NVIDIA GeForce RTX 5090 Founders Edition Review: Still King Of PC Gaming
NVIDIA's GeForce RTX 5090 may be the fastest GPU on the planet at this time, but there's always some wiggle room to tweak things and further improve performance. With that in mind, we spent a little time trying to overclock the GeForce RTX 5090 to see what kind of additional performance we could wring from the card.
Overclocking The Blackwell-Powered GeForce RTX 5090
Like previous-gen GeForces, new RTX 50 series cards like the GeForce RTX 5090 feature GPU Boost, which scales frequencies and voltages dynamically within pre-determined power and temperature limits, based on the GPU's workload and other environmental factors at the time. Should a temperature or power limit be reached, GPU Boost will drop down to the previous boost frequency / voltage stepping, to bring power and temperatures down gradually and not cause any significant performance swings. If there is power and thermal headroom available, however, the GPU will boost as high as it can for as long as it can to maintain optimal performance. The RTX 50 series can dynamically switch frequencies and enter / exit power states much more quickly than previous-gen GeForces, but ultimately behaves in a similar way.
When overclocking the 5090, the maximum boost clocks, memory clock, GPU voltage, and power limit can be tweaked to increase performance, as is the case with most GeForce cards.
With its bleeding-edge cooling solution, Blackwell's power-saving Max-Q technologies, and its high-performance liquid metal TIM, the GeForce RTX 5090 runs relatively cool and quiet at stock settings. According to the NVIDIA App, the GPU topped out at only 67°C with nearly 100% GPU utilization over an extended period of time, and the cooling fans only spun up to about 1400 RPM. In its stock configuration, the GeForce RTX 5090 is relatively tame in terms of temperatures and fan noise -- the fans actually spin down and are silent while idle. The fans are definitely audible while gaming though, but not what we'd consider loud, especially when enclosed in a chassis. Like most other high-end GeForces, the GeForce RTX 5090 is typically voltage or power limited while gaming, and it is those limits that will usually be the gating factors in any overclocking efforts.
The NVIDIA App takes a lot of the grunt work out of overclocking by offering a couple of sliders and an Automatic Tuning function that can dial in the power limit and various frequencies with just a few clicks.
We didn't have a lot of time to manually tweak our GeForce RTX 5090 and still pull off this article, so we enlisted the help of NVIDIA's auto tuning option and upped the card's power limit in an attempt to boost performance. Auto Tuning will take about 30 - 45 minutes to complete as the app methodically tests an array of frequencies at various voltages over the entire curve, but in the end, you'll have optimal, if somewhat conservative, increased GPU and memory frequencies that shouldn't affect stability. Ultimately, our particular card was comfortable with +99MHz on the GPU frequency, with a 200MHz boost to the memory after the auto tuner did its thing.
With the GeForce RTX 5090, the power target can also be increased by up to 4%. The maximum GPU voltage can also be increased with a simple slider.
After tweaking those options in the NVIDIA App, we assumed the card was poised to hit higher frequencies and that it had some more power headroom to play with, but things didn't play out quite that way. The GPU clock didn't seem to boost as high as the stock configuration and power consumption (as reported by the app) was somewhat lower. Now, if the auto tuner was able to under-volt the GPU slightly, lower power consumption would be expected and welcomed, since that would translate into additional power headroom, but when we ran some benchmarks with these tweaked settings, performance didn't change much.


After upping the power target and running NVIDIA's auto tuning option, we re-ran a couple of benchmarks and technically saw some increased performance, but the differences were minimal. We're not absolutely certain the NVIDIA App is performing as intended with the GeForce RTX 5090 just yet (we have asked NVIDIA for input), or if these increases were just due to variability in the benchmarks, but are reporting these numbers for the sake of completeness for now. We reserve the right to revisit RTX 5090 overclocking in the future. :)
Total System Power Consumption
Throughout all of our benchmarking and testing, we also monitored noise output and tracked how much power the GPUs were consuming in our test system. Our goal here is to give you an idea as to how much power each GPU used while idle and also while under a heavy workload. These power numbers were captured during MLPerf Client and Horizon Zero Dawn Remastered testing, and are representative of total system power consumption, not the GPUs on their own.
