What each BC-250 actually holds — start here

[MTSistemi]

Every BC-250 came off the same line and no two behave the same. This category is where we collect what each one actually holds, so nobody has to guess a starting point.

What we know so far

Board	GPU	CPU	Compute units	Cooling
ours (development board)	2230 MHz @ 1000 mV	3700 MHz @ −150 mV	40 / 40	single blower
@cosmiccowboy68	2200 MHz @ 1000 mV	3500 MHz, no undervolt	40 / 40	two stock blowers
a third, reported by email	artefacts at 40 CUs even at 1100 mV	3500 MHz, no undervolt	38 / 40	stock

Two boards out of three hold all forty compute units. The third produces visible artefacts on the desktop at 40 and is stable at 38 — one WGP pair switched off. That is why the unlock is a switch you turn on rather than a default we impose.

How to add yours

The Tuner has a Share to the silicon DB button under My silicon, and after the Find my max wizard it offers the same thing. It opens a new discussion here with the values already filled in, including the kernel and the package version, and you read the whole thing before posting. Nothing else about your machine is collected.

By hand is fine too. What is useful:

• GPU maximum stable, MHz and mV — the pair matters, not the number alone
• CPU maximum stable, and the undervolt offset if you use one
• how many compute units hold, and which WGP you had to drop if not all forty
• how it is cooled, and what failed when you went further
• uname -r and the SkillFishOS package version

One thing worth knowing before you report

Artefacts that build up over seconds and then take the display down usually mean the chip ran out of voltage, not that a compute unit is dead — going from 24 to 40 CUs is roughly two thirds more shader hardware drawing current on an APU where the CPU and the GPU share one power budget. Rule that out first: put the CPU undervolt back to zero, then try one step more GPU voltage, then one step less GPU clock.

The driver declares the range it will accept for this chip — SCLK 350–2230 MHz, VDDC 700–1129 mV — and refuses anything outside it, which is where the Tuner's own limits come from. The 85 °C cap and the fan curve stay in charge of the heat.

If it still falls over at a comfortable clock, then it is the silicon, and the CU test in the Tuner lights each extra pair on its own and stresses it until something gives.