Nvidia GeForce RTX 4090 Review
This is our first look at Nvidia's GeForce 40 series, starting with the flagship GPU, the GeForce RTX 4090. Today we're about to find out all you need to know about this next generation graphics card – most importantly its gaming performance.
The new GeForce RTX 4090 is baked on the AD102 silicon which measures 608.4 mm2, about 3% smaller than the GA102 used by the RTX 3080 and 3090. Nvidia has moved from the Samsung 8N process with Ampere to TSMC's 4N process for Ada Lovelace. Quite incredibly, this has seen transistor count increase by 170% from 28.3 billion to an insane 76.3 billion.
When compared to the RTX 3090 Ti, there's 52% more streaming multiprocessors, CUDA cores, Tensor cores and RT cores and texture units. The ROP count has increased by 57% and the boost clock has been wound up by 35%. The same 21 Gbps GDDR6X memory is in use and there's still 24GB of it on a 384-bit wide memory bus, resulting in the same 1008 GB/s of memory bandwidth. The GPU also keeps the same the PCI Express 4.0 x16 interface.
Nvidia claims a total graphics power rating of 450 watts for the RTX 4090, the same rating given to the 3090 Ti, though the maximum GPU temperature has been slightly downgraded from 93C to 90C. The minimum power supply requirement is 850 watts, which is what we'll be using for all of our testing.
In terms of design, the Founders Edition 4090 looks similar to the 3090 with some fairly significant changes. The most noticeable is the graphics card's width. On paper, Nvidia claims a triple-slot form factor for both models which is accurate, but where the 3090 measures 52mm wide, the new RTX 4090 is 17% wider at 61mm. Despite the size increase, both models weigh about the same at 2190 grams.
That's surprising, but what's not all that surprising is the absence of the NVLink connector which Nvidia has now killed off and will instead rely on the PCIe 4.0 bus. The only other major change is the 16-pin power input which has been upgraded to the PCIe 5.0 spec, otherwise known as the super catch 12VHPWR power connector.
A single PCIe 5.0 power connector can deliver up to 600 watts, whereas previously that would require four 8-pin power connectors. You won't require a new PCIe 5.0 compliant PSU though as the RTX 4090 comes with a 4x 8-pin to single 16-pin adapter, similar to the 3x 8-pin to 16-pin adapter supplied with 3090 Tis.
We should note that it was falsely reported that the 12VHPWR power connector could only survive 30 cycles (30 connect and disconnects), but that's not the case and the connector longevity will be similar to that of the 8-pin connectors, which is to say you're never going to wear one out.
Besides the increase in cores, inclusion of 4th-gen tensor cores and 3rd-gen RT cores, GeForce 40 also introduces DLSS 3, a feature that for now is exclusive to the GeForce 40 series.
This new upscaling/frame rate multiplying technology is exciting and while we'll briefly show some results in this review, our full analysis of DLSS 3 is coming up soon. DLSS 3 requires a significant amount of testing and analysis which is far beyond the scope of a day-one review.
For testing, all GPUs were set to run at the official clock specifications (no factory overclocking), the CPU powering the test system is the Ryzen 7 5800X3D with 32GB of dual-rank, dual-channel DDR4-3200 CL14 memory on the MSI MPG X570S Carbon Max WiFi motherboard.
Benchmarks
Starting with Watch Dogs: Legion at 1440p, the GeForce RTX 4090 doesn't look that impressive here (...wait for it). Sure, it's the fastest GPU we've ever seen but a 9% boost over the 6950 XT is hardly awe-inspiring.
We saw a more substantial 22% uplift over the RTX 3090 Ti, but given today's GPU prices that's not amazing. Having said that, the issue appears to be less with the RTX 4090 and more with the 5800X3D which has become the primary system bottleneck...
Therefore when increasing the resolution to 4K we get to see how brutally fast the RTX 4090 really is. We're now looking at a monstrous 60% boost over the 6950 XT and a 64% uplift from the RTX 3090 Ti, that's unbelievable and a truly high refresh rate experience at 4K.
It appears as though we have a lot more CPU headroom in Rainbow Six Extraction at 1440p and so the RTX 4090 is 59% faster than the RTX 3090 Ti and 71% faster than the RTX 3090 and 6950 XT. This is a serious performance uplift.
Jumping up to 4K didn't change the results radically as we weren't CPU limited at 1440p. The RTX 4090 was once again ~60% faster than the 3090 Ti, though the margin against the Radeon 6950 XT kept growing and now we're looking at a 102% deficit here for AMD. That's brutal to say the least.
Performance Summary: 13 Game Average
Here's a look at the 13 game average which is calculated using the geomean. The GeForce RTX 4090 pumped out 219 fps making it on average 44% faster than the 6950 XT and 45% faster than the 3090 Ti.
Those are huge margins but still under-represent the RTX 4090 as the 5800X3D (a pretty fast gaming CPU) was limiting performance on multiple occasions.
It's crazy to think that one of the fastest gaming CPUs out there can be a serious bottleneck for the RTX 4090 at 1440p, and in many cases we're using the highest visual settings for testing.
The GeForce RTX 4090 is on average 59% faster than the RTX 3090 Ti and 71% faster than the 6950 XT at 4K gaming. Those are massive margins. Just as impressive is the fact that the RTX 4090 averaged 145 fps at 4K. This is truly the first 4K GPU capable of delivering a high refresh rate gaming experience. Very impressive stuff by Nvidia.
We haven't touched on 1080p results for the 13 games tested, though we did gather all that data even if the results were often heavily CPU limited. Overall, the RTX 4090 was 28% faster than the 3090 Ti at 1080p using the Ryzen 7 5800X3D and 24% faster than the 6950 XT, so it's still faster and if anything, a great tool for testing next-gen CPU performance.
Ray Tracing Performance + DLSS
For ray tracing and upscaling results we'll start with F1 22. At 1440p, the RTX 4090 was just 11% faster than the 3090 Ti and 24% faster than the 6950 XT, when comparing standard rasterization performance (ultra high quality preset but with RT effects disabled).
With RT + DLSS enabled the RTX 4090 was 17% faster than the 3090 Ti and 29% faster than the 6950 XT which was using FSR for upscaling.
Comments