Voltron Data throws its weight behind AMD for GPU-accelerated SQL

Support for AMD's Instinct GPUs is coming to Voltron Data's accelerated SQL engine Theseus in the latest sign Nvidia's CUDA moat is getting shallower. Introduced in late 2023, Theseus — named for the ship of Theseus since it's constantly being stripped down and rebuilt — uses GPUs to accelerate SQL queries, enabling large …

[DS999]

I guess this was bound to happen

People used to run their databases with CPUs back when the storage was hard drives. NVMe storage has accelerated the IOPS capacity 10,000 fold and while CPUs are faster and cores more numerous than they were in the hard drive days they aren't even 100x faster let alone 10,000x. So it makes sense to find ways to accelerate the SQL query side as much as the storage side has been accelerated.

[werdsmith]

Re: I guess this was bound to happen

It has always remained a problem where more powerful hardware is thrown at databases to try and improve shitty sub-optimal queries with massive select lists from outer joins.

Don’t encourage them.

[elsergiovolador]

Re: I guess this was bound to happen

Most data engineers today don’t know what a full table scan is, and they don’t care - as long as the SQL vomits something vaguely plausible and doesn’t crash, it’s good enough for a stakeholder slide. Query runs for hours? Perfect, more time for meetings and coffee.

Now the industry just throws silicon at the problem. Instead of investing in talent, we’re optimising for lowest-common-denominator usage. GPU acceleration isn’t enabling better work - it’s compensating for the fact that nobody’s paid enough to care about writing efficient queries anymore. And honestly, why should they?

[vistisen]

Re: I guess this was bound to happen

We refer to badly written sqls that use joins with no conditions as 'happy slapping'

[DS999]

Re: I guess this was bound to happen

That's true for ALL kinds of software. The very slow improvement of disk IOPS over the years acted as a limiting factor for people writing bad SQL queries in a way that faster CPUs and more RAM didn't for people writing more general software. Now the floodgates are opened.

If you're concerned about that, maybe keep your old disk array around and force the DBAs to use it on their development system :)

[werdsmith]

Re: I guess this was bound to happen

DBAs?

Queries are dynamically generated by applications behind a front end where the end users ask for the information themselves by dragging stuff around on the screen with a mouse. These are managers with these tools now, not even the comfort of a professional data analyst (report writer in honest language).

[HuBo]

Good move

Great to see Voltron taking advantage of AMD's recent ROCm-DS hipDF lib to broaden the GPU base on which its TPC-H record breaking software runs! Aramburu and Patterson (and team) have done great work in this imho.

They're running the accelerated Theseus SQL engine atop "open sources distributed file systems and file formats" afaik, and given Intel's DAOS io500 performance (Distributed Asynchronous Object Storage -- that was initially going to be a 3D XPoint and/or Optane thing), I wonder what kind of perf they may get on top of that, or if (as DS999 suggests above) the io is really not that much of a bottleneck anymore here.