Quick Links
Graphics card releases always generate a lot of hype, and the rumor mill usually begins churning out content a year before they're released. Well, the official word on NVIDIA's RTX 4000-series is finally available --- here are some important takeaways if you're looking to upgrade.
GPU Launches Are Always Exciting
The launch of the RTX 3000-series was marred by bad timing and bad luck: a global supply chain disruption, combined with a hot cryptocurrency market, drove up prices and hurt availability. It was hardly possible to get a 3000-series GPU at MSRP until nearly two years after their initial release.
Limited supplies and inflated prices have compelled gamers to hang on to their GPUs longer than they might normally, and the general sentiment leading up to NVIDIA's GTC has been "I'll wait for the 4000-series to release before I decide."
The rumors haven't done anything to hurt hype either --- a series of reasonably credible leaks suggested that NVIDIA's flagship Lovelace device, dubbed the RTX 4090, packs some serious punch. That means the lower-tier devices, like the 4060, 4070, and 4080 will probably wipe the floor with their 3000-series counterparts in real-world performance.
So, hype and rumor aside, what do we know for sure after the GTC Keynote?
Welcome NVIDIA's RTX 4090 and RTX 4080s
Leaks and rumors are one thing, but now we've finally got some concrete numbers from NVIDIA. The RTX 4090 and 16 gigabyte (GB) 4080 are both monsters --- they're 11.97 inches long, 5.4 inches wide, and take up three entire slots in your case.
What Are the RTX 4090's Specifications?
We know for sure that the RTX 4090 will have:
- GPU Memory: 24GB of GDDR6X
- CUDA Cores: 16,384
- FP32: 90+ TFLOPs
- TDP: 450 Watts
- MSRP: $1,599
The RTX 4090 has launched with a price tag of $1,600. That is $100 more than the RTX 3090, which debuted at $1,500, and $500 more than the going price of a 3090 Ti in September 2022. Considering the performance jump, the additional hundred dollars doesn't hurt so badly, especially considering many feared the RTX 4090 would wind up priced at about $2,000.
The RTX 4090 becomes available October 12th, 2022.
What Are the RTX 4080's Specifications?
Two different RTX 4080 GPUs are being released. Despite sharing a name, there are some pretty critical differences between the two variants. They probably wouldn't both be called 4080s in any other generation of cards.
The beefier 4080:
- GPU Memory: 16GB of GDDR6X
- CUDA Cores: 9,728
- TDP: 320 Watts
- MSRP: $1,199
The lighter 4080:
- GPU Memory: 12GB of GDDR6X
- CUDA Cores: 7,680
- TDP: 285 Watts
- MSRP: $899
NVIDIA's website doesn't give any physical dimensions for the 12GB 4080 and instead only says it "varies by manufacturer." That may mean there won't be a Founders Edition available.
The pricing is a noticeable jump from the RTX 3080 (10GB), which originally sold for $699. The change may be a hard pill to swallow for a lot of gamers and push more people towards the future RTX 4070 card.
The two RTX 4080s are being released in November.
Video Memory is Good, but It Isn't Everything
Modern gaming can easily use up 6 to 8 gigabytes of video memory, especially if you're gaming at 4K. When you're looking at GPU memory usage it is important to note that applications will often reserve more memory than it is actually using just to be safe.
That means that when you see your 3060 or 3080 with 12 gigabytes of VRAM sitting there with 10 gigabytes in use, you probably don't need to worry --- you've got more headroom than you think you do.
The RTX 4090 has 24 gigabytes of video memory, and the 4080 will be available with 12 or 16 gigabytes. The real question: Do you actually need it?
If you plan on gaming at 4K, then it might be worth springing for the 16-gigabyte variant of the 4080, but it is exceedingly unlikely that any game will actually be able to utilize the whopping 24 gigabytes available on the RTX 4090 in the near future. You'll probably need the extra memory (and additional speed!) that the RTX 4090 is packing if you want to game at 8K, however.
If you're interested in artificial intelligence (AI) and machine learning (ML) applications, like generating enormous ultra-high-definition images in Stable Diffusion, the question gets more complicated. AI and ML both benefit from enormous amounts of VRAM, which makes the RTX 4090 much more appealing. However, you can get that much video memory in a 3090 Ti for at least $500 less --- it'll just be a bit slower.
Improvements to Ray Tracing
Good lighting is an essential part of any game's environment, and ray-traced lighting has been hailed as the golden standard for years. The major snag? Ray-traced lighting is computationally expensive. NVIDIA developed special Ray Tracing Cores (RT Cores) that are optimized to perform a specific kind of ray tracing calculation much more efficiently.
NVIDIA has included special ray tracing cores on their GPUs since the release of the RTX 2000-series GPUs. The RTX 4000-series comes loaded with the 3rd generation of Ray Tracing cores, and they promise to increase performance from the previous generation by two to three times.
If you care about ray tracing, then the RTX 4000-series GPUs are a very promising upgrade from previous generations.
What Is New with Deep Learning Super Sampling 3 (DLSS 3)?
DLSS is an AI rendering technology that allows some (any of the RTX cards) NVIDIA GPUs to use AI to upscale a frame that was rendered at a lower resolution to a higher resolution. The resulting performance gains can be enormous.
NVIDIA revealed DLSS 3 at their Keynote address and showcased a clip of DLSS 3 being used in Cyberpunk 2077. In their example, the normal framerate at 4K with full ray tracing was in the low 20s --- which is aggravatingly laggy for most people. DLSS 3 managed to push the framerates all the way into the mid-80s, which takes it from unpleasant-to-play territory to perfectly acceptable for most people.
That is very nearly four times the FPS with no noticeable impact on visual fidelity. NVIDIA's promotional material suggests that sort of FPS jump (two to four times) is to be expected with DLSS 3.
Should I Buy a 4000-Series GPU?
It is early, and there aren't enough real-world samples to be sure, but it looks likely that the 4000-series will represent one of the biggest generation-to-generation jumps in performance we've seen.
If you're willing to pay top dollar and are still using a GTX 1000-series GPU, or even an RTX 2000-series card, then the upgrade today is a no-brainer. The RTX 4090 looks like an incredible piece of hardware for anyone trying to run games at 4K at 144+ FPS or 8K at 60 FPS. Just keep in mind that it would be a waste if you're not in UHD territory.
More budget-conscious buyers, especially those on 1080p or 1440p, are going to want to wait until NVIDIA unveils the RTX 4070 and the RTX 4060 later in 2022 or early in 2023. The specs on the 12GB RTX 4080 are close to what one might expect from a 4070 if previous generations of GPUs were any indication --- clearly, something is a bit different this time around. It remains to be seen how powerful the RTX 4070 and the RTX 4060 are going to be when compared to their bigger siblings.
Even with that caveat, the improvements to NVIDIA's Tensor Cores (which power DLSS 3) and Ray Tracing Cores ensure that all of the RTX 4000-series GPUs are going to be compelling upgrades for every gamer out there, especially if they're not already using an RTX 3000-series card.
You can read more about the RTX 4000 series GPUs on NVIDIA's website.