Review of ROG Matrix reveals ASUS changed the design after announcement
The first ASUS ROG Matrix RTX 5090 review shows disappointing liquid metal application and great cooling efficiency but puts the price into question
German overclocker Roman “der8auer” Hartung has published what appears to be the first full review of the ASUS ROG Matrix GeForce RTX 5090, including benchmarks in its 800W power mode using an ASUS BTF motherboard. The card targets collectors and overclockers with a limited run of 1,000 units and a price around €4,000/$4,000. It celebrates 30 years of ASUS GPU designs and sits above the already high-end ROG Astral RTX 5090 series.
The Matrix card uses dual power inputs . A standard 12V-2×6 connector on the top edge provides up to 600 W. A second GC-HPWR style connector on the underside feeds power through a BTF motherboard slot. With both connected, the card exposes an 800W power target that is not available on regular RTX 5090 models. Without a BTF board the card behaves like a 600W air-cooled RTX 5090, regardless of its BIOS.
P-Mode (left) and Q-Mode (right), Source: Der8auer
Hartung confirms that ASUS’ per-pin current sensing works as intended. Under heavy load the GPU clock stays above 3,000 MHz while total power draw in “P” (performance) BIOS mode typically sits in the 680–730 W range. Current on individual 12V-2×6 pins remains lower than on an ROG Astral card, because part of the load is shifted to the BTF slot. That is the key safety difference versus users who flash the Matrix BIOS onto non-BTF boards and run all power through a single 12V-2×6 connector.
ASUS GC-HPWR power adapter, Source: Der8auer
The review also covers the first retail look at the final cooler. Compared to the Gamescom prototype, ASUS switched to nine-blade front fans and a revised rear fan, and uses an aluminium fin stack instead of the earlier full-copper heatsink idea, mainly to control weight and cost. The shroud exposes parts of the fin stack at the front through small slots, and the card keeps the four-fan layout with one rear fan pulling air through the dense heatsink.
Extremely quiet in Q-Mode, quieter than ROG Astral in P-Mode
In P mode the card pulls roughly 720W with the fans at about 1,600 RPM, and Der8auer measured around 46 dB. He reports this as about 6 dB quieter than his ROG Astral RTX 5090 sample while drawing roughly 120W more. Coil whine is present but described as low to mid level and acceptable for a card in this class. In Q (quiet) mode the power limit drops to 600W, clocks fall by 150–200 MHz, fans slow to around 1,400 RPM, and noise drops to about 36dB , which makes the card hard to hear in a normal open bench setup.
[CAPTION_STARTSource: Der8auer
Overclocking and performance
In 3DMark Speedway, the card improves from 158 FPS stock to about 163 FPS after a simple OC with maxed voltage slider, +160 MHz on the core and maximum memory offset in GPU Tweak III. In gaming tests, the Matrix can edge out the RTX Pro 6000 in some titles, such as Assassin’s Creed Mirage at 150 FPS, but trails in others, like Star Wars Outlaws at 98 FPS while consuming more power. Compared to the RTX 5090 Founders Edition, Hartung sees around 10% higher frame rates at the cost of about 26%.
[CAPTION_STARTSource: Der8auer
The review also touches on the early Matrix BIOS that has already appeared on TechPowerUp . Enthusiasts have flashed it to other RTX 5090 models such as AORUS Master with some success, but it does not work on ASUS Astral cards. Hartung warns that running a Matrix BIOS on a non-BTF card pushes the full extended power limit through the 12V-2×6 connector alone, which increases the risk of connector damage. The Matrix is designed to split load between the top connector and the BTF slot, which is not possible on standard boards.
[CAPTION_STARTSource: Der8auer
Der8auer took the card apart to inspect ASUS’ liquid metal implementation. The Matrix uses liquid metal on the GPU die with a paste “frame” around the package instead of a glued sealing ring. He found a large amount of liquid metal squeezed out to the sides and an unusual contact pattern on both the die and cooler base. Temperatures during testing were still within expected ranges, yet he describes the application as below what he would expect from a €4,000 flagship and plans a follow-up with a re-application test.
Source:
