What you need to know
- NVIDIA has officially unveiled the new GeForce RTX 4090 and RTX 4080 graphics cards.
- The graphics cards are built on NVIDIA's Lovelace architecture, and deliver huge gains in energy efficiency and performance over the RTX 30 cards.
- The RTX 4090 is launching October 12 for $1,499, the RTX 4080 will go on sale in November, starting off at $899 for the 12GB version and $1,199 for the 16GB variant.
NVIDIA has officially announced the GeForce RTX 40 series, kicking things off with the RTX 4090 and two variants of the RTX 4080. The new graphics cards are powered by the Lovelace architecture and deliver huge gains across the board over the RTX 30 cards.
NVIDIA is touting an eye-watering 2x gains for shader workloads, 2.8x gains for RT cores, and up to a 2x increase in energy efficiency over the RTX 30 series. The efficiency gains are particularly interesting, and the switch to TSMC's 4n node — used by the best Android phones — has allowed NVIDIA to crank things up a gear.
We've rated NVIDIA's RTX 30 series as among the best graphics cards available to buy right now, and we're excited to see the gains on offer with the RTX 40 series. While the cards won't be available for a few weeks, let's deliver into the details and see what's changed.
NVIDIA RTX 40 GPUs: Everything you need to know
As is usually the case with every new generation, NVIDIA is focusing on the high-end category, introducing the RTX 4090 and RTX 4080 at launch. We'll get the RTX 4070 and RTX 4060 and other variants down the line, but for now, the focus is on showcasing what the Lovelace architecture brings to the table.
Here's a high-level overview of what you're getting with the RTX 4090 and RTX 4080:
| Category | RTX 4090 | RTX 4080 (16GB) | RTX 4080 (12GB) |
|---|---|---|---|
| CUDA Cores | 16384 | 9728 | 7680 |
| Shader Cores | 128 | 76 | 60 |
| Clock Speeds (Boost) | 2520MHz | 2505MHz | 2640MHz |
| Memory | 24GB GDDR6X | 16GB GDDR6X | 12GB GDDR6X |
| Memory Bus | 384-bit | 256-bit | 192-bit |
| Bandwidth | 1008GB/s | 736GB/s | 504GB/s |
| Power | 450W | 340W | 285W |
The RTX 4090 offers the same memory bus and video memory as the RTX 3090, and it slots in at the same $1,499 figure. What's different this time is that it gets more shader cores, and the node shift to 4nm will result in less power draw for the same workloads.
The 12GB RTX 4080 should hit the sweet spot for most gamers, delivering the best that the new Lovelace architecture has to offer for $899. The effective bandwidth is just over half of what you get with the RTX 4090, but that is still more than adequate for a vast majority of gamers. Considering how well the RTX 3080 has aged in the last two years, there is a lot to be excited for with the RTX 4080.
Lovelace architecture: Big performance & power draw
Ada Lovelace is the name of NVIDIA's architecture that powers the GeForce RTX 40 series. NVIDIA has showcased considerable gains over Ampere with the new series of GPUs set to offer up to 83 TFLOPS of computing power for PC games — double that of RTX 30 cards.
NVIDIA is touting up to 191 TFLOPS for the RT cores, a 2.8x increase over the RTX 4090, with the third-gen cores slated to deliver realistic lighting effects. DLSS is one of the best features introduced by NVIDIA in recent years, and the feature is getting a huge boost, with the brand highlighting 2x gains thanks to a new "neural network to boost frame rates while maintaining image quality."
There's also AV1 decode this time, and this should be a good addition for content creators looking to cut down on their encode times.
But the bigger story here is the TMSC 4nm node that NVIDIA is using for these cards, a significant upgrade over the Samsung 8nm node used for Ampere cards. This improvement alone provides vital space for NVIDIA to work in additional performance without sacrificing too much on power draw.
NVIDIA GeForce RTX 4090: The 8K powerhouse
NVIDIA says the RTX 4090 will deliver 100fps at 4K gaming, and it should be a viable candidate for 8K gaming in the near future. The card has significantly more shader cores, but the memory bus and video memory are identical.
What's interesting is that the new RTX 4090 needs 450W of power — the same as the current-gen card — but NVIDIA notes that it delivers 2x gains at the same power. This is the key benefit of switching to a more efficient node; the same workloads consume less power.
| Category | RTX 4090 | RTX 3090 |
|---|---|---|
| Architecture | Lovelace | Ampere |
| CUDA Cores | 16384 | 10496 |
| Clock Speeds | 2520MHz | 1695 |
| Memory | 24GB GDDR6X | 24GB GDDR6X |
| Memory Bus | 384-bit | 384-bit |
| Memory bandwidth | 1008GB/s | 936GB/s |
| Power | 450W | 450W |
NVIDIA GeForce RTX 4080: Turbocharged 4K gaming
NVIDIA says the RTX 4080 outshines the RTX 3090 Ti with DLSS enabled, and the card should be ideal for 4K gaming. The RTX 3080 fared relatively well in this area, and given the gains on offer with the RTX 4080, we should see significant gains.
The RTX 4080 comes in at a conservative 340W, and the 12GB version is under the 300W margin at 285W. If you haven't done so, we'd suggest checking your power supply and using our best PSUs guide to help you pick the right option.
Here's the rundown of how the two variants of the RTX 4080 hold up against the RTX 3080:
| Category | RTX 4080 (16GB) | RTX 4080 (12GB) | RTX 3080 |
|---|---|---|---|
| Architecture | Lovelace | Lovelace | Ampere |
| CUDA Cores | 9728 | 7680 | 8,704 |
| Clock Speeds | 2505MHz | 2640MHz | 1708GHz |
| Memory | 16GB GDDR6X | 12GB GDDR6X | 10GB GDDR6X, 320-bit |
| Memory Bus | 256-bit | 192-bit | 320-bit |
| Memory bandwidth | 736GB/s | 540GB/s | 760GB/s |
| Power | 340W | 285W | 320W |
RTX 40: Prices and availability
The RTX 4090 will be available starting October 12, with the RTX 4080 launching in November. Here's a breakdown of the cost for the RTX 40 Founder's Edition GPUs from NVIDIA:
- GeForce RTX 4090 Founder's Edition: $1,499
- GeForce RTX 4080 Founder's Edition (16GB): $1,199
- GeForce RTX 4080 Founder's Edition (12GB): $899
As has been the case with recent NVIDIA GPU launches, you'll be able to purchase a card from NVIDIA (one of the Founder's Editions) or go with an AIB partner like ASUS, PNY, Gigabyte, MSI, ZOTAC, and others. EVGA will not be a launch partner for NVIDIA, nor will the company be making any NVIDIA graphics cards for the foreseeable future.
