Once upon a time, you had to carefully budget your microcontroller’s resources if you wanted to do something as simple as flash a bunch of LEDs. These days, they’re powerful enough to humiliate the game consoles of yesteryear. [alexkid77] demonstrates this well, having the ESP32-P4 run Tomb Raider.
Now, [alexkid77] hasn’t gone so far as to create a PlayStation emulator on the ESP32 or anything quite like that. Instead, this is a port—and not of the original Tomb Raider release, either. [alexkid77] started with OpenLara—the classic game running in an open-source engine. With the ESP32-P4 having two cores running at 400 MHz each, there was plenty of processing power on tap to run the engine with a software renderer at 320×240, which is hardware scaled up to 1024×600 via the Pixel Processing Accelerator (PPA) built into the chip. There’s also stereo audio with an ES8311 codec hooked up, while input is via a USB HID keyboard.
It’s funny to think that it could actually be cheaper and quicker to get Tomb Raider running on an ESP32 and a cheap LCD display versus actually going out to buy a PlayStation and an original game disc. But that’s the way the cookie crumbles in 2026. At least you don’t have to play it on an S3 Verge.
S3 Virge
Definitely better looking than the S3 Virge. It was my first 3D card, if you can call it a 3D card. It was basically software rendering with blurring to smooth out the jagged edges.
OpenLara targets the GBA with a 16 MHz CPU and 256 kb ram and no GPU. The original PlayStation had a 33 MHz CPU. How is it impressive to get this running on a dual core 400 MHz beast?
Vertical video? 👎
Youtube Short? 👎
This culturual/social phenomenon is known as VVS, the Vertical Video Syndrome.
It’s been explained here: https://www.youtube.com/watch?v=f2picMQC-9E
https://en.wiktionary.org/wiki/vertical_video_syndrome
“S3 Verge” is, I believe, the French spelling of it.
ESP32-P4 = Pentium 4
Running the same software too
Kinda crazy
I do mave my doubts.
The Pentium IV had implemented a very sophisticated x86 design no ARM core comes currently close to.
https://www.eeeguide.com/architecture-of-pentium-4-processor/
The Google LLM describes the performance comparison as such:
“The ESP32-P4 uses a dual-core RISC-V CPU running up to 400 MHz with hardware H.264 encoding and security accelerators,
while a Pentium 4 typically runs at 2.4–3.8 GHz with a single-core NetBurst architecture;
the Pentium 4 generally delivers higher single-threaded performance for general computing tasks.
[..]
The ESP32-P4 excels in specialized applications like IoT
and multimedia processing due to its dual-core architecture and hardware acceleration features.
In contrast, the Pentium IV is better suited for general computing tasks,
benefiting from higher clock speeds and single-threaded performance.
Each processor serves its purpose effectively within its intended domain.”
And I do support that view, I used to own a Prescott-based PC.
The hot-headed Pentium IV had an excellent single-core performance.
It had all the x86 SIMDs of its day, including SSE3.
DVD Playback worked like a charm, even with all the filters enabled.
It even ran Windows Vista without a hitch. 🥲
Why do you think that typing a question into an LLM creates some sort of info/comment that readers here would find at all useful? You obviously don’t have the expertise to answer the question yourself, so it stands to reason that you similarly wouldn’t have the expertise to determine the credibility of the LLM’s claim in the area. Check the docs, and keep the LLM use to yourself. (Hackaday does NOT need an LLM-generated comments-section).
Um, counter question: Why do you think I need to justify a quote, which I even labeled correctly?
AFAIK, there are no traditional benchmarking websites that compare performance of microcontrollers to PC CPUs. It’s apples and oranges.
PS: I didn’t even ask the for Google LLM for help, Google itself generated the text. One that was useful this time, even.
And it matches my existing understanding of the matter, so I’ve quotet it here.
Things that are reasonable or logical are so, no matter their origin.
And I think one should not automatically conclude from oneself to others.
Personally, I’m into vintage computing, am an ex Pentium IV user
and I read the articles about it a long time ago already.
The Pentium IV’s NetBurst architecture was a highly complex thing, a dead end in x86 history.
It had a very complicated pipelining and prediction scheme.
+1
As a 16-years retired senior engineer in IT architecture and an EE grad, I find Google’s AI comparisons very useful while realizing that the results may be flawed/tainted. I certainly AM NOT opposed to HAD generating posts that utilize AI in new creative ways including software generation: code is nothing but glue between the computing hardware and the machine’s output. I remember similar arguments about the use of “reusable” code libraries and that the “programmer” was compromised because s/he did not write the entire program and did not understand fully the library black-box except for the published APIs.
AI comparisons are not even in the same category as professional AI tools for software coding.
The computing world where I worked has changed and the microprocessor and microcomputing state of the art have advanced, thus it is often advantageous to have a non-biased (hopefully) viewpoint even if algorithmically crafted. With Google, there is a potential for the advertising $ide of the house-of-profit to sell methods to bias AI toward a biased conclusions, something that requires observation and awareness by the human.
As with any tool, AI can be useful or just used as a wocky hammer; it is up to the human to decide how the results are interpreted, which is why everyone should have some understanding and exposure to the use of AI starting early in formal education. Old farts are likely already firmly set in concrete with their impression of AI and are (sadly) unlikely to be swayed.
https://en.wikipedia.org/wiki/AI-induced_psychosis
Thanks, I’ve just read the summary generated by Gemma 4!
It was very interesting!
I’m surprised the bigger variants of microcontrollers don’t have a 128-way SIMD block or something for accelerating graphics operations. does it add too many transistors? does it need too much RAM to operate? I always thought IP blocks like the mali G31 were too small for SoCs with a quad A53 and would be better off drawing UI on 3D printer panels instead of web pages on android tablets.
My belief is that the silicon revolution in uC capability is linked heavily to the marketplace and not to the technology to produce complex chips.
As such, there is a perceived demand for a uC which has a lifecycle of potential demand over time which when viewed financially should create a positive cash flow for “n” units.
Add to the lifecycle the competition pressures from other manufacturers, there is always uncertainty in the market and competition to secure fad-time for chip manufacturing at a scale that minimize die cost. Thus, chip production is not steady but completed in fab runs of a fixed quantity to avoid stockpiling costs and to match production with pre-sales and anticipated demand.