Bring data-center AI to your media workflow, without the server room
The DVEO GB300 AI Station is a deskside AI supercomputer powered by the NVIDIA GB300 Grace Blackwell Ultra Superchip, built on NVIDIA DGX Station architecture. It delivers up to 20 PFLOPS of AI performance and 748 GB of unified coherent memory, engineered for the AI media workloads that matter most: dubbing, subtitling, video upscaling, transcoding, and large-model inference.
Most AI media pipelines hit a memory wall, not a compute wall. Running large dubbing and subtitle models simultaneously forces constant data paging between GPU and system memory, adding latency and operational complexity. The GB300 AI Station eliminates that bottleneck. The NVIDIA NVLink C2C interconnect unifies 252 GB of HBM3e GPU memory and 496 GB of LPDDR5X CPU memory into a single coherent pool. Your models load once, stay resident, and process continuously, at broadcast scale.
Key Features
- NVIDIA GB300 Grace Blackwell Ultra Superchip: Integrates a high-performance NVIDIA Grace CPU and NVIDIA Blackwell Ultra GPU via NVLink C2C, delivering exceptional compute density in a single coherent architecture.
- 748 GB unified coherent memory: 252 GB HBM3e (7.1 TB/s) + 496 GB LPDDR5X (396 GB/s) accessible as a single pool. Run 1-trillion-parameter AI models without paging or pipeline interruption.
- Up to 20 PFLOPS of AI performance: Enough to run AI dubbing, subtitling, upscaling, and inference concurrently without resource contention.
- 800 Gb/s networking: NVIDIA ConnectX-8 SuperNIC with two 400 GbE QSFP ports for high-throughput live media ingest and delivery.
- Deskside form factor: 245 × 528 × 595 mm. No server room, no raised floor, no dedicated cooling infrastructure required.
- NVIDIA AI Software Stack: Full support for fine-tuning, inference, and data science workloads out of the box, compatible with DVEO's AI for Media pipeline.
- Up to 1 trillion parameters: The coherent memory pool supports next-generation LLM inference and multimodal AI without multi-node clustering.
Applications
- AI Dubbing: Run large speech synthesis and translation models fully resident in the 748 GB memory pool. Near-real-time multilingual dubbing output for broadcast delivery.
- AI Subtitling: Generate broadcast-grade subtitles for live and VOD content with ASR and translation models integrated into SRT/HLS/DASH workflows.
- Video Upscaling to 4K/8K: Neural upscaling pipelines for archival SD and HD content, at the throughput required for real-time or faster-than-real-time output.
- Large-Model Inference: Run LLMs and multimodal AI models locally — for content analysis, automated QC, and metadata generation — without cloud latency or egress costs.
- GPU-Accelerated Transcoding: Pair with DVEO Brutus Cloud for AI-assisted multi-bitrate transcoding and packaging for OTT and FAST delivery.
- Video Restoration: Restore archival footage for HD and 4K rebroadcast using AI restoration model chains running without pipeline interruption.
Technical Specifications
- Platform: NVIDIA DGX Station Architecture
- Superchip: NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip
- GPU: NVIDIA Blackwell Ultra GPU (1×)
- GPU Memory: 252 GB HBM3e — 7.1 TB/s bandwidth
- CPU Memory: Up to 496 GB LPDDR5X — up to 396 GB/s bandwidth
- Total Coherent Memory: 748 GB (CPU + GPU unified via NVLink C2C)
- AI Performance: Up to 20 PFLOPS
- Max Model Size: Up to 1 trillion parameters
- Networking: 2× 400 GbE QSFP (NVIDIA ConnectX-8 SuperNIC) + 1× 10 GbE RJ45
- NVMe Storage: 2× M.2 PCIe 5.0 (CPU) + 2× M.2 PCIe 6.0 (training data)
- PCIe Expansion: 1× Double-Wide PCIe 5.0 ×16 + 2× Single-Wide PCIe 5.0 ×16
- Management: Aspeed AST2600 BMC
- Power Supply: 1600W 80 PLUS Titanium ATX
- Dimensions: 245 mm (W) × 528.4 mm (H) × 595 mm (D)
- OS Compatibility: Ubuntu, RedHat Enterprise Linux, NVIDIA DGX OS
- Software: NVIDIA AI Software Stack + DVEO AI for Media (dubbing, subtitling, upscaling, restoration)
Maximize your AI media potential with the DVEO GB300 AI Station. Contact us today to learn more or to request a quote.