NVIDIA increased the TGP of the GeForce RTX 5090 by 125W over the RTX 4090 and that absolutely shows in our power consumption tests. The GeForce RTX 5090 consumed the most power by far under load. Idle power was also somewhat higher, but not much of concern.
These kind of power consumption numbers don't tell a good efficiency story for Blackwell and the GeForce RTX 5090, but for the people shopping for this class of product, we're not sure it'll matter much. In practice, the GeForce RTX 5090 is the fastest GPU on the market, and it runs relatively cool and quiet, which is usually what matters to most hardcore enthusiasts. It's just not the most efficient GPU comparably, and it consumes the most power.
GeForce RTX 5090 Review Summary & Conclusion
NVIDIA's GeForce RTX 5090 is the fastest, most powerful, and feature-rich consumer GPU in the world as of today, period. There’s no other way to put it. The NVIDIA GeForce RTX 5090 Founders Edition card itself is also a refined piece of hardware. To design a card that offers significantly more performance than an RTX 4090, at much higher power levels, in a roughly 33% smaller form factor is no small feat of engineering. The card also looks great in our opinion. On its own, the GeForce RTX 5090 is currently unmatched in the consumer GPU market – nothing can touch it in terms of performance, with virtually any workload – AI, content creation, gaming, you name it.It's not all sunshine and rainbows, though. In many cases, the previous-gen GeForce RTX 4090 offered nearly double the performance of its predecessor (the RTX 3090) when it debuted, at lower power, while using the exact same settings and workloads. However, if you compare the GeForce RTX 5090 to the RTX 4090 at like settings, the RTX 5090 is “only” about 25% - 40% faster and consumes more power. The RTX 5090’s $1,999 MSRP is also significantly higher than the 4090’s $1,599 price tag. Considering the Ada and Blackwell GPUs at play here are manufactured on the same TSMC process node, NVIDIA was still able to move the needle considerably, but the GeForce RTX 5090 doesn’t represent the same kind of monumental leap the RTX 4090 did when it launched, if you disregard its new rendering technologies at least.
You can’t disregard those new capabilities, though. Neural Rendering, Mega Geometry, DLSS 4 with multi-frame generation, the updated media engine, and all that additional memory and memory bandwidth all have to be taken into consideration. When playing a game that can leverage Blackwell’s new features, the GeForce RTX 5090 can indeed be more than twice as fast as the RTX 4090.
The use of frame generation has spurred much discussion since its introduction, and we understand the concerns regarding input latency and potential visual artifacts that come from using frame-gen. But the fact remains, using AI and machine learning to boost game and graphics performance in the most effective and efficient way forward at this time. Moving to more advanced manufacturing process nodes doesn’t offer the kind of power, performance and area benefits it once did, so boosting performance must ultimately come mostly from architectural and feature updates. And everyone in the PC graphics game is turning to AI. We specifically asked about the importance of traditional rasterization moving forward and were told development is still happening, and it will remain necessary for “ground truth” rendering to train the models, but ultimately AI will be generating more and more frames in the future.
While that may not sit well with everyone, keep in mind, the problems currently associated with today’s upscalers and frame generation are not insurmountable. With each new update, visual quality continues to improve. And there will be ways to better address input latency concerns as well.
We could spend multiple pages debating the future of rendering and PC gaming, but as of today, the current "ground-truth" is that the GeForce RTX 5090 sits unchallenged at the top of the consumer GPU market, and nothing can touch its performance in virtually every workload. If you’ve got the budget and are shopping for the current king of the GPU hill, the GeForce RTX 5090 is it.
