cancel
Showing results for 
Search instead for 
Did you mean: 
Silent_Scone
Super Moderator

 

The RTX 4090 wasn’t just powerful, it was a game-changer. My own ROG 4090 TUF OC has been an absolute beast, handling everything I’ve thrown at it without breaking a sweat. Whether it's 4K maxed-out gaming or heavy-duty content work, it’s rarely felt like I was even pushing it. The jump from the previous generation was so massive that, honestly, it felt like we’d hit a ceiling.

So when talk of the 5090 started, I’ll admit, I was curious, but skeptical. How do you top something that already feels like it’s in a league of its own? It wasn’t just going to be about raw speed; it had to be smarter, more refined. That’s exactly where the RTX 5090 steps in.

The previous generation RTX 4090 set a new benchmark for enthusiast-level GPUs, delivering unprecedented performance that reshaped expectations for gaming and content creation alike. It quickly established itself as the ultimate powerhouse, effortlessly handling the most graphically intense games and applications with remarkable ease. With its groundbreaking efficiency, powerful ray-tracing capabilities, and strong AI-driven features, the RTX 4090 was nothing short of a technological achievement.

Now, building upon the formidable legacy established by the RTX 4090, the RTX 5090 arrives, ready to elevate the gaming and benchmarking experience even further. The RTX 5090 enhances its predecessor's capabilities significantly through advancements in architecture, CUDA core count, clock speeds, improved memory technology, and enhanced thermal management.

Specifically, the RTX 5090 offers substantial upgrades over the RTX 4090, featuring 24,576 CUDA cores compared to the 4090's 16,384, representing a remarkable increase in raw processing capability. Clock speeds also see significant improvement, with the RTX 5090 providing a base clock of 2.5 GHz and a boost clock up to 2.9 GHz, versus the RTX 4090's base clock of 2.23 GHz and boost clock of 2.52 GHz. Additionally, the RTX 5090 introduces 32GB of next-generation GDDR7 memory, significantly surpassing the 24GB of GDDR6X memory found in the RTX 4090. The upgrade to GDDR7 boosts memory bandwidth substantially, greatly benefiting ultra-high-resolution gaming scenarios such as 4K and beyond, where increased bandwidth facilitates smoother performance, quicker texture loading, and better handling of complex scenes and high-resolution assets.

pd-back.png

A standout feature is its unique quad-fan layout, with the fourth fan acting as a rear-mounted "pull" fan to assist with airflow through the heatsink. This addition increases total airflow volume by approximately 20%, helping maintain higher boost frequencies under sustained loads while reducing acoustic output.

pd.png

The ROG Astral RTX 5090 is a showcase of thoughtful engineering, combining premium materials with a cooling and power delivery design built for sustained high performance. The cooler assembly is constructed entirely from metal and weighs nearly a kilogram, giving it a solid, industrial feel that's immediately reassuring. At the heart of its thermal solution is a large vapor chamber that draws heat away from the GPU core, memory modules, and VRM components. This is paired with eight heatpipes and a dense fin stack, all kept in check by a quad-fan setup—including a rear-mounted pull fan that improves airflow through the heatsink by roughly 20%.

 

 

 

 

Silent_Scone_0-1743141684558.png

 

 

 

To further improve thermal transfer, the Astral makes use of a premium phase-change thermal pad between the GPU and the cooling module. This material melts slightly under heat and conforms to microscopic surface gaps, dramatically improving contact and heat conductivity. The result is enhanced thermal dissipation under heavy loads, helping maintain consistent performance and long-term reliability.

 

Silent_Scone_0-1743142271067.jpeg

 

Under the hood, the power delivery system is no less impressive. The GPU is driven by a 24-phase VRM controlled by a Monolithic Power Systems MP29816 controller, with each phase relying on MP86670 DrMOS modules rated for 80 amps. That same class of power stage is used in the dedicated 7-phase memory VRM, which is managed by a Monolithic MP2898 controller and tasked with supplying clean, stable power to the card’s 16 GDDR7 chips. This tight integration between thermal and electrical design ensures the Astral is more than ready to handle the RTX 5090’s substantial power demands, while leaving room for overclocking headroom and long-term stability.

