ASUS Crams NVIDIA's GB300 Blackwell Ultra Superchip Into A Desktop PC
ASUS has officially unleashed into the realm of high-performance computing with its ExpertCenter Pro ET900N G3. What makes the ET900N G3 special isn't just that it sports an NVIDIA GB300 Grace Blackwell Ultra, but also that it's the first time the chip is being offered by a third-party brand and, in this case, is found in a svelte (and some might say sedate) desktop form factor.
Designed from the ground up almost specifically for AI computation in mind, the highlight of the ET900N G3 is indeed the NVIDIA GB300 Grace Blackwell Ultra, a hybrid superchip that combines NVIDIA's Grace ARM-based CPU with the Blackwell Ultra GPU.
Designed from the ground up almost specifically for AI computation in mind, the highlight of the ET900N G3 is indeed the NVIDIA GB300 Grace Blackwell Ultra, a hybrid superchip that combines NVIDIA's Grace ARM-based CPU with the Blackwell Ultra GPU.
One of the most striking features of this new machine is its memory capacity and performance. The system boasts up to 784GB of unified cache-coherent memory (a term you might recall from our Meteor Lake architecture rundown), most critical for handling massive AI models and datasets. This memory is intelligently distributed, with 288GB of HBM3E memory directly attributed to the GPU, complemented by 496GB of LPDDR5X RAM embedded around the Grace CPU. The idea behind the coherent memory architecture is to ensure smooth data flow and minimize operational bottlenecks.
In terms of raw computational power, the ExpertCenter Pro ET900N G3 is no slouch. It is engineered to deliver 20 PetaFLOPS (PFLOPS) of FP4 precision compute. therefore making the machine ideal in accelerating tasks like large language model (LLM) training, complex simulations, and advanced data analytics.
ASUS has also equipped the ET900N G3 with high-speed connectivity to match, including NVIDIA's ConnectX-8 SuperNIC that offers 800Gb/s. Indeed, with this kind of networking bandwidth/interface, scaling AI workloads across multiple systems or integrating seamlessly into data center environments shouldn't be a problem.
The workstation runs on NVIDIA's DGX OS, a specialized Ubuntu fork optimized with NVIDIA-specific kernel tweaks, drivers, and optimizations, ensuring the hardware and software work in perfect harmony.
The motherboard supports pretty decent expandability too: three full-length PCIe x16 slots, providing ample room for additional GPUs or specialized accelerators, plus three M.2 storage expansion slots. Power-wise, the system is equipped with standard ATX and EPS 12V connectors, alongside three dedicated 12V-2x6 GPU power connectors capable of supplying up to 1,800 watts.
While pricing and general availability details for the ExpertCenter Pro ET900N G3 are yet to be officially released, this custom-designed system, based on NVIDIA's DGX blueprint, is anticipated to come with a premium price tag, potentially exceeding $30k. It is primarily targeted at professional AI researchers, developers, and/or any organization needing advanced computing needs.
The motherboard supports pretty decent expandability too: three full-length PCIe x16 slots, providing ample room for additional GPUs or specialized accelerators, plus three M.2 storage expansion slots. Power-wise, the system is equipped with standard ATX and EPS 12V connectors, alongside three dedicated 12V-2x6 GPU power connectors capable of supplying up to 1,800 watts.
While pricing and general availability details for the ExpertCenter Pro ET900N G3 are yet to be officially released, this custom-designed system, based on NVIDIA's DGX blueprint, is anticipated to come with a premium price tag, potentially exceeding $30k. It is primarily targeted at professional AI researchers, developers, and/or any organization needing advanced computing needs.
