Retired Subway Menu PCs Become Gaming Rigs, but GTA V Runs at Just 15 FPS

Tom Hardware has documented a gamer turning three retired Subway menu computers into gaming rigs, despite GTA V reaching only about 15 frames per second. Minecraft managed roughly 25 fps, but its performance varied enough to challenge any claim that these machines became practical gaming PCs.

TechTuber Garkus rescued the computers after they had reportedly spent about a decade driving digital menu boards above Subway counters. He upgraded their memory and installed three different operating systems before testing games. The experiment preserved working equipment, yet it also exposed the difference between extending a computer's life and changing its intended job.

These were not conventional office desktops awaiting a better graphics card. They were thin, fanless Advantech signage players designed for continuous, predictable media playback. Their passive construction favored quiet operation and low maintenance, while their integrated Intel graphics placed a firm ceiling on demanding games.

Three Subway Menu PCs Get Three New Operating Systems

The experiment changed three specialized signage players into general-purpose computers, but it did not change their underlying performance envelope.

According to the original Subway PC test, Garkus received three computers and their matching LG digital signage displays. The systems still booted into Windows and entered kiosk mode, where they displayed Subway menu content.

Labels on the machines identified them as Advantech DS-081 units. More specifically, their processors matched the DS-081GB-U2A1E configuration, which uses an Intel Core i3-6100U. That is a sixth-generation Skylake mobile processor with two cores and four threads.

Each computer reportedly arrived with 4GB of DDR4 memory and a 64GB SATA solid-state drive. Garkus added another 4GB memory module, bringing the tested configuration to 8GB. This was a sensible upgrade because the processor's integrated graphics also use system memory.

The machines then followed three different software paths. One retained Windows 10, another received Bazzite Linux, and the third became a Hackintosh running macOS Monterey. A Hackintosh is a non-Apple computer configured to run macOS, usually through unofficial compatibility work.

That three-system arrangement made the project more interesting than a single game benchmark. It showed that retired commercial hardware can accept several familiar desktop environments. It also revealed how storage interfaces, drivers, and operating-system support affect reuse.

Garkus initially tried to install SteamOS on one machine. According to the report, that attempt stalled because the system relied on SATA storage. He then selected Bazzite, a Linux distribution built around a console-like PC gaming experience.

The operating-system variety should not be mistaken for a controlled comparison. The machines were similar, but the available reporting does not establish identical game versions, background activity, drivers, or thermal conditions. Those missing controls matter when frame rates are already extremely low.

The systems still completed an important transition. They went from playing fixed menu content to launching desktop software and games. That is real reuse, even if it falls well short of producing a dependable modern gaming computer.

The original gaming video presents the project as an experiment rather than a formal laboratory benchmark. Its entertainment value comes partly from uncertainty. Viewers see whether a machine built to advertise sandwiches can negotiate gaming workloads it was never expected to encounter.

That distinction frames the entire Tom Hardware story. Success means the computers avoided disposal and performed new tasks. It does not mean a memory upgrade converted them into suitable replacements for current consoles or gaming desktops.

The Advantech DS-081 Was Built for Uptime, Not GTA V

The DS-081's greatest strengths as a signage player become restrictions when a game demands sustained graphics performance.

Advantech designed the DS-081 as an ultra-slim digital signage player. Its official specification lists a chassis measuring 180 by 190 by 19 millimeters. The enclosure weighs 0.87 kilograms and uses a fanless thermal solution.

Fanless operation means the system moves heat through its chassis without an active cooling fan. That reduces noise and removes one mechanical component that can gather dust or fail. Those benefits are valuable when a computer operates behind a display for long periods.

The tradeoff is limited thermal headroom. A compact passive enclosure cannot remove heat like a large gaming case filled with fans. Its processor must remain inside a comparatively modest power range, especially during sustained workloads.

The Core i3-6100U runs at 2.3 GHz and pairs with Intel HD Graphics 520. The integrated graphics processor, or iGPU, sits within the same package and shares memory with the CPU. It does not have the dedicated video memory or computing resources found on a discrete gaming graphics card.

Tom Hardware highlighted a configurable processor power range between 7.5 and 15 watts. That modest target makes sense for commercial signage, where efficiency and reliability outweigh high frame rates. It leaves little room for intensive three-dimensional rendering.

