CachyOS: New GeForce graphics driver for AMD's Zen 4 and Zen 5
 |
| Source: Gigabyte |
Blackwell Ultra is coming soon to desktop PCs. The AI workstations designed by Nvidia and its partners rely on a Grace Blackwell Ultra superchip with 288 GB of HBM3E and 72 ARM processor cores with 496 GB of LPDDR5X.
As part of its GPU Technology Conference 2026, which will be held at the San Jose Convention Center until March 19, Nvidia has announced that the desktop PCs with Blackwell Ultra alias GB300 will be available for pre-order soon. The AI workstation designed by Nvidia and its partners relies on a Grace Blackwell Ultra Superchip, which has 288 gibytes of HBM3E memory and a total of 72 ARM processor cores with 496 gibytes of LPDDR5X.
Based on the Nvidia DGX Station, which serves as a reference design, workstations such as the MSI XpertStation WS300, the Asus ExpertCenter Pro G3 and the Gigabyte W775-V10-L01 have already been announced. AI developers can accordingly fall back on 784 gibytes of system memory, since the two memory pools of HBM3E and LPDDR5X can be used in combination.
- Nvidia GB300 Grace Blackwell Ultra Desktop Superchip
- Nvidia Grace CPU with 72 Neoverse V2 ARM processor cores
- Nvidia Blackwell Ultra GPU with 288 GiByte HBM3E memory
- 496 GiByte LPDDR5X as system memory for the CPU
- ConnectX-8 SuperNIC network chip with 800 Gb/s
Over three times PCIe 5.0 ×16 additional graphics cards or AI accelerators can be installed, while three M.2 slots are provided for PCIe SSD. The power supply is provided by standard ATX + EPS12V and three 12VHPWR connections of the second generation ("12V-2×6"), which alone can supply up to 1,800 watts.
 |
| Source: Gigabyte |
| Nvidia GB300 | Nvidia GB200 | |
|---|---|---|
| Codename | Blackwell Ultra | Blackwell |
| Construction | Dual-die graphics chip | Dual-die graphics chip |
| transistors | 208 billion | 208 billion |
| Streaming Multiprocessors | 160 | 144 |
| FP4 Tensor Core (Dense/Sparse) | 15/20 PFLOPS | 10/20 PFLOPS |
| FP8/FP6 Tensor Core (Dense/Sparse) | 5/10 PFLOPS | 5/10 PFLOPS |
| INT8 Tensor Core (Dense/Sparse) | 157.5/315 TFLOPS | 105/210 TFLOPS |
| FP16/BF16 Tensor Core (Dense/Sparse) | 2.5/5 PFLOPS | 2.5/5 PFLOPS |
| TF32 Tensor Core (Dense/Sparse) | 1.25/2.5 PFLOPS | 1.25/2.5 PFLOPS |
| FP32 | 80 TFLOPS | 80 TFLOPS |
| memory | 288 GiB HBM3E | 192 GiB HBM3E |
| Memory bandwidth | 8 TB/s | 8 TB/s |
| NVLink interface | 1.8 TB/s | 1.8 TB/s |
| interface | PCIe 6.0 with 256 GiB/s | PCIe 5.0 with 128 GiB/s |
| TDP | 1400 W | 1200 W |
The in-house Linux distribution Nvidia DGX OS 7 is used as the operating system, which in turn is based on Ubuntu 24.04 LTS ("Noble Numbat") and has been optimized accordingly for use on Nvidia's AI systems.
With regard to the expected price, Nvidia is still keeping a low profile and, as usual in this category, will probably only communicate it on request if interested. However, the AI workstation in desktop format is likely to cost several tens of thousands of US dollars, since a GB200 superchip is already being traded for up to 70,000 US dollars.
Your opinion is in demand this CachyOS: New GeForce graphics driver for AMD's Zen 4 and Zen 5!
How do you feel about this issue? The Shero King editorial team is pleased about your well-founded opinion in the comments to this message. To comment, you need to click on Shero King or be logged in to the Extreme forum.
Post a Comment