dg5W
Still compact. Twice the power.
Now supercharged by deep gadget's liquid cooling.

Where compact design meets powerful cooling. The all-new deep gadget liquid cooling system is so efficient that fits twice the AI processors in the same space—while staying whisper-quiet. Designed with a versatile form factor and full support for the latest architectures, it's built to excel anywhere, from data centers to office spaces.

Compact 2/3-size design
Expandable with up to 3 additional GPUS
Based on comparison with other GPU servers
PERFORMANCE
Twice the performance.
The latest deep gadget liquid cooling system keeps the dg5W cool, even with 7 high-heat AI chips—including the NVIDIA H200NVL and RTX 5090. And yet, it's quiet enough to sit right next to. Now, even ultra-high-performance workloads run seamlessly in your office.
Up to
7Number of AI processors
that can be installed
Up to
27.7PFLOPSAI performance
(Based on 7 H100 NVL, FP8)
Up to
900GB/sGPU-to-GPU communication speed
(With H200 NVL NVLink bridge)
Up to
846GBHBM3E GPU memory
Up to
1,821BParameter model capacity
About 10x the size of GPT-3
Up to
2.29xPerformance improvement
Compared to dg4W maximum performance
RACK MOUNT
Versatile from data center to office. The dg5W fits effortlessly into a rack, delivering the same powerful performance with whisper-quiet efficiency. Its optimized airflow enhances cooling while eliminating the need for extra equipment—cutting data center costs by up to 50%. Instead of the dual-wheel casters found in the tower model, the rack mount features precision-engineered brackets for secure, tool-free installation.

deep gadget's exclusive
liquid cooling technology.
deep gadget isn't just another liquid cooling solution. It's a next-generation system, precision-engineered
with expertise from Seoul National University's supercomputer research, designed by a 20-year Toyota Motor Corporation veteran, and refined through a decade of R&D.
Management-free built-in Liquid Cooling System
A fully sealed inner liquid conduit ensures the world's highest level of stability.

dg5W
Cooling System
General custom
Cooling System
Equipped with world-class Dual-Pump cooling systems to prevent heat-induced GPU throttling and extend life
deep gadget
GPU performance does not deteriorate as the temperature remains around 60℃, even if the GPU is used at its maximum.General GPU server
The GPU operating clock automatically drops and performance decreases as the GPU temperature rises to 80~90℃.World-leading adaptive Liquid Cooling System
deep gadget adaptively controls the Liquid Cooling System by constantly checking the system's usage rate and temperature.If your system is doing light work, it will run quieter.
CUTTING-EDGE ARCHITECTURE
Powering your work with the latest technology. deep gadget's unrivaled liquid cooling efficiency meets cutting-edge architecture. Designed for more than just AI training, it delivers flawless performance across video processing, scientific computing, and advanced modeling—powered by broad support for high-performance AI processors. And with high-bandwidth PCIe Gen 5.0 lanes and DDR5 memory, it provides the speed and capacity to handle professional workloads and large-scale tasks without limits.