Intel HD Graphics 520 contains 24 execution units and can reach a reported graphics frequency of 1 GHz. Raw specifications do not predict every game's result, but they explain why additional system memory cannot solve the central problem. More RAM reduces one bottleneck without creating more graphics cores.

Advantech nevertheless gave the DS-081 useful expansion and connectivity options. Its specification lists two DDR4-2133 SO-DIMM slots supporting up to 32GB. It also includes a 2.5-inch SATA III bay and a full-size mSATA connection.

Externally, the computer provides two HDMI 1.4 outputs, two Gigabit Ethernet ports, four USB 3.0 ports, audio connections, and an RS-232 serial interface. The dual HDMI outputs support a maximum listed resolution of 4096 by 2304 at 24 Hz.

That display capability illustrates a common specification trap. Driving a high-resolution menu or video does not imply that the same computer can render a complex game at that resolution. Decoding prepared media and generating an interactive scene are different workloads.

A digital menu typically displays static assets, simple animations, or compressed video. The workload is controlled, repeatable, and unlikely to change from one second to the next. A game must calculate player input, geometry, physics, visibility, effects, and frame composition continuously.

The DS-081 was also specified for operating temperatures between 0 and 40 degrees Celsius with stated airflow conditions. Its commercial design prioritized predictable deployment on a wall, desk, rail, or VESA mount. Nothing in that mission required responsive AAA gaming.

This is the primary conflict behind the experiment. Salvage supplies the computer at little or no acquisition cost, but the original engineering choices remain attached. A specialized low-power machine cannot escape those choices through software enthusiasm alone.

Tom Hardware Finds the Upgrade Ceiling at 15 FPS

Doubling the memory improved the platform's basic usability, yet graphics capacity remained the decisive constraint.

The easiest result came from Peggle, a lightweight two-dimensional puzzle game. The retired signage computer reportedly handled it without difficulty. That outcome established that the hardware could support casual gaming within an appropriately modest workload.

Garkus then moved to GTA V, which sharply increased the demand. The game reached only about 15 fps. It had reportedly defaulted to high graphics settings, required a long preparation process under Bazzite, and crashed when exiting.

Fifteen frames per second means the display receives a new rendered image about every 67 milliseconds. A stable 30 fps halves that interval, while 60 fps reduces it to roughly 17 milliseconds. Lower rates make movement feel less responsive and often expose uneven frame delivery.

The high default settings complicate the result. Reducing resolution, texture quality, shadows, population density, and distance scaling might improve performance. However, the published account does not include a standardized low-settings rerun that readers can reproduce.

GTA V is also an older title, but age does not make its workload trivial. Its open world places simultaneous demands on CPU processing, memory bandwidth, and graphics rendering. The HD Graphics 520 must share that limited memory bandwidth with the processor.

Minecraft later produced roughly 25 fps, although the rate reportedly fluctuated widely. The test did not disclose the resolution, render distance, quality selections, game build, or modification list. Those variables have a large effect on Minecraft performance.

The absence of settings prevents a clean comparison with another computer. It also makes “25 fps” less informative than it first appears. A quiet scene at a low render distance can behave very differently from a populated environment with new terrain generation.

The Tom Hardware account raises another significant testing issue. Garkus apparently ran Mario Kart Wii emulation while GTA V was still compiling shaders. Shader compilation converts graphics instructions into forms the GPU driver can execute, and it can consume processing resources.

Running another workload during that process makes the outcome entertaining but scientifically weak. It becomes difficult to separate the computer's normal gaming capacity from temporary background demand. The resulting numbers should therefore be treated as observations, not definitive benchmark scores.

Even so, the broad conclusion remains difficult to overturn. Moving GTA V from 15 fps to a consistently smooth experience would require a major performance increase. Settings changes can help, but the old iGPU, shared memory, and low-power processor still define the ceiling.

The memory upgrade demonstrates why component changes need context. Expanding from 4GB to 8GB gives the operating system and games more room. It can reduce paging, improve multitasking, and give the integrated graphics a less restrictive shared pool.

