(Credit: John Burek)
Nvidia’s CES keynote announcement of its upcoming GeForce RTX 50-series graphics cards was exciting, but with few details revealed except names and prices, I was eager to learn more. Sooner than expected, that wait is over. I tracked down core counts and clock speeds, giving us a better picture of these GPUs' capabilities and the first solid basis for comparison with the outgoing RTX 40-series.
Nvidia GeForce RTX 50 Series Explained
RTX 5090
Naturally, the biggest focus of the RTX 50-series has been on its new flagship, the Nvidia GeForce RTX 5090. This card has a monster of a graphics chip with 21,760 CUDA cores clocked at 2.01GHz with a boost clock of 2.41GHz. It also has 32GB of GDDR7 memory and a 512-bit wide memory interface.
Though incredibly powerful, this isn't the graphics card that gamers and gaming enthusiasts should consider, as its $1,999 price tag and advanced AI processing technology all but ensure it will be snatched up by businesses and be relatively rare in gaming PCs. Anyone who does buy it should ensure they have a potent power supply, as Nvidia lists this card as requiring a 1,000-watt power supply and says the card can consume up to 575W by itself.
RTX 5080
Far more obtainable for gamers will be the Nvidia GeForce RTX 5080, which is essentially half of an Nvidia GeForce RTX 5090. The RTX 5080 will feature 10,752 CUDA cores clocked at 2.3GHz with a boost clock of 2.62GHz. That’s only a modest increase from the current Nvidia GeForce RTX 4080, which has 9,728 CUDA cores, and the Nvidia GeForce RTX 4080 Super, which has 10,240, but everything from Nvidia so far suggests the AI hardware has improved substantially. According to an Asus product page, the RTX 5080 GPU will have 30Gbps GDDR7 RAM connected to the GPU die over a 256-bit memory interface. The math gives us an effective bandwidth of 960GB/s between the GPU die and the memory.
The last-gen RTX 4080 and RTX 4080 Super also featured a 256-bit wide memory interface, but those used slower GDDR6X memory, and their bandwidth topped out at 716GB/s and 736GB/s, respectively. This means the new RTX 5080 has a more significant 30% bandwidth advantage over the RTX 4080 Super compared with the more meager 5% increase in CUDA cores.
Nvidia has not released details on the rest of its shading resources. Still, often the proportion of texture mapping units (TMUs) and raster operation processors (ROPs) to CUDA cores doesn’t change from one generation to the next. Nvidia packs most of its shading resources into a logic block called a streaming multi-processor (SM) for ease of scalability.
If this remains unchanged from the RTX 40 series, the RTX 5080 will have 84 SMs with 336 TMUs. The ROPs aren’t packaged inside the SMs, which makes them harder to estimate, but they are instead closely associated with the memory interface and are likely to remain unchanged or relatively close to the previous generation in overall count. That suggests the RTX 5080 will have just 112 ROPs.
Recommended by Our Editors
Other than those specs, I found just two other noteworthy pieces of information on the product page. First, the RTX 5080 supports a PCIe 5.0 x16 connection, an upgrade from the PCIe 4.0 x16 connection used on the last-gen RTX 40-series cards. In fact, all of these new graphics cards use PCIe 5.0 connections up and down the stack. We'll see whether this holds up with eventual 5060 and 5050 cards.
Second, the recommended PSU Nvidia listed on its spec page is 850W, an increase from the 750W rating Nvidia set on the RTX 4080 and RTX 4080 Super. This suggests the new RTX 5080 will be more power-hungry with a max power consumption of 360W.
RTX 5070 Ti
For now, those are all the details I could find on the new RTX 5080, but similar info for both the Nvidia GeForce RTX 5070 and the GeForce RTX 5070 Ti is available. These GPUs will have 6,144 and 8,960 CUDA cores, respectively. One of the more interesting details we didn't know before is that both cards will also ship with GDDR7 RAM. When new high-end memory technologies first come out, it’s common that they are only used on the highest-end products in a line due to cost and supply limitations, so this wasn’t at all a guaranteed thing before now.
The RTX 5070 Ti is clocked at 2.3GHz with a max turbo of 2.45GHz. It is paired with 16GB of GDDR7 memory that Gigabyte suggests are 28Gbps memory chips, which would give it a bandwidth of 896GB/s. Using similar math as I used for the RTX 5080 above, I would estimate that this card will have 70 SMs, 280 TMUs, and 112 ROPs, though this is speculative. Nvidia recommends a 750W PSU for the RTX 5070 Ti and lists its max power draw at 300W.
RTX 5070
As for the Nvidia GeForce RTX 5070 (no-Ti), its 6,144 CUDA cores are clocked at 2.16GHz with a higher turbo of 2.51GHz than the RTX 5070 Ti to help make up for the lower core count. I would estimate it to have 48 SMs based on its CUDA core count with 192 TMUs and around 80 ROPs. This model also has a 192-bit wide memory interface and will connect to 12GB of 28Gbps GDDR7, according to Gigabyte, giving it an effective bandwidth of 672GB/s. Power consumption is lower here, with a max power draw of 250W, and the recommended PSU power rating is 650W.
It's also worth remembering that Nvidia said its GeForce RTX 5070 would be able to match the RTX 4090 in games, which, if true, is all the more impressive based on its hardware resources and power draw. Both are far closer to the last-gen Nvidia GeForce RTX 4070, with the core count even lower than the Nvidia GeForce RTX 4070 Super. The RTX 4070 Super also has similar clocks to the RTX 5070, which suggests it could be faster than the RTX 5070 in some situations. The RTX 5070 is likely only able to match the RTX 4090 when using its AI hardware to gain an advantage, so it'll be exciting to see whether the RTX 5070 can beat the RTX 4070 Super when DLSS 4 isn't used.
For now, that’s all the information I've gleaned about the new RTX 50 series, but stay tuned for more updates as we approach the launch of these graphics cards this month.