 

Silent_Scone_0-1743143592615.png

 

The ROG Astral RTX 5090 features a patented vapor chamber designed with precision-milled pathways that allow heat pipes to integrate seamlessly without flattening. This innovative design enhances heat dissipation efficiency, resulting in a thermal performance improvement of over 10%. By maximizing the contact surface area between the heat pipes and the vapor chamber, this approach ensures more effective heat transfer away from critical components, contributing to lower GPU temperatures and improved overall performance.

 

Power Detector+ Integration

Normally, verifying proper contact on a 12VHPWR connector requires physically checking each pin with tools like a current clamp, a tedious and impractical task, especially in tight PC cases. Power Detector+ completely removes that hassle by offering digital, real-time monitoring of every pin through software.

ASUS’s Power Detector+ offers real-time pin-level monitoring of the 12VHPWR connector, directly within GPU Tweak III. Users can add Power Detector+ to the software’s on-screen display (OSD), making it easy to keep tabs on connection integrity during gaming or stress testing. With visual indicators showing whether each pin is making proper contact, it eliminates the guesswork from high-power GPU installations.


GPU Tweak 3 OSD SettingsGPU Tweak 3 OSD Settings

 

Integration into the OSD is possible and will indicate if any improper amperage is detected on any of the 6 pins.

GPU Tweak 3 OSDGPU Tweak 3 OSD

 

Additionally, HWiNFO version 8.23 and later now support Power Detector+ telemetry, allowing users to incorporate this diagnostic data into broader system monitoring.

HWINFO v8.23HWINFO v8.23

 

 

GPU Tweak III & BIOS Profiles

Dual BIOS Profiles:

  • Performance Mode (Default): Prioritises performance.

  • Quiet Mode (Silent): Optimises for quieter operation by reducing fan speeds.

Silent_Scone_0-1744572438677.jpeg

 

GPU Tweak III Profiles:

  • Default Mode (Enabled via GPU Tweak III):

    • Boost Clock: 2560 MHz

    • Power Limit: 600W (max allowed by 16-pin 12V-2×6 connector)

    • Cooling Behavior: Balanced for performance, with a more aggressive fan curve to maintain lower temps


    OC Mode (Enabled via GPU Tweak III):

    • Boost Clock: 2610 MHz

    • Power Limit: Still capped at 600W

    • Cooling Behavior: Same as Default, but allows the GPU to push slightly higher frequencies within thermal and power constraints


    Silent Mode (Enabled via GPU Tweak III):

    • Boost Clock: 2550MHz 

    • Power Limit: 480W

    • Cooling Behavior: Fan curve is tuned for quieter operation

 

 

CS2 1440P.jpg

CS2 4K.jpg

Silent_Scone_1-1744436237082.png

CP2077 4K.jpg

Silent_Scone_0-1744436199390.png

Monster Hunter Wilds 4K.jpg

SOTR 1440P.jpg

SOTR 4K.jpg

WuKong 4K.jpg

WuKong 1440P.jpg

Steel Nomad.jpg

Time Spy Extremem.jpg

Port Royal.jpg

Power draw figures are averages

 

1440P

In Cyberpunk 2077 at 1440p RT Psycho settings with Frame Generation, we observed substantial performance gains through manual overclocking, achieving a significant boost from 182.14 FPS to 195.15 FPS, highlighting an approximate 7% increase. Notably, GPU temperatures remained well-managed across scenarios, with our custom OC significantly lowering temperatures due to increased fan speed. Interestingly, GPU power draw saw moderate increases, from 446W in Silent Mode to 469W with Custom OC at 1440p and up to 538W at 4K, indicating efficient scaling relative to the performance gains.

Monster Hunter Wilds at 1440p Max Settings further emphasized the RTX 5090's raw power, delivering a pronounced uplift in average FPS from 203.98 (Silent Mode) to 232.91 (Custom OC). This substantial 14% performance improvement comes with only a slight increase in GPU power draw, from 311W in Silent Mode to 317W in Custom OC, underscoring the GPU's efficiency.