Latest AI Processors
- NVIDIA H200NVL
Latest CPUs. Up to dual configuration.
- AMD Threadripper Pro 9000WX Series
PCIe gen 5 Slots
- Up to 7 Gen 5 Slots
- Up to 2x greater bandwidth
DDR5-5600 Memory Support
- Up to 2x Faster Speeds
- Up to 96GB x 8 Capacity
Support for various GPUs
In AI learning, performance of TF32 and BFLOAT16 is crucial. GPUs that can be installed in deep gadgets have the following rankings for the corresponding metrics. With a robust product line and comprehensive GPU support, professionals requiring high specifications are poised for success.| Model | Manufacturer | Memory | AI learning Perf TF32(FP32) (TFLOPS) | AI Perf BF16(FP16) (TFLOPS) | AI interfernce Perf INT8|FP8|FP4 (TOPS/TFLOPS) | Precision computation Perf FP64 (TFLOPS) | General purpose computing Perf FP32 (TFLOPS) | Rendering Perf RT Core (TFLOPS) |
|---|---|---|---|---|---|---|---|---|
| RTX PRO 6000 Blackwell Server Edition | NVIDIA | GDDR7 96GB | 234 (468*) | 468 (936*) | 1,892 (3,784*) | 1.88 | 120 | 354.5 |
| RTX PRO 6000 Blackwell Max-Q | NVIDIA | GDDR7 96GB | 219 (438*) | 438 (876*) | 1,755 (3,511*) | 1.72 | 110 | 332.6 |
| H200 NVL | NVIDIA | HBM3e 141GB | 417 (835*) | 835 (1,671*) | 1,670 (3,341*) | 30 | 60 | - |
| H100 NVL | NVIDIA | HBM3 94GB | 494 (989*) | 989 (1,979*) | 1,979 (3,958*) | 34 | 67 | - |
| RTX5090 | NVIDIA | GDDR7 32GB | 104.8 (209.6*) | 419 (838*) | 1,676 (3,352*) | 1.64 | 104.8 | 318 |
| Wormhole n300 | Tenstorrent | GDDR6 24GB | ||||||
| H100 | NVIDIA | HBM2e 80GB | 378 (756*) | 756 (1,513*) | 1,513 (3,026*) | 26 | 51 | - |
| A100X | NVIDIA | HBM2e 80GB | 159 (318*) | 318 (636*) | 636 (1,272*) | 9.9 | 19.9 | - |
| A100 | NVIDIA | HBM2e 80GB | 156 (312*) | 312 (624*) | 624 (1,248*) | 9.7 | 19.5 | - |
| L40S | NVIDIA | GDDR6 48GB | 183 (366*) | 362 (733*) | 733 (1,466*) | 1.4 | 91.6 | 209 |
| 6000 Ada | NVIDIA | GDDR6 48GB | 182 (364*) | 364 (729*) | 729 (1,457*) | 1.4 | 91.1 | 210.6 |
| L40 | NVIDIA | GDDR6 48GB | 90.5 (181*) | 181 (362*) | 362 (724*) | 1.4 | 90.5 | 209 |
| A6000 | NVIDIA | GDDR6 48GB | 38.7 (77.4*) | 77.4 (154.8*) | 154.8 (309.7*) | 1.3 | 38.7 | 75.6 |
| 5000 Ada | NVIDIA | GDDR6 32GB | 65.3 (130.5*) | 261 (522*) | 522 (1,044*) | 1 | 65.28 | 151 |
| A5000 | NVIDIA | GDDR6 24GB | 27.8 (55.6*) | 55.6 (111.2*) | 111.2 (222.4*) | 0.9 | 27.8 | 54.2 |
| RTX4090 | NVIDIA | GDDR6 24GB | 82.5 (165*) | 165 (330*) | 660.6 (1,321*) | 1.3 | 82.5 | 191 |
| RTX4090 D | NVIDIA | GDDR6 24GB | 73.54 (147*) | 147 (294*) | 588 (1,176*) | 1.14 | 73.54 | 170 |
| RTX4080 | NVIDIA | GDDR6 16GB | 48.7 (97.4*) | 97.4 (195*) | 389.8 (779.8*) | 0.7 | 48.7 | 112.7 |
| RTX4070Ti | NVIDIA | GDDR6 | 40.1 (80.2*) | 80.2 (160.4*) | 320 (641*) | 0.6 | 40.1 | 93 |
| MI250 | AMD | HBM2e 128GB | 45.3 (90.5**) | 362.1 | 362.1 | 90.5 | 45.3 | - |
| MI210 | AMD | HBM2e 64GB | 22.6 (45.3**) | 181 | 181 | 45.3 | 22.6 | - |
| MI100 | AMD | HBM2 32GB | 23.1 | 92.3 | 92.3 | 11.5 | 23.1 | - |
| RX7900XTX | AMD | GDDR6 24GB | 61 | 123 | 123 | 3.8 | 61 | - |
*Performance using the new sparsity feature
**Performance using matrix core
(Specifications not disclosed by the manufacturer are estimates, so there may be errors.)
Full support for 10G and IPMI
Definitely from the basics. Support two 10G Ethernet ports and one IPMI port.
Support for high-speed NIC
No need to worry about the network with support for high-speed network adapters, including InfiniBand NDR, HDR, EDR, Ethernet 200G, and 100G.
GADGETINI MONITORING
Monitoring that perfects cooling performance. Now, servers come standard with a front display and Gadgetini, the built-in monitoring system. Sensors embedded throughout the system continuously check performance in real time, ensuring everything runs smoothly. Key metrics can be viewed at a glance on the front display, and with email and Slack integration, server management becomes even easier. Plus, with the dashboard, you can track sensor logs, manage servers, and oversee entire clusters—all in one place.