It does not double processor throughput or graphics execution resources. It also does not add dedicated video memory. Once the dominant bottleneck moves elsewhere, further memory expansion produces diminishing returns.

That is why “Can it launch?” and “Can it play well?” require different answers. The computers launched GTA V and Minecraft, which is impressive for repurposed menu hardware. Their frame rates still place both experiences below what most players expect from a dedicated gaming system.

Cheap Hardware Reuse Collides With Predictable Performance

The project succeeds as reuse precisely where it fails as a gaming conversion: it exposes which jobs fit the hardware and which do not.

Old computers often retain more utility than their original owners need. A device retired from commercial service can become a home server, media endpoint, network appliance, learning machine, or emulation box. Those uses can extend its practical life without demanding current gaming performance.

The DS-081 is especially interesting because it offers conventional x86 compatibility. Unlike a locked media appliance, it can boot familiar desktop operating systems and accept standard storage and memory. Its ports also support monitors, controllers, drives, audio equipment, and networking.

Those advantages give hobbyists room to experiment. Windows preserves broad software compatibility, while Bazzite offers a gaming-oriented Linux environment. macOS Monterey adds another technical challenge, although unsupported Hackintosh installations carry maintenance and compatibility risks.

However, a usable second life must fit the machine's fixed resources. The Core i3-6100U, HD Graphics 520, passive cooler, and compact chassis arrive as one system. Replacing enough of those elements would stop being a practical conversion.

An external graphics processor is not an obvious answer. The published DS-081 specification does not list Thunderbolt support, and its internal Mini-PCIe expansion was intended for other uses. Experimental adapters would add complexity, cabling, power requirements, and uncertain compatibility.

At that point, the project would trade the machine's compact simplicity for a fragile collection of parts. The environmental case would also become less clear if many new components were required to chase acceptable game performance.

Garkus concluded that the systems were better suited to emulating older consoles through the PlayStation 2 and original Xbox period. Emulation makes one computer imitate another platform, adding processing overhead beyond the original game's workload.

That suggestion needs cautious interpretation. Tom Hardware noted that PlayStation 2 emulation would probably require lighter games and reduced settings. Emulator behavior varies by title, so one successful game cannot establish compatibility across an entire console library.

Earlier systems offer safer targets. Many classic arcade, 8-bit, 16-bit, and early three-dimensional games demand far less from the hardware. Lightweight native PC games also avoid some of the overhead and compatibility complexity introduced by emulation.

Media playback remains another natural use because it resembles the machine's first job. The two HDMI outputs can support a dual-screen information display, workshop dashboard, exhibition player, or scheduled video station. These tasks align with the DS-081's original strengths.

Its two Ethernet ports create further possibilities. A technically experienced owner might use one as a network testing device, monitoring station, or lightweight home service host. Security updates and trustworthy software remain essential when repurposing any retired commercial computer.

The 64GB drive is restrictive for large game libraries but adequate for a lightweight operating system and focused application. Replacing it with larger compatible storage would improve flexibility. Storage expansion still would not address the gaming bottleneck.

The most sensible reuse plan begins with a workload, not the novelty of ownership. A person should ask whether the machine's processor, memory, storage, interfaces, and software support match a real need. If they do, old signage hardware can remain valuable.

That approach also avoids reducing sustainability to a slogan. Keeping one computer in service is useful when it replaces another device or performs meaningful work. Running an unsuitable machine at its limits simply to imitate a better system offers less practical benefit.

The Subway conversion therefore pressures a familiar online hardware narrative. Extremely cheap computers attract attention because their acquisition cost feels like the main constraint. In reality, setup time, missing accessories, software support, and performance stability can matter more.

The Benchmark Gaps Matter More Than the Headline Numbers

The reported frame rates establish a credible performance warning, but they do not form a reproducible comparison between operating systems or configurations.

Neither 15 fps nor 25 fps should be read as a precise universal result for every DS-081. They describe what appeared during Garkus's particular experiment. A rigorous test would document the operating system, driver version, game build, resolution, quality preset, and measurement method.

It would also control background processes and temperatures. Passive systems can change behavior as heat accumulates, especially during sustained rendering. Reporting clock rates, package power, and thermal throttling would show whether cooling reduced performance over time.