Bring data-center AI to your media workflow, without the server room
The DVEO GB300 AI Station is a deskside AI supercomputer powered by the NVIDIA GB300 Grace Blackwell Ultra Superchip, built on NVIDIA DGX Station architecture. It delivers up to 20 PFLOPS of AI performance and 748 GB of unified coherent memory, engineered for the AI media workloads that matter most: dubbing, subtitling, video upscaling, transcoding, and large-model inference.
Most AI media pipelines hit a memory wall, not a compute wall. Running large dubbing and subtitle models simultaneously forces constant data paging between GPU and system memory, adding latency and operational complexity. The GB300 AI Station eliminates that bottleneck. The NVIDIA NVLink C2C interconnect unifies 252 GB of HBM3e GPU memory and 496 GB of LPDDR5X CPU memory into a single coherent pool. Your models load once, stay resident, and process continuously, at broadcast scale.
Key Features
- NVIDIA GB300 Grace Blackwell Ultra Superchip: Integrates a high-performance NVIDIA Grace CPU and NVIDIA Blackwell Ultra GPU via NVLink C2C, delivering exceptional compute density in a single coherent architecture.
- 748 GB unified coherent memory: 252 GB HBM3e (7.1 TB/s) + 496 GB LPDDR5X (396 GB/s) accessible as a single pool. Run 1-trillion-parameter AI models without paging or pipeline interruption.
- Up to 20 PFLOPS of AI performance: Enough to run AI dubbing, subtitling, upscaling, and inference concurrently without resource contention.
- 800 Gb/s networking: NVIDIA ConnectX-8 SuperNIC with two 400 GbE QSFP ports for high-throughput live media ingest and delivery.
- Deskside form factor: 245 × 528 × 595 mm. No server room, no raised floor, no dedicated cooling infrastructure required.
- NVIDIA AI Software Stack: Full support for fine-tuning, inference, and data science workloads out of the box, compatible with DVEO's AI for Media pipeline.
- Up to 1 trillion parameters: The coherent memory pool supports next-generation LLM inference and multimodal AI without multi-node clustering.
Applications
- AI Dubbing: Run large speech synthesis and translation models fully resident in the 748 GB memory pool. Near-real-time multilingual dubbing output for broadcast delivery.
- AI Subtitling: Generate broadcast-grade subtitles for live and VOD content with ASR and translation models integrated into SRT/HLS/DASH workflows.
- Video Upscaling to 4K/8K: Neural upscaling pipelines for archival SD and HD content, at the throughput required for real-time or faster-than-real-time output.
- Large-Model Inference: Run LLMs and multimodal AI models locally — for content analysis, automated QC, and metadata generation — without cloud latency or egress costs.
- GPU-Accelerated Transcoding: Pair with DVEO Brutus Cloud for AI-assisted multi-bitrate transcoding and packaging for OTT and FAST delivery.
- Video Restoration: Restore archival footage for HD and 4K rebroadcast using AI restoration model chains running without pipeline interruption.
