iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
After more than a year away from C and C++, Dijo Torres is starting again with small math exercises—and using questions about local AI inference to find a path into accelerator hardware and the computations behind it. His DEV Community post is a personal learning kickoff, not an embedded-systems tutorial or hardware guide.
Why Torres is returning to C++
Torres says it had been more than a year since he last coded in C or C++. Rather than begin with a large project, he chose small mathematical problems as a way back into C++ practice.
That modest starting point soon connected to a larger curiosity: how long would his PC’s GPU take to process a request from a locally run large language model? Following that question led him to consider GPUs, TPUs, and NPUs, and to ask what kinds of computations make these processors useful.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsFrom local inference to matrix multiplication
The thread running through the post is a progression of questions. Local inference prompts curiosity about accelerator hardware; learning that matrix multiplication is central to many machine-learning workloads leads to questions about how to perform it efficiently; those questions open onto the systems and libraries built around accelerator computation.
#1 Best Overall
Torres names GEMM—general matrix multiplication—and BLAS, a family of basic linear algebra routines, alongside hardware and software concepts including systolic arrays, XLA, MXUs, cuDNN, and CUTLASS. These names sketch the territory he wants to explore; the post does not compare them or explain how to select a tool for a particular project.
The AlphaTensor example that caught his attention
One result in the post illustrates why faster matrix multiplication is an interesting problem. Torres describes a 2022 Google DeepMind paper in which reinforcement learning was used to find matrix multiplication algorithms requiring fewer multiplications than previously known algorithms. For a 4×4 modular-arithmetic example, his post reports 47 multiplications, compared with Strassen’s 49. That figure is the post’s account of the example, not an independently verified performance result or a claim about every matrix multiplication workload.
Rank #2
What the post is—and what it is not
This is an introduction to a learning process, not a technical reference for embedded programming. It does not offer a complete curriculum, teach the named libraries, or evaluate hardware. Its value is in showing how one concrete question—what happens when a local model runs on a PC?—can lead from C++ practice toward the math and systems behind machine-learning computation.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →A commitment to learn in public
Torres closes with the intention to share what he learns as he goes: “Starting today, I’m going to document everything I learn here, in public.” The post is therefore best read as the opening note in a continuing personal journey, rather than a report of completed embedded-systems work.
Rank #3
- Used Book in Good Condition
Read Dijo Torres’s original post on DEV Community.
Quick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