Frame-rate averages alone would still be incomplete. A proper comparison would include one-percent-low results, which summarize the slower frames that often correspond with visible stutter. Frame-time graphs would reveal whether a nominal average concealed severe inconsistency.

GTA V's high default settings create the largest immediate uncertainty. Fifteen fps at high settings is poor, but it does not establish the best playable configuration. Testing lower resolutions and presets would determine whether the machine can approach a stable 30 fps.

Minecraft presents a different uncertainty because its workload changes dramatically with configuration. Render distance, world generation, entities, graphics mode, and software edition all matter. The lack of these details prevents the result from serving as purchasing guidance.

The operating systems also introduce unequal conditions. Windows, Bazzite, and macOS use different graphics stacks, drivers, and compatibility layers. A game running through translation software can behave differently from the same title running natively.

Bazzite itself was selected after SteamOS reportedly rejected the SATA-only arrangement. That makes the Linux installation a practical workaround, not proof that Bazzite delivers better performance. Comparing it with Windows would require the same hardware, settings, game sequence, and measurement tools.

The Hackintosh build is even less representative. Apple did not design macOS Monterey for this Advantech model, and current compatibility depends on community boot tools and patches. A successful installation says more about platform flexibility than long-term reliability.

There is also no evidence that all three physical units performed identically. Commercial devices that operated for years can differ in storage health, thermal interface condition, firmware, and component aging. Those differences can influence stability even when model labels match.

The Tom Hardware report handles this limitation with appropriate skepticism. It notes the missing quality and resolution information, questions the benchmarking practice, and avoids presenting the numbers as controlled laboratory results.

That restraint matters because unusual hardware projects often spread faster than their methodology. A striking frame-rate number can become detached from its settings and conditions. Readers then repeat an observation as though it were a product specification.

The strongest conclusion is therefore narrower. These particular fanless Core i3 signage systems ran several games after modest upgrades and extensive setup. They did not show convincing evidence of smooth modern gaming.

That claim does not require perfect benchmarking. Peggle's success, GTA V's severe slowdown, Minecraft's inconsistency, and the platform specifications all point in the same direction. Better testing would refine the boundary rather than erase it.

What the Next Subway PC Tests Need to Prove

Three follow-up signals would determine whether these machines are useful gaming projects or mainly memorable salvage experiments.

The first signal is a controlled GTA V rerun. Garkus would need to publish the resolution, preset, driver, game version, and background workload. Tests should include the original high settings and a deliberately reduced configuration.

If the system reaches a stable 30 fps under reasonable visual settings, the current judgment would soften. It would indicate that configuration, rather than hardware alone, caused much of the poor first result. Continued performance near 15 fps would strengthen the existing conclusion.

The second signal is repeatable emulation testing across several console generations. Each game should use named emulator versions and disclosed rendering settings. Results should separate easy titles from demanding ones rather than assigning one capability to an entire platform.

Consistent performance across representative PlayStation 2 games would support Garkus's proposed use case. Frequent slowdowns or per-game workarounds would push the practical target toward older consoles and lightweight native games.

The third signal is sustained thermal and stability testing. A useful repurposed computer should run its chosen workload for hours without throttling, crashing, or corrupting data. Temperature, clock, and power measurements would expose whether the passive chassis remains comfortable during gaming.

That test is important because the DS-081 was designed for continuous signage operation, not continuous three-dimensional rendering. Its durable construction does not automatically guarantee full gaming performance under a different workload.

Owners should also consider software longevity. Windows 10 support status, Linux driver behavior, and unofficial macOS compatibility can all change the maintenance burden. A project that works today can become harder to secure or update later.

The most promising path remains a role that respects the original design. These computers can support older games, simple local multiplayer, media playback, dashboards, or focused network services. Such uses ask less from the integrated graphics while benefiting from compact, silent hardware.

Anyone inspired by the Tom Hardware experiment should begin by recording the exact model and testing one repeatable workload. Add memory or storage only after identifying the limiting resource. Then compare the result with the time, energy, and reliability required.

A rescued PC does not need to compete with a new gaming tower to justify its second life. It only needs a job it can perform consistently. The Subway machines already proved they can escape kiosk mode. The next question is whether owners will choose workloads that turn that escape into durable, useful reuse.