NVIDIA GeForce RTX 5090 Founders Edition Review – An Engineering Feat Marred By Blackwell’s Shortcomings

At a time where NVIDIA is valued at $3 trillion, the RTX 50 series aka “Blackwell” is officially introduced – and it shouldn’t surprise anyone that the focus on this generation is none other than AI itself. Still, as the RTX 4090’s successor, the GeForce RTX 5090 Founders Edition we’re looking at today does has quite some changes inside and out, so let’s take a close look.

Unboxing

The unboxing experience is quite different than before, now coming in a rather nondescript box that reveals the inner packaging housing the GPU itself. This does remind me of Apple’s product packaging design, and the inner box certainly can work as a decoration if you do own one of these cards.

NVIDIA provides the new 12V-2×6 to 4x PCIe 8-pin adapter along with the RTX 5090, and it comes with several changes: one, the cables are now based on paracord materials (like the ones you find in high-end mice) which gives it better flexibility in tight spaces; secondly, the 16-pin connector is also reinforced with bigger plastic housing to maintain a safe bending radius.

Walkaround

The biggest change RTX 5090 made compared to the RTX 4090 is the cooler design, and there’s quite a lot to talk about here. In both RTX 3090 and RTX 4090 FE cards, NVIDIA employed a push-pull fan system that combines flow-through and blower-style airflow to keep the cards cool; this time around, the RTX 5090 uses a double flow-through design thanks to the new ultra-compact PCB that sits in between both fans.

That design also allowed the card to shrink down to two slots despite needing to cool a whopping 575 watts, and it fits right within the dimensions as specified in NVIDIA’s own “SFF-Ready Enthusiast GeForce Card” standard. Like the predecessor, the 12V-2×6 connector is located at the top angled rightward, capable of delivering up to 600W through a single cable.

Continuing with the minimalist design aesthetic (and the new cooler not requiring exhaust to be placed sideways), the I/O is simply a flat piece of PCIe bracket with four display output ports: three DisplayPort 2.1, and one HDMI 2.1. When was the last time you see a flagship GPU being this thin?

Similar to predecessors, the lighting consists of the ‘X’ pattern at both side of the card, along with the ‘GeForce RTX’ wordmark at the top. All lighting zones are white, with no native customization present (though third-party RGB customization might work).

Specifications

NVIDIA GeForce RTX 5090 Founders Edition

Full specifications available on product page.

GPU Core & VRAM: NVIDIA GeForce RTX 5090
GPU Core Variant	GB202-300-A1
Microarchitecture	Blackwell
Process Node	TSMC 4NP (enhanced 5nm-class)
Transistors	92.2 billion
Die size	750mm²
Streaming Multiprocessors (SMs)	170
ROPs / TMUs	176 / 680
CUDA Cores	21760
Tensor Cores	680
RT Cores	170
Cache	96MB L2
VRAM Configuration	32GB GDDR7 512-bit
Memory Bandwidth	28 Gbps, 1792 GB/s peak
Add-in board: NVIDIA GeForce RTX 5090 Founders Edition
Fan Layout	Dual flow-through axial fans
Base/Boost Clocks	2017 / 2407 MHz
TDP (TGP)	575W
Factory Recommended PSU	1000W
Dual BIOS Mode	No
Display Outputs	3x DisplayPort 2.1b 1x HDMI 2.1b *Max output resolution: 7680×4320 (8K)
Power Connector	1x 12V-2×6 (12VHPWR) connector
Bus Interface	PCIe 5.0 x16
Dimensions	2-slot, 304 x 137 x 40 mm

Test System

CPU	Intel Core i9-13900K
Cooling	Cooler Master MasterLiquid PL360 Flux 30th Anniversary Edition Thermal Grizzly Kryonaut
Motherboard	ASUS ROG Maximus Z790 Apex
GPU	> NVIDIA GeForce RTX 5090 Founders Edition
Memory	Kingston FURY BEAST RGB DDR5-6800 CL34 (2x16GB) *configured to DDR5-6400 CL32 XMP profile
Storage	ADATA LEGEND 960 MAX 1TB
Power Supply	Cooler Master MWE Gold 1250 V2 Full Modular (ATX12V 2.52) 1250W
Case	VECTOR Bench Case (Open-air chassis)
Operating System	Windows 11 Home 24H2

Performance

All benchmarks are done in out-of-the-box settings under Performance Mode – for gaming benchmarks, upscaling and frame generation features such as NVIDIA DLSS, AMD FSR, or Intel XeSS are turned off unless otherwise specified in the chart.