Shadow of the Tomb Raider results demonstrate that even at extreme frame rates (378 FPS at 1440p Highest preset), there is minimal variance between Silent, OC, and Custom OC modes, indicating the a CPU bottleneck Power draw was similarly stable, hovering around 379W in standard modes and slightly lower at 369W under Custom OC due to optimized settings. 

 

4K

At 4K resolution, the ROG Astral RTX 5090 continues to demonstrate outstanding performance, flexing its muscle in the most demanding titles while scaling cleanly across Silent, OC, and Custom OC modes.

In Black Myth: Wukong, the card edges up from 89 FPS in Silent Mode to 97 FPS with Custom OC, with minimum framerates improving from 77 to 82 FPS. Power consumption rises modestly from 487W to 510W.

Shadow of the Tomb Raider at 4K Highest settings shows the card pushing well past the 300 FPS mark. Average FPS increases from 305 in Silent Mode to 317 in Custom OC, with a consistent climb in minimum framerate as well. Power draw scales from 511W to 543W, showing a modest power cost for a measurable performance uplift.

In Monster Hunter Wilds, we see more modest gains 150.27 FPS in Silent Mode up to 159.92 FPS with our custom OC but performance is impressively stable across all modes. Power draw rises just slightly, topping out at 395W, which reinforces how efficient the GPU is when fully utilised in well optimised games in tandem with upscaling methods.

Cyberpunk 2077, still one of the toughest titles on the market, scales well at 4K RT Psycho settings with Frame Generation enabled. The RTX 5090 climbs from 108.26 FPS to 115.38 FPS, while power draw stretches from 508W to 538W an expected increase for a game this taxing.

In Counter-Strike 2, a title that thrives on ultra-high framerates, the card flexes even harder, moving from 315 FPS up to 337 FPS. Minimum FPS also sees a notable jump, all while keeping power draw under control at 475W, reinforcing how well this card performs in competitive scenarios.

 

Thermal Performance and Cooling Efficiency

Across all tested scenarios, the ROG Astral RTX 5090 maintained excellent thermal performance despite the substantial power draw and aggressive overclocking. In Cyberpunk 2077 at 1440p, GPU core temperatures dropped from 66.8°C in Silent Mode to 63.5°C under our Custom OC, thanks to the increased fan profile. This trend held at 4K as well, with temperatures holding steady around 65.3°C under full load with the custom overclock, which his impressive given the nearly 540W power draw.

In Monster Hunter Wilds, thermal output remained modest even under heavy load, with Silent Mode temps at 61.7°C, and barely increasing to 62.2°C with our Custom OC and 60% fixed profile, showing that the Astral’s cooler has substantial headroom.

The synthetic benchmarks confirmed this pattern. Even under sustained workloads in Steel Nomad and 3DMark Time Spy Extreme, the cooler consistently kept temps in the low 60s without throttling.

Overall, the ROG Astral RTX 5090 offers outstanding performance potential, particularly when manually tuned, reinforcing its status as a high-tier enthusiast GPU ideal for gamers demanding peak performance in the most graphically intensive scenarios. Despite the impressive performance gains, the GPU maintains relatively balanced power consumption, demonstrating good efficiency even under demanding workloads.

 

OC Scanner

ASUS GPU Tweak III's OC Scanner is a user-friendly feature designed to automatically determine the optimal overclocking settings for NVIDIA GPUs. By initiating a series of stress tests, it identifies the maximum stable clock speeds your specific GPU can handle, accounting for the inherent variability between individual units. OC Scanner isn't just a shortcut for applying a fixed core clock offset — it's a smarter, adaptive tool that tailors the overclock to the GPU sample. While a manual offset simply adds a set frequency to the existing boost clock across the voltage curve, OC Scanner stress-tests the GPU and dynamically adjusts the voltage-frequency curve (V/F curve). It evaluates stability at different voltage points and applies a frequency increase where headroom exists, but leaves other areas untouched to maintain stability.

OC Scanner also applies a modest, conservative memory overclock by default, typically around +400 MHz. However, this is a predefined uplift and not the result of any real-time stress testing or validation of your specific GPU’s memory modules. Users seeking the maximum stable memory overclock should still perform manual tuning and testing.

