Introduction
At CES, NVIDIA made the announcement for their new GeForce RTX Blackwell lineup featuring the NVIDIA RTX 5090, RTX 5080, RTX 5070Ti and RTX 5070. During the event that focused a lot of AI training and simulation, we got a glimpse on pricing and availability of the cards.
We will go over the architectural changes to the new Blackwell family of GPUs as well as what we think of the claimed performance increase.
GeForce RTX Blackwell Technological Advancements
The flagship Blackwell GPU features 92 billion transistors, updated SMs (streaming multiprocessors), next generation Tensor Cores for AI computing, next gen RT Cores (ray-tracing) for high fidelity graphics computing, and up to 21760 CUDA cores (found in the RTX 5090).
CUDA Cores, Tensor Cores, and RT Cores work in tandem to deliver unparalleled graphics and productivity performance via the all new DLSS 4.0. While DLSS is mainly used in gaming, the advancements in AI translate to VR and scene generation like 3D modeling and the architectural design space.
To deliver these high performance and high degree of complexity data calculations to the user NVIDIA RTX 50-series use GDDR7, a new memory module, capable of over 1.8TB of memory bandwidth.
Here is the complete table of specifications from the released RTX 50-series GPUs and their MSRPs with the NVIDIA RTX 4090 for reference.
NVIDIA RTX 5090, RTX 5080, RTX 5070Ti, and RTX 5070 Specifications
Model | Interface | Memory | Memory Bandwidth | CUDA Cores | Tensor Cores | RT Cores | Single Precision (TFLOPS) | AI Throughput (TOPS) | Power | MSRP |
|---|---|---|---|---|---|---|---|---|---|---|
NVIDIA RTX 5090 | PCIe 5.0 x16 | 32GB GDDR7 | 1792 GB/s | 21,760 | 680 | 170 | 104.8 | 3352 | 575W | $1999 |
NVIDIA RTX 5080 | PCIe 5.0 x16 | 16GB GDDR7 | 960 GB/s | 10752 | 336 | 84 | 56.3 | 1801 | 360W | $999 |
NVIDIA RTX 5070Ti | PCIe 5.0 x16 | 16GB GDDR7 | 896 GB/s | 8960 | 280 | 70 | 43.9 | 1406 | 300W | $749 |
NVIDIA RTX 5070 | PCIe 5.0 x16 | 12GB GDDR7 | 672 GB/s | 6144 | 192 | 48 | 30.9 | 988 | 250W | $579 |
NVIDIA RTX 4090 | PCIe 4.0 x16 | 24GB GDDR6X | 1001 TB/s | 16384 | 512 (Last Gen) | 128 (Last Gen) | 82.6 | 1321 | 450W | $1599 |
NVIDIA is stressing that the generational improvements to RTX Blackwell is only possible with AI, so take this with a grain of salt. These were measurements of FPS, and utilizing a different versions of frame generation technique showcases the improvements in the software pipeline. For strictly throughput based workloads that cannot take advantage of tensor cores, expect a different level of gen-over-gen performance increase.
That isn’t saying the RTX 5090 will not be the most powerful GPU! But it does caution the performance increases that NVIDIA claims, especially the claim of the RTX 5070 having RTX 4090 performance levels. This is only in highly specific DLSS 4.0 scenarios.
Release Dates
The RTX 5090 GPU RTX 5080 GPU will be available on Jan. 30 at $1,999 and $999, respectively. The RTX 5070 Ti GPU and RTX 5070 GPU will be available starting in February at $749 and $549, respectively. All but the RTX 5070Ti will have a Founder’s Edition model, the 2-slot form factor by NVIDIA.
All RTX Blackwell GPUs will also be available from board partners like ASUS, MSI, GIGABYTE, PNY, ZOTAC, and more. Likely, these board partners will not be able to replicate the 2-slot width design that the NVIDIA Founder’s Edition cards will ship with.
If you have any questions on the new RTX Blackwell GPUs, or would like to get a quote on a system featuring these new GPUs, contact us today for more information. SabrePC let’s you customized your own desktop, workstation, or server; visit one of our solution pages to customize a platform today. We have the RTX 50-series GPUs available to pre-order and select as a GPU in our configurator. Pre-order now!