Synthetic Benchmarks

Starting off, the RTX 5090 FE’s performance doesn’t indicate a big jump from predecessor like previous-gen GPUs do, as the performance uplift across the board is generally moderate. That said, the card does scale better at higher resolutions and heavier scenarios, such as Time Spy Extreme and Steel Nomad. The more interesting part is the difference between RTX 5080 and RTX 5090: despite the latter doubling in cores, performance doesn’t scale to the same extent, though 50-60% faster over RTX 5080 is still quite a gap.

(Note: the ROG RTX 4090 has a slightly higher TGP at 500W – 50W more than stock – which means performance is a few percent higher than the stock figures.)

Gaming Benchmarks

The RTX 5090 is the fastest GPU in town, capable of delivering well over 60FPS at the most graphical-heavy games short for Black Myth: Wukong at native resolution, which is one of the most difficult games to run without any upscaling assistance. Generally, it’s best suited for 4K resolutions, but some games at 1440p can still make use of RTX 5090’s available compute power pretty effectively before hitting the CPU bottleneck.

Still, it’s worth noting that for all the power draw it gained, the gen-on-gen improvement is lackluster this time around. If you already have, say, the RTX 4090, upgrading to the RTX 5090 simply doesn’t make sense (unless you really want the Multi Frame Generation feature), and the power efficiency improvements, based on our observations during testing, has be minimal at best.

Thermals

With all 575 watts of power packed into a card just two slots thick, it wouldn’t be surprising that thermals will be pretty close to the limit – still, the fact that this is even possible in the first place is a testament to the FE card’s engineering, which is simply impressive from a thermal management standpoint. In a hypothetical scenario where a workload maxes out the GPU like Furmark, the core temperature managed to stay below 80°C throughout the 10-minute test, while Time Spy Extreme didn’t manage to max out the power draw, resulting in lower temperatures.

All of the cooling is done by the dual flow-through fans which is virtually silent throughout our testing. You’ll have to be listening closely to barely notice the low-pitch ‘whoosh’ noise, and it’s safe to say you won’t hear much when it’s installed in a PC case. And it’s still a dual-slot card!

Software

For tuning, the FE card relies on the new NVIDIA App that finally includes the GPU tuning section, so you won’t need to install things like MSI Afterburner to get things going. The app also includes a one-click overclock scan, which takes a few minutes to complete – once done, it’ll give you the best overclock available to the specific GPU that you have, depending on how lucky you are in terms of silicon lottery.

Verdict

The NVIDIA GeForce RTX 5090 Founders Edition comes with an eye-watering price of RM10,390, assuming the card stays on MSRP (which, it most likely won’t at this time). For the price, the key selling point at this moment is the compactness of the card, which is an impressive engineering feat on Team Green’s part – so if you have a SFF PC build in mind, this card might be the ideal option if maximum power is what you’re looking for.

But before you hit the purchase button (assuming you’ve got one locked in), here’s a few things worth pointing out. First, the gen-on-gen performance is not to the level we’ve seen in recent generations, making Blackwell pretty hard to justify; second, adding on top of that, the only major upgrade you’ll get is more VRAM (now 32GB GDDR7), and Multi Frame Generation, which now enables up to 4x more frames instead of just 2x in the conventional Frame Generation already found in RTX 40 series cards. In either case, they won’t provide any improvement in input responsiveness if that’s what you’re looking for.

There’s also one specific group of users that RTX 5090 is potentially aiming for: those who work with AI models. 32GB of VRAM is plenty for some pretty large models, and this is a capacity that used to be strictly on workstation cards that are magnitudes more expensive than even this GPU’s price tag. There’s a lot of raw compute horsepower that even games aren’t quite able to extract (the GPU almost never draws full power in games), but local AI models is perhaps quite capable of tapping into that.

So, is it a worthwhile purchase if you have the money? I’d say, unless you’re the niche I just mentioned, you probably don’t need this card, and it’s probably not worth your money either. If you want a big jump in gaming performance specifically, I suggest wait until RTX 60 series comes out – if that delivers, of course.

Special thanks to NVIDIA for providing the GeForce RTX 5090 Founders Edition GPU for this review.