 

ezgif-1c61d02de8ebd7.gif

ezgif-45da3fd18d038d.gif

Silent_Scone_0-1744400170937.png

Silent_Scone_1-1744400262586.png

Silent_Scone_2-1744400291838.png

 

Voltage Frequency Curve Overclocking

Voltage/Frequency (V/F) curve tuning in GPU Tweak III allows advanced users to control how the GPU boosts under load precisely. Instead of applying a flat offset, you can adjust the relationship between voltage and frequency at various points across the curve. This is useful for either overclocking—by pushing higher frequencies at a given voltage, or undervolting, where you maintain performance while reducing voltage and heat. GPU Tweak III’s Curve Optimizer interface provides a visual editor, letting you raise or lower frequency points individually or shift entire segments of the curve. It’s a powerful way to fine-tune stability, thermals, and efficiency beyond what automatic tools like OC Scanner can achieve.

Uncapped potential

For those curious about the upper limits of RTX 5090 performance, here is a Steel Nomad benchmark with power limits lifted (internal use). The ROG Astral RTX 5090 achieved a score of 16,633, marking an 8% increase over our already strong custom OC result of 15,339. However, that performance uplift came at the cost of an incredible 68% increase in power draw, highlighting the steep curve of diminishing returns when pushing well beyond NVIDIA’s official limits.

To achieve this, the GPU core was overclocked by 300+ MHz, and the memory received an aggressive offset exceeding +3000 MHz, contributing to an average core clock of 3,290 MHz. With fans locked in at 80% for the test, the GPU maintained an average of 67°C  with the aid of an AC unit. This test showcases the headroom available in the silicon and the Astral’s thermal and electrical design margin when unleashed beyond typical constraints. During the test, GPU power levels peaked at around 980W.

 

Silent_Scone_0-1744351815021.png

 

DLSS CNN vs. Transformer Models

Deep Learning Super Sampling (DLSS) has evolved significantly since its original introduction. NVIDIA's initial implementation relied on Convolutional Neural Networks (CNN), which analyzed frames using spatial data from nearby pixels to reconstruct a higher-resolution image. While effective, the CNN approach sometimes resulted in soft edges, subtle ghosting, or lost detail in complex visual scenes.

With DLSS's recent shift to a Transformer-based neural network, NVIDIA leverages a more advanced architecture capable of understanding spatial relationships and context over larger image areas. Transformers can more effectively interpret fine textures, intricate geometry, and subtle color variations, producing noticeably sharper and more stable images. This new model significantly reduces artifacts, preserves fine details, and delivers superior temporal stability, especially in challenging visual scenarios such as fast-moving objects, dense foliage, and text-heavy environments.

The following comparisons highlight the practical visual improvements provided by the Transformer-based DLSS over the traditional CNN-based approach.

 

CNN__Transformer.png

CNN__Transformer #2.png

The shift from CNN-based DLSS to the newer Transformer-based model brings noticeable improvements across the board. The Transformer model significantly enhances the fine-detail reconstruction, sharpening edges and textures that previously appeared to suffer from excessive aliasing or indistinctness, especially noticeable in foliage, text, and character details. The Transformer model greatly reduces ghosting and visual artifacts, resulting in cleaner transitions and improved stability during motion-heavy scenes. Whilst these changes are subtle, they are a welcome improvement towards the goal of seamless integration into the rendering pipeline. 

 

Wrapping Up the Astral Experience

The ROG Astral RTX 5090 not only surpasses the formidable legacy of its predecessor but also sets a new benchmark for enthusiast GPU performance. The unprecedented frame rates at ultra-high resolutions, sophisticated cooling design, and transformative visual enhancements from the latest DLSS technology represent a meaningful leap forward. However, even at its already substantial power draw, testing indicates that the RTX 5090 still has more performance headroom to offer, limited only by current power constraints. ASUS’s robust engineering comfortably supports this level of performance, highlighting the further potential if removing power limits. Overall, the ROG Astral 5090 OC has incredible cooling performance that's capable of handling way beyond what NVIDIA are willing to put on the table.

 

Resources: 

ROG Astral RTX 5090 OC Support Page

GPU Tweak III

How to overclock your GPU with GPU Tweak III

Futuremark - 3DMark Suite

 

2 Comments