The Nvidia GeForce RTX 4090 takes the pole position in our GPU benchmarks hierarchy and ranks as the fastest of the best graphics cards. It’s also expensive, and the initial supply sold out quickly. But assuming you can find 4090 cards in stock and can afford one, which should you buy? The MSI RTX 4090 Suprim Liquid X takes a slightly different approach from traditional cards, pairing a reasonably large 240mm radiator and all-in-one (AIO) cooling solution with Nvidia’s current reigning monarch. We have other 4090 cards on the way, but let’s see how the MSI RTX 4090 Suprim Liquid X stacks up to the competition.
Considering the fact that the latest CPUs ‘only’ use about 250W but recommend at least 240mm liquid cooling, you might be wondering how the MSI card can manage 450W. It does quite fine, because it’s not just about cooling capacity, it’s about thermal density. AMD’s Zen 4 chips can use 170W (more with boost), but that’s concentrated primarily in the compute chiplet that only measures 70mm^2. Intel’s Core i9-12900K pulls 125W and peaks at 250W, in a 215mm^2 chip. The AD102 GPU, by contrast, measures 608mm^2, so while the wattage is higher, the thermal density is quite a bit lower than either of those CPUs.
In general, it’s not too difficult to cool silicon that uses roughly 1 W/mm^2. AMD’s Zen 4 gets closer to 2 W/mm^2, while under load Intel’s CPUs are around 1W/mm^2. AD102 at 450W ends up around 0.75 W/mm^2. Of course, there are hot spots in the chips — cache doesn’t use nearly as much power as the various ALUs — but we’re not terribly worried about the cooling potential of the MSI Suprim Liquid X.
The real question? Is cooling with a 240mm radiator better than straight air-cooling solutions, though, like those found in the Asus RTX 4090 ROG Strix or even the reference RTX 4090 Founders Edition? Well, based on our testing results, the answer has to be no. Performance ends up basically tied with the Founders Edition, a card that theoretically costs $150 less.
That’s not to say the MSI card is bad. Getting a dual-slot card, albeit with the need to accommodate a 240mm radiator, is a nice change of pace from all the 3- to 4-slot models we’re seeing. It might just fit into a smaller case thanks to splitting off the cooling from the main card body. Plus, some people think liquid cooling just looks cooler, and depending on the street price, it might be worth the extra expense.
|Graphics Card||MSI RTX 4090 Suprim Liquid X||Asus RTX 4090 ROG Strix OC||RTX 4090 Founders Edition|
|Boost Clock (MHz)||2625||2610 (Gaming mode)||2520|
|2640 (OC mode)|
|VRAM Speed (Gbps)||21||21||21|
|TFLOPS FP32 (Boost)||86||85.5||82.6|
|TFLOPS FP16 (FP8)||688 (1376)||684 (1368)||661 (1321)|
We’ve completed testing on three RTX 4090 cards now, shown in the above table, but more are on the way. As we finish testing and reviewing those, we’ll put the performance results in our main Nvidia RTX 4090 review (on page 9). So if you’re stopping by to read this article a few weeks or months in the future, head over there if you want to see results from all the 4090 cards we’ve tested.
As is usually the case, the only differences on the specs sheet for the various cards are in the clock speeds, TBP (total board power) ratings, and the resulting impact on compute performance. The MSI card has the highest default clocks so far at 2625 MHz, a 4.2% increase over the reference card’s 2520 MHz boost clock. In practice, however, we should note that we typically saw clock speeds of 2750–2790 MHz while gaming on all three of the above cards.
If you want more information on what makes the RTX 4090 tick, check out our Nvidia Ada Lovelace Architectural deep dive. We also ran quite a few additional tests on the RTX 4090 Founders Edition, including professional content creation workloads and DLSS 3 testing. For our AIB card reviews, given the relatively limited differences we see in performance, we’ll focus more on the card design and aesthetics, as well as any other extras.