The caveat is that these are one writer's measurements on one machine. They are not lab-controlled results, and the numbers below should be read that way.
Why the 386H is the "small" Panther Lake iGPU
Intel's Panther Lake chips, now sold as Core Ultra Series 3, ship with two very different graphics tiers. The integrated graphics are based on the Xe3 architecture, but the 386H has only 4 Xe cores. The top X-series parts are a different story. The Ultra X9 has 12 Xe3 cores, whereas the Ultra 9 386H has only 4. The X-series chips carry the branded Arc B390 graphics.
That is where the article's "a third of the cores" framing comes from. It is a core-count comparison and says nothing precise about frame rates. Four cores versus twelve does not mean one-third the gaming performance, since clocks, memory bandwidth and power limits all matter.
Intel's own specification page for the 386H lists four Xe cores on the graphics side and labels the GPU "Intel Graphics" rather than Arc. Intel's footnotes add that Arc branding on H-series parts requires at least 16 GB of system memory in a dual-channel configuration. So a single-stick box like this one is not even eligible for the Arc badge, which makes it a fair stand-in for the worst case.
Third-party spec aggregators also note the 386H is commonly paired with a discrete GPU in gaming laptops. One of them advises that anyone wanting to game on integrated graphics should choose the X models. The MS-03 test is interesting precisely because it ignores that advice.
The test setup: one stick of RAM
The author tested on Windows 11 with the performance BIOS profile, mostly at 1080p. The main handicap was a single DDR5 SODIMM. An iGPU has no dedicated video memory and uses system RAM, so a single-channel setup cuts the bandwidth available to it.
Minisforum's product page lists the MS-03 with two SO-DIMM slots, so the one-stick configuration is a choice for this test unit. It is not a limit of the hardware. Minisforum lists support for up to 128 GB. The retail listing shown alongside the XDA piece describes 32 GB of DDR5-5600, but the article does not say how the review unit was configured beyond "one stick."
The reported results
All figures are author-reported and approximate:
| Game | Settings | Reported average |
|---|---|---|
| Black Mesa | 1080p native, low | about 70 FPS |
| GTA IV | 1080p, low | about 100 FPS |
| GTA IV | 1440p, low | about 60 FPS |
| GTA V Enhanced | 1080p, minimum preset (opening sequence) | about 90 FPS |
| Battlefield 6 | 1080p output, XeSS Ultra Performance, online Domination | about 35 FPS |
Older games
Black Mesa is the 2020 remake of Half-Life on Valve's Source engine. It ran at native resolution with no upscaling, which makes the roughly 70 FPS a clean data point.
GTA IV is the more interesting case. The 2008 PC port is notorious for running poorly on hardware that should handle it easily. The author notes that frame rate fell roughly in proportion to the extra pixels when moving from 1080p to 1440p. He reads that as a sign the GPU, not the CPU, was the limiting factor. That is a reasonable inference, but it is an inference and not a separately measured bottleneck analysis.
GTA V Enhanced
This was the headline result. The author ran the game's minimum preset and saw about 90 FPS through the opening sequence. He cautions that open-world gameplay in Los Santos could run worse. Even a one-third drop would still leave it near 60 FPS. The minimum preset and the opening sequence are both forgiving, so treat this as a best-case slice and not a full-game verdict.
Battlefield 6
This is where the small iGPU runs out of road. With XeSS set to Ultra Performance at 1080p output, the author saw about 35 FPS in an online Domination lobby.
The author points out what that mode means. Ultra Performance renders at one-third of the output resolution on each axis. At 1080p that is 640 by 360 internally, or one-ninth of the pixels, upscaled to the display. On his 32-inch 4K OLED, he said the image looked terrible. He judged infantry play unsuitable, with vehicle combat possibly more tolerable.
EA's Battlefield 6 FAQ confirms that XeSS 2 is among the supported upscalers, so the feature is legitimate. The same FAQ also needs a correction to the article's framing. EA's published PC minimum lists an Nvidia RTX 2060, a Radeon RX 5600 XT 6GB or an Intel Arc A380 for graphics, with 16 GB of RAM and Windows 10. The FAQ does not state a 30 FPS or 1080p target, and it does not mention a dual-channel memory requirement. The XDA piece's claims that the result "clears" an EA 30 FPS minimum and that the minimum calls for dual-channel memory are therefore not backed by the EA page. Treat them as the author's framing and not as official spec language.
One practical note for anyone considering this game on PC: EA's FAQ says Secure Boot is required, tied to its Javelin anti-cheat. That is separate from graphics performance, but it is worth checking before you blame your iGPU for a game that will not launch.
What the single-channel caveat does and doesn't mean
The author argues that these numbers are a floor and that a second memory stick would significantly lift performance. The theory is sound. Single-channel operation gives the iGPU roughly half the theoretical bandwidth of dual-channel at the same memory speed. But bandwidth does not translate directly into frame rate. The piece offers no before-and-after measurement, so any specific uplift is unproven here. Expect some improvement in bandwidth-bound scenes, but do not assume a doubling.
How far to trust the numbers
The test leaves out several things that would make it reproducible:
- No graphics driver version.
- No memory capacity, speed or timings for the test stick.
- No 1% low or minimum frame rates, only averages.
- No power or thermal logs. Minisforum lists Balance mode limits of 70 W PL1 and 80 W PL2, and the author says only that he used the "performance" BIOS profile.
- Some game settings beyond the stated presets.
Averages without frame-time data can hide stutter. A 90 FPS average that dips badly in busy scenes feels very different from a steady 90.
Bottom line
The supported takeaway is modest but useful. The four-core Xe3 iGPU in the Core Ultra 9 386H handled older and lighter titles well, even on a single memory stick. It struggled with a current AAA shooter, and only aggressive upscaling made Battlefield 6 run at all.
If you are speccing a mini PC or laptop, the practical lessons are:
- Check the graphics tier, not just the CPU name. A 386H and an X9 388H share a family but not a GPU.
- Populate both memory channels. It is the cheapest likely gain for any iGPU system, even if the size of the gain varies by game.
- Match expectations to the game. Older and esports-style titles are realistic targets. Recent AAA releases are not.
- Do not treat one reviewer's averages as guarantees. Treat them as a signal that integrated graphics have come a long way.