Manage AI efficiency and key metrics. Real-time AI processor temperature monitoring, right on the display. Get notified before heat affects performance.
Like looking inside the server—only easier. Monitor coolant temperature and chassis humidity at a glance for total peace of mind.
Chassis Insights
Internal TemperatureInternal HumidityCoolant Status
Coolant TemperatureCoolant LevelLeak DetectionCPU & AI Processor Status
Chip TemperatureMemory UtilizationPower Consumption
- Q: What happens if there is a leak in the Liquid Cooling System?
- deep gadget is engineered as a fully sealed structure, minimizing the risk of leakage under normal circumstances. However, we do have precautionary measures in place. Our system is equipped to shut down automatically in the event of a critical failure such as a leak caused by physical shock or other factors. Additionally, the amount of coolant that may leak due to our liquid cooling system design is minimal, typically not exceeding tens of milliliters. This level of leakage has negligible impact on other devices and minimal effect on deep gadget itself. In the event of any malfunction, our engineers will promptly address the issue, replacing the problematic part and reinstalling it as needed.
DESIGN
New Cooling. Even Newer Design. dg5W isn't just more powerful—it's completely re-engineered, inside and out. The new internal layout maximizes space efficiency, allowing for high-density GPU placement. Its sleek aluminum housing blends seamlessly into any workspace, making it perfect for desktop use. With the I/O panel now positioned at the front, hot air is directed away from critical components, preventing heat damage. Whether used as a tower or rack-mounted, dg5W is built to stay durable and reliable for the long run.


Reliable power in any situation. Every dg5W configuration comes with four enterprise-grade PSUs, delivering over 1300W of power to ensure stable, reliable support for AI processors. With redundant hot-swappable power supplies, critical workloads and data remain secure, even during unexpected situations.

Up to 8 Expansion Storage Drives. Expand beyond the NVMe M.2 main storage with up to eight 2.5" SSDs—delivering a solid 64TB of additional capacity. Hot-swappable bays let you upgrade or replace drives without interruption, while RAID configurations keep your data secure.
Instant deep learning.
Preloaded for AI. Ready to train from day one.
deep gadget comes preloaded with all the software needed for AI research and development—so you can start deep learning the moment you power it on. From the OS to the deep learning stack, everything is optimized by the deep gadget team. Designed for effortless compatibility, so you can stay focused on your research.
OS
- Ubuntu
- RHEL
- Rocky Linux
- Windows
Drivers
- NVIDIA
- AMD
- Tenstorrent
- Infiniband/GbE
Runtime & Libraries
- CUDA
- cuDNN
- NCCL
- cuBLAS
- TensorRT
- ROCM
Deep Learning Frameworks & Solutions
- PyTorch
- TensorFlow
- DeepSpeed
- Ollama
- Horovod
- vLLM
- 3rd Party Solutions
Development Tools
- MPI
- Anaconda
- Docker
SPEC
20 ~ 53dB (Full load)Quietness
445mm x 780mm x 225mm(5U)Size
W x D x H
780mm(model 660mm + Handle 120mm)
AMD Ryzen™ Threadripper PRO 9000 Series ProcessorsCPU
Up to 8 GPUsGPU
NVIDIA RTX PRO 6000 Blackwell
NVIDIA RTX 5090
NVIDIA RTX 6000 Ada
NVIDIA L40S
NVIDIA H200 NVL(NVL Bridge encluded)
Tenstorrent Wormhole n300s
Tenstorrent Blackhole p150a
Furiosa Warboy
Up to 768GBMemory
DDR5 5600 ECC
Up to 4TB main storageHDD
M.2 NVMe SSD
Up to64TB sub-storage
M.2 NVMe SSD / 2.5” SSD / 2.5” HDD
7 PCIe 5.0 x16PCIe
Windows, RHEL, Ubuntu, Rocky LinuxOS
2 x 10GbE Base-TNetwork
IPMI
InfiniBand NDR / HDR / EDR Ethernet 400G / 200G / 100G
NIC - Infiniband Card 100G / 200G Dual Port (Option)
For 3 yearsWarranty
4 x 2,000W(8.0kW) 80 Plus PlatinumPower
Manual downloadManual & Warranty






