Sidebar

[Jeff P]

The servers that are running AI/deep learning algorithms in the data centers I described (above) are configured to use a completely different info model than what you are describing --

Under a Standard Info Model --

• Just about everything is executed by the CPU - including requests from the server's operating system - plus requests from all of the users/visitors browsing all of the sites that are hosted on the server.

This info model works well if you are running standard or non-AI/deep learning programs on the server.

The programs being run are executing tasks such as verifying logon credentials, managing session state of site visitors, performing mathematical calculations, retrieving files, overwriting files, rendering requested web pages, reading data from databases, writing data to databases, and performing automated backups, etc.

Generally, this works rather well because the CPU is perfectly capable of handling all of the requests being made.

But this standard info model doesn't work so well if the number of requests rises beyond a certain point. If that happens - and a good example might be a DOS or denial of service attack: The CPU reaches a point where it simply cannot keep up with all of the new requests coming in -- and the server slows to a crawl.

This standard info model also doesn't work so well if the server's primary purpose is to run AI/deep learning algorithms.

Don't get me wrong - under a standard info model where everything is executed by the CPU:

The server is perfectly capable of running AI/deep learning algorithms -- just not quickly.


Why use High End GPUs? --

• High end GPUs are faster and more powerful than a CPU. They are able to process a greater number of decisions/transactions per second than a CPU.

• The AI/deep learning algorithms researchers are working on are often called upon to process a LOT more transactions/calculations before reaching a simple classification decision (such as 0 or 1) than non-AI/deep learning algorithms.

For these two reasons, pretty early on, researchers working on AI discovered that GPUs are better than CPUs for running AI/deep learning algorithms.

Once chip makers like AMD, Intel, Nvidia, and others realized that - they began working on GPUs designed specifically for AI.

Somewhere along the way - Nvidia began snatching up market share from AMD and Intel in this new space -- Imo because they've been cranking out GPUs with specs that are far and away superior to those made by AMD and Intel. (That doesn't mean AMD or Intel or someone else couldn't announce a major breakthrough tomorrow. But right now as I type this Nvidia is the clear leader in this space.)



Alternate Info Model --

The servers in the data centers I described in my previous post (above) are configured in such a way that nearly all of the requests being made by the AI/deep learning algorithms are executed by banks of High End GPUs - instead of by the CPU.

As a result the AI/deep learning algorithms are running at max possible speed (up to the limits imposed by the latest available technology.)

Hope I managed to type most of that out in a way that makes sense,



-jp

.