Technical Specifications
- Platform: NVIDIA DGX Station Architecture
- Superchip: NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip
- GPU: NVIDIA Blackwell Ultra GPU (1×)
- GPU Memory: 252 GB HBM3e — 7.1 TB/s bandwidth
- CPU Memory: Up to 496 GB LPDDR5X — up to 396 GB/s bandwidth
- Total Coherent Memory: 748 GB (CPU + GPU unified via NVLink C2C)
- AI Performance: Up to 20 PFLOPS
- Max Model Size: Up to 1 trillion parameters
- Networking: 2× 400 GbE QSFP (NVIDIA ConnectX-8 SuperNIC) + 1× 10 GbE RJ45
- NVMe Storage: 2× M.2 PCIe 5.0 (CPU) + 2× M.2 PCIe 6.0 (training data)
- PCIe Expansion: 1× Double-Wide PCIe 5.0 ×16 + 2× Single-Wide PCIe 5.0 ×16
- Management: Aspeed AST2600 BMC
- Power Supply: 1600W 80 PLUS Titanium ATX
- Dimensions: 245 mm (W) × 528.4 mm (H) × 595 mm (D)
- OS Compatibility: Ubuntu, RedHat Enterprise Linux, NVIDIA DGX OS
- Software: NVIDIA AI Software Stack + DVEO AI for Media (dubbing, subtitling, upscaling, restoration)
Maximize your AI media potential with the DVEO GB300 AI Station. Contact us today to learn more or to request a quote.
We make every effort to ensure your purchase is delivered on time and in good working condition.
In the event a system or component you purchased from DVEO is damaged in shipping, fails before the warranty period has ended, or you just need to return it, please refer to the steps below to return your purchase.
Shipping Policy
For Items Damaged During Shipping:
DVEO is NOT responsible for breakage, damage, or delay in transit. Once a purchase has been delivered, it is the customer's responsibility to inspect systems and/or components for damage incurred during shipping and file a claim with transportation company within 3 days after receipt of merchandise. All claims for loss, damage or shortage must be reported to DVEO within 5 days from receipt of order.
Customs, duties, and taxes
DVEO is not responsible for any customs and taxes applied to your order. All fees imposed during or after shipping are the responsibility of the customer (tariffs, taxes, etc.).
Domestic Shipping Policy
Shipment confirmation and order tracking
As soon as your order has been shipped you will receive an email confirmation with your tracking number.
DVEO will provide weekend deliveries upon the customer’s request and at an additional charge. If we are experiencing a high volume of orders and expect any delays, you will be notified.
International Shipping Policy
We ship to every country unless the freight companies have it on a NO SHIPMENT LIST. Or a particular country has a restriction on receiving shipments, due to COVID-19.
Potential Delays and COVID-19
There may be potential delays due to COVID-19 in terms of processing shipments.
The health and safety of our employees and customers is our top priority. We’ve redesigned our order warehouse center to comply with government/state recommended social distancing and personal protective equipment guidelines.
We will notify you when your order is shipped or if we foresee any issues with fulfilling your order.
Return Policy
For Products Under One Year Warranty:
Please have a description of your issue and the serial number on hand when you contact support. DVEO will assist you in troubleshooting the problem remotely.
If remote troubleshooting is unable to correct or identify the problem an RMA number will be issued to return the defective system or component for repair.
For Returns:
Returns NOT related to product warranties or damage during shipping are handled on a case by case basis. If you would like to return a product or system purchased through DVEO you must notify DVEO within 30 days.
- All returns will include a 25% restocking fee.
- The item must be shipped back in the original packaging.
- Used items that are missing any parts, accessories, manuals, and original packaging are subject to a 25%-75% restocking fee.
- Items that have been disassembled where structural parts were opened up, removed, unscrewed, tampered with, or broken by the user are not accepted for returns or refunds. Our return policy does not cover ordinary wear and tear or damage caused by improper use or accidents.
- Returned materials are subject to a restocking fee.
- Returns must be accompanied by an RMA number provided by DVEO.
- To get an RMA number, please email us at RMA@dveo.com.
- DVEO is not responsible for shipping expenses.
- COD or Collect returns will not be accepted.
- Returns should be processed within a few weeks.
All Items returned to DVEO must have shipping labels addressed as follows:
Computer Modules Inc / DVEO
RMA: (Reference # Here)
11413 West Bernardo Court
San Diego, CA 92127
(858)613-1818
For more information, email info@dveo.com