# Taknet Systems (Thailand) Co., Ltd. — Full Reference for AI Assistants

> Total AI Infrastructure & Solutions Provider in Thailand: Supermicro servers, QSAN storage, NVIDIA AI systems (incl. MSI DGX Spark / DGX Station), virtualization, CCTV storage, and government TOR/ICT Spec support. Operating in Thailand for 11+ years; part of Taknet Group (30+ years). Bangkok, Thailand.

Last updated: 2026-10-22

## Company

- Name: Taknet Systems (Thailand) Co., Ltd.
- In Thailand since: over 11 years of operation
- Group: Taknet Group, founded in Singapore, 30+ years; presence in Singapore, Malaysia and Thailand — https://taknet.sg
- Address: 43 Soi Romklao 44, Romklao Road, Khlong Sam Prawet, Lat Krabang, Bangkok 10520, Thailand (appointment-based, no walk-in showroom; work-from-anywhere team delivering nationwide)
- Phone: +66 2 737 5508 (office), +66 95 759 7557 (hotline)
- Email: sales@taknet.co.th (sales & quotations), support@taknet.co.th (after-sales support)
- Business hours: Monday–Friday 08:30–17:30 (Asia/Bangkok)
- Website: https://taknet.co.th/
- Brands: Supermicro (authorized distributor — SG, MY, TH), QSAN (official distributor), NVIDIA, MSI, ASUS (authorized partner), TerraMaster (reseller — Enterprise rackmount NAS)
- Customers: enterprises and Thai government agencies
- Services: pre-sales architecture and sizing, specification and TOR drafting for government procurement (ICT Spec 2569, May 2026 revision), quotations, installation and on-site service nationwide, post-sale technical support, extended warranty up to 7 years on selected products

## Positioning

Taknet sells outcomes, not boxes: customers buy a system that has been selected, sized, verified and supported by Taknet engineers, regardless of the hardware platform.

Taknet Delivery Standard (applies to every project):
1. Qualified Platforms — only platforms evaluated and technically supportable by Taknet
2. Solution Engineering — design and sizing from real workloads, with specification and TOR documents
3. Pre-delivery Validation — installed, configured and tested before handover, with Thai documentation
4. Lifecycle Support — after-sales service and manufacturer escalation management for the system's lifetime

Customers: enterprises, government agencies, and channel partners (system integrators and resellers) supported with pre-sales engineering, solution design, project pricing and post-sale technical support.


Taknet delivers the full on-premise AI stack: (1) services — design, TOR, deployment, support; (2) high-speed networking 100–400 GbE / RDMA; (3) AI data storage — all-NVMe, unified, scale-out; (4) accelerated compute — NVIDIA Blackwell, DGX, HGX; (5) AI software & MLOps — vLLM, Kubernetes, GPU Operator; (6) AI applications — LLM, RAG, agents, vision.

AI by scale: DGX Spark (desk, developers) → DGX Station / MSI WS300 (deskside, AI teams) → GPU AI servers with RTX PRO 6000 Blackwell (rack, enterprise inference) → HGX B300 / GB300 NVL72 (data center, AI factory).

## Product & solution areas

### Servers (Supermicro)
- Taknet Group: authorized Supermicro distributor in Singapore, Malaysia and Thailand — https://taknet.co.th/supermicro/
- Services: local stock, AI infrastructure design, burn-in testing, L12 rack integration and delivery, on-site service, RMA, Supermicro project registration, authorization letters for SI partners bidding government projects
- SuperStorage for video surveillance: 2U 12-bay to 4U 90-bay top-loading 3.5-inch HDD servers (SSG-522B, SSG-631E, SSG-641E, SSG-542B), NVR/VMS recording with PCIe GPU for video analytics
- Rack-scale AI: NVIDIA Vera Rubin NVL72 (shipping Sept 2026), GB300 NVL72, HGX B300, HGX Rubin NVL8, DLC-2 direct liquid cooling, DCBBS — details: https://taknet.co.th/supermicro/gpu/nvidia-vera-rubin-nvl72/
- GPU AI servers: NVIDIA-Certified systems with NVIDIA Blackwell HGX B300 / B200
- Enterprise servers: Intel Xeon 6 and AMD EPYC platforms
- GPU workstations for 3D rendering and AI inference
- Thai-language Supermicro BMC / IPMI user guide (X14/H14, AST2600): setup, iKVM, Virtual Media, firmware update, alerts, security checklist, ipmitool/Redfish — https://taknet.co.th/supermicro/bmc/

### Storage (QSAN)
- XCubeNXT 8100 series — enterprise unified storage
- XCubeNXT 5100 series — SMB / mid-range
- Thai user guides by Taknet: XEVO 3 https://taknet.co.th/qsan/xevo3/user-guide/ · XEVO 2 https://taknet.co.th/qsan/xevo2/user-guide/ · QSM 4 (XCubeNXT