Todd Mostak presents a monologue detailing the growing divide between data volume expansion and CPU processing capabilities. The host is absent during this introductory presentation segment.2:19–5:10 · Matt as informed peer 0/10 Architectural Differences: CPU vs. GPU Cores Todd Mostak explains GPU versus CPU architecture and how MapD uses GPU RAM for fast SQL query execution. This is a continuous solo presentation without host participation.5:10–9:47 · Matt as informed peer 0/10 MapD Immerse Hybrid Rendering Strategy Todd Mostak demonstrates live filtering on a 1.2 billion record NYC taxi dataset. The host does not intervene or comment during the live demonstration.9:47–13:35 · Matt as informed peer 0/10 Demo: Credit Card Data and ML Pipeline Todd Mostak showcases credit card transaction analytics and Python notebook integration for machine learning forecasting. The segment consists entirely of product demonstration monologue.13:35–21:13 · Matt as informed peer 4/10 Closing Thoughts and Value Proposition Matt Turck opens Q&A with targeted technical questions regarding data virtualization and algorithm execution, referencing database pioneer Mike Stonebraker. Todd Mostak answers collaboratively alongside audience queries.0:10–2:19 · Guest teaching 2/10 The Deluge of Data Growth Todd Mostak presents a monologue detailing the growing divide between data volume expansion and CPU processing capabilities. The host is absent during this introductory presentation segment.2:19–5:10 · Guest teaching 3/10 Architectural Differences: CPU vs. GPU Cores Todd Mostak explains GPU versus CPU architecture and how MapD uses GPU RAM for fast SQL query execution. This is a continuous solo presentation without host participation.5:10–9:47 · Guest teaching 2/10 MapD Immerse Hybrid Rendering Strategy Todd Mostak demonstrates live filtering on a 1.2 billion record NYC taxi dataset. The host does not intervene or comment during the live demonstration.9:47–13:35 · Guest teaching 2/10 Demo: Credit Card Data and ML Pipeline Todd Mostak showcases credit card transaction analytics and Python notebook integration for machine learning forecasting. The segment consists entirely of product demonstration monologue.13:35–21:13 · Guest teaching 3/10 Closing Thoughts and Value Proposition Matt Turck opens Q&A with targeted technical questions regarding data virtualization and algorithm execution, referencing database pioneer Mike Stonebraker. Todd Mostak answers collaboratively alongside audience queries.0:10–2:19 · Guest disagreement 0/10 The Deluge of Data Growth Todd Mostak presents a monologue detailing the growing divide between data volume expansion and CPU processing capabilities. The host is absent during this introductory presentation segment.2:19–5:10 · Guest disagreement 0/10 Architectural Differences: CPU vs. GPU Cores Todd Mostak explains GPU versus CPU architecture and how MapD uses GPU RAM for fast SQL query execution. This is a continuous solo presentation without host participation.5:10–9:47 · Guest disagreement 0/10 MapD Immerse Hybrid Rendering Strategy Todd Mostak demonstrates live filtering on a 1.2 billion record NYC taxi dataset. The host does not intervene or comment during the live demonstration.9:47–13:35 · Guest disagreement 0/10 Demo: Credit Card Data and ML Pipeline Todd Mostak showcases credit card transaction analytics and Python notebook integration for machine learning forecasting. The segment consists entirely of product demonstration monologue.13:35–21:13 · Guest disagreement 1/10 Closing Thoughts and Value Proposition Matt Turck opens Q&A with targeted technical questions regarding data virtualization and algorithm execution, referencing database pioneer Mike Stonebraker. Todd Mostak answers collaboratively alongside audience queries.0:10–2:19 · Matt pushing back 0/10 The Deluge of Data Growth Todd Mostak presents a monologue detailing the growing divide between data volume expansion and CPU processing capabilities. The host is absent during this introductory presentation segment.2:19–5:10 · Matt pushing back 0/10 Architectural Differences: CPU vs. GPU Cores Todd Mostak explains GPU versus CPU architecture and how MapD uses GPU RAM for fast SQL query execution. This is a continuous solo presentation without host participation.5:10–9:47 · Matt pushing back 0/10 MapD Immerse Hybrid Rendering Strategy Todd Mostak demonstrates live filtering on a 1.2 billion record NYC taxi dataset. The host does not intervene or comment during the live demonstration.9:47–13:35 · Matt pushing back 0/10 Demo: Credit Card Data and ML Pipeline Todd Mostak showcases credit card transaction analytics and Python notebook integration for machine learning forecasting. The segment consists entirely of product demonstration monologue.13:35–21:13 · Matt pushing back 2/10 Closing Thoughts and Value Proposition Matt Turck opens Q&A with targeted technical questions regarding data virtualization and algorithm execution, referencing database pioneer Mike Stonebraker. Todd Mostak answers collaboratively alongside audience queries.

speaking balance: gold is Matt, purple is the guest (3 minute bins)

0:00 · Matt 0% · guest 100%0:00 · Matt 0% · guest 100%3:00 · Matt 0% · guest 100%3:00 · Matt 0% · guest 100%6:00 · Matt 0% · guest 100%6:00 · Matt 0% · guest 100%9:00 · Matt 0% · guest 100%9:00 · Matt 0% · guest 100%12:00 · Matt 0% · guest 100%12:00 · Matt 0% · guest 100%15:00 · Matt 24.5% · guest 75.5%15:00 · Matt 24.5% · guest 75.5%18:00 · Matt 4.3% · guest 95.7%18:00 · Matt 4.3% · guest 95.7%21:00 · Matt 18.9% · guest 81.1%21:00 · Matt 18.9% · guest 81.1%
Sharpest disagreement ▶ 18:52 Playful rejection of official Stonebraker branding

Todd Mostak humorously corrects the host's framing regarding Mike Stonebraker's involvement, clarifying that MapD is not officially a Stonebraker company.

 Hardest push from Matt ▶ 15:38 Probing for missing architectural components

Matt Turck presses Todd on where complex analytics algorithms actually run, questioning what intermediate infrastructure might be missing.

 Biggest teaching moment ▶ 15:55 Explaining cohort execution natively in SQL

Todd Mostak clarifies that analytical models like cohort analysis do not require external engines but are run natively using SQL queries inside MapD.

 Matt holds his own ▶ 18:46 Referencing Mike Stonebraker's database lineage

Matt Turck demonstrates deep industry knowledge of database startups by asking whether MapD represents Mike Stonebraker's tenth database venture.
the scores for every segment, with the reasoning behind each
ChapterTopicMatt as informed peerGuest teachingGuest disagreementMatt pushing backWhy
The Deluge of Data Growth 0200 Todd Mostak presents a monologue detailing the growing divide between data volume expansion and CPU processing capabilities. The host is absent during this introductory presentation segment.
Architectural Differences: CPU vs. GPU Cores 0300 Todd Mostak explains GPU versus CPU architecture and how MapD uses GPU RAM for fast SQL query execution. This is a continuous solo presentation without host participation.
MapD Immerse Hybrid Rendering Strategy 0200 Todd Mostak demonstrates live filtering on a 1.2 billion record NYC taxi dataset. The host does not intervene or comment during the live demonstration.
Demo: Credit Card Data and ML Pipeline 0200 Todd Mostak showcases credit card transaction analytics and Python notebook integration for machine learning forecasting. The segment consists entirely of product demonstration monologue.
Closing Thoughts and Value Proposition 4312 Matt Turck opens Q&A with targeted technical questions regarding data virtualization and algorithm execution, referencing database pioneer Mike Stonebraker. Todd Mostak answers collaboratively alongside audience queries.

Statements from this episode (13)

Assertion Not checkable as stated
Mostak: GPUs deliver 100x speedups over CPUs for analytics workloads
“Using graphics cards to accelerate data warehouse and visual analytic workloads, literally getting hundred-x speed-ups over what you would get on a CPU solution.”
Todd Mostak Apr 6, 2017 ▶ 0:18
Assertion Supported
Mostak: Enterprise data volumes are growing by 40 percent year-over-year
“Basically, it's doubling in less than every three years, or 40% year-over-year growth.”
Todd Mostak Apr 6, 2017 ▶ 0:32
Prediction Didn’t hold up
Mostak: Storage costs will drop below eight dollars per terabyte by 2020
“They're saying that by 20 20, you'll be able to buy a terabyte of storage for less than eight dollars.”
Todd Mostak Apr 6, 2017 ▶ 0:59
Assertion Supported
Mostak: CPU power growth lags behind 40% annual data volume growth
“Data is growing at 40% year over year, CPU processing power is only growing at a 20% year over year rate.”
Todd Mostak Apr 6, 2017 ▶ 1:13
Assertion Partly supported
Mostak: A single 2017 GPU server rivals a 2007 top-ten supercomputer
“In fact, you're basically getting a supercomputer in a box, something that would have been a top 10 supercomputer 10 years ago on a single server today.”
Todd Mostak Apr 6, 2017 ▶ 2:48
Disclosure
Mostak: MapD streams live Kafka data for a top social media giant
“One of our clients is one of the largest social media companies in the world, and we actually pull in streaming data via Kafka.”
Todd Mostak Apr 6, 2017 ▶ 3:57
Assertion Not checkable as stated
Mostak: MapD scans GPU-cached data at up to six terabytes per second
“So we can actually, once the data's in GPU RAM, we can scan that data At rates approaching six terabytes a second.”
Todd Mostak Apr 6, 2017 ▶ 4:44
Assertion Not checkable as stated
Mostak: MapD queries billions of records in milliseconds on a single server
“And that's why we can run queries over billions of records in milliseconds, even on a single server.”
Todd Mostak Apr 6, 2017 ▶ 4:50
Assertion Supported
Mostak: MapD queries all data live without relying on pre-aggregation
“One of the great things about MapD is that we're not just pre-aggregating the data, that we're actually querying all the data live.”
Todd Mostak Apr 6, 2017 ▶ 11:03
Prediction Not checkable as stated
Mostak in 2017: GPUs will dominate computing over the next decade
“We're at an inflection point where the growth in data is outpacing the growth in compute, and I believe, I strongly believe, I'm a little biased, but, ah, that GPUs are set to kind of become a dominant force in compute over the coming decade. Ah, they already …”
Todd Mostak Apr 6, 2017 ▶ 13:38
Assertion Not checkable as stated
Mostak: Verizon's queries on billions of records previously took over an hour
“So Verizon, before using our platform, some of their queries over tens of billions of records of call data records Would literally take, ah, an hour or more on a standard kind of data warehouse platform.”
Todd Mostak Apr 6, 2017 ▶ 14:22
Assertion Supported
Mostak: Tableau data extracts fail on datasets exceeding 100 million rows
“Tableau extracts are great, but they kind of fall over, either when the extract gets over a hundred million rows, a few hundred million rows, or when you have, like, streaming data doing the system.”
Todd Mostak Apr 6, 2017 ▶ 17:16
Assertion Not checkable as stated
Mostak: Machine learning cannot yet automatically derive insights from raw data
“Machine learning has not gotten to the point where you can just point at a data and say, give me the insights.”
Todd Mostak Apr 6, 2017 ▶ 19:27
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 400 conversations transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.

Try the podcast clipper Learn more