NVIDIA will unveil its GeForce RTX 50 series “Blackwell” graphics cards at CES 2025 next month. Company CEO Jensen Huang is expected to show off the RTX 5090 and RTX 5080 during his keynote in Las Vegas. According to WCCFTech, the chipmaker will announce both desktop and notebook SKUs. However, the RTX 5080 (desktop) will be the first to hit retail in January 2025.
NVIDIA RTX 5080 16 GB: The First Blackwell Gaming GPU
The GeForce RTX 5080 16 GB is expected to leverage the fully enabled GB203 die, including 84 SMs and 10,752 shader cores. That’s 11% higher than the RTX 4080 Super (10,240 cores), indicating a mild to moderate generational uplift. The next-gen GDDR7 memory will be the primary upgrade, increasing the bandwidth to 1024 GB/s (vs. 736 GB/s on the 4080S).
| NVIDIA RTX 5090 | NVIDIA RTX 5080 | NVIDIA RTX 5070 Ti | NVIDIA RTX 5070 | |
|---|---|---|---|---|
| GPU | GB202-300 | GB203-400 | GB203-300 | GB205 |
| SMs | 170 | 84 | 70 | 50 |
| CUDA Cores | 21,760 | 10,752 | 8,960 | 6,400 |
| Memory (GDDR7) | 32 GB @ 28 Gbps | 16 GB @ 32 Gbps | 16 GB @ 28 Gbps | 12 GB @ 28 Gbps |
| Memory Bus | 512-bit | 256-bit | 256-bit | 192-bit |
| Bandwidth | 1792 GB/s | 1024 GB/s | 896 GB/s | 672 GB/s |
| TBP | 600W | 400W | 350W | 250W |
| Launch | Q1 2025 | Q1 2025 | Q1 2025 | Q1 2025 |
The TBP (Total Board Power) is allegedly set to increase to 400W (vs. 320W on the RTX 4080 Super). This implies substantially higher boost clocks. The L2 cache and ROP count should also be scaled up accordingly, but we have no exact figures at the moment. The memory bus width (256-bit) and buffer size (16 GB) are reportedly not seeing any increase.
The GeForce RTX 5080 16 GB is scheduled to hit retail shelves by mid-January 2025. Knowing NVIDIA, it will be marketed as the fastest gaming GPU (briefly) till the RTX 5090 arrives. Expect a 30-50% performance uplift over the RTX 4080/4080 Super at 4K. This would place it *just* ahead of the RTX 4090 at a much more affordable $1000-1200 price point.
NVIDIA RTX 5090 32 GB: One GPU to Rule them All
The GeForce RTX 5090 32 GB will land by the end of Jan (or early Feb). It is rumored to feature GB202 die, including 21,760 shaders across 170 SMs. That’s 33% more than the RTX 4090 (16,384). Additionally, these shaders will run at markedly higher clocks (~3 GHz), granting a substantial performance boost in all workloads.
| NVIDIA RTX 5090 | NVIDIA RTX 5080 | NVIDIA RTX 5070 Ti | NVIDIA RTX 5070 | |
|---|---|---|---|---|
| GPU | GB202-300 | GB203-400 | GB203-300 | GB205 |
| SMs | 170 | 84 | 70 | 48 |
| CUDA Cores | 21,760 | 10,752 | 8,960 | 6,144 |
| Memory (GDDR7) | 32 GB @ 28 Gbps | 16 GB @ 32 Gbps | 16 GB @ 28 Gbps | 12 GB @ 28 Gbps |
| Memory Bus | 512-bit | 256-bit | 256-bit | 192-bit |
| Bandwidth | 1792 GB/s | 1024 GB/s | 896 GB/s | 672 GB/s |
| TBP | 600W | 400W | 350W | 250W |
| Launch | Q1 2025 | Q1 2025 | Q1 2025 | Q1 2025 |
The 32 GB GDDR7 memory buffer will be paired with a 512-bit memory bus (or 384-bit?), resulting in a peak memory bandwidth of 1792 GB/s. The result will be an epic monstrosity that may be the first true 8K gaming GPU, courtesy of the ~80% bandwidth hike.
Of course, the L2 cache and ROPs will be scaled up as well. The RTX 5090 will have a TBP of up to 600W. Expect a performance increase of up to 2x over the RTX 4090, and a steep $2,000+ price-tag to go along with it.
NVIDIA RTX 50 Laptop GPUs
Partial specs of the RTX 50 mobility GPUs have also leaked out (via Huang514613). Interestingly, NVIDIA has decided to equip all the Blackwell laptop graphics chips with GDDR7 memory. The RTX 5080 and 5090 will get 16 GB of GDDR7 memory, 12 GB for the RTX 5070 Ti, and 8 GB for the RTX 5070, 5060, and 5050. The memory clocks will likely decrease down the stack.
The RTX 5080 and 5090 laptop GPUs will have a TBP of 175W, 140W for the RTX 5070 Ti, and 135W and lower for the remaining SKUs. ASUS plans to pair the RTX 50 mobility GPUs with Intel’s Core Ultra 9 285HX/275HX and AMD’s Ryzen 7 7945HX/7940HX processors.
The Ryzen 9 AI Max+ 395 (Strix Halo) sits at the top of the stack with 40 CUs or 2,560 shaders based on the RDNA 3.5 graphics architecture. It will be paired with a 2K screen with a 180Hz refresh rate, versus 240Hz displays for the RTX 50 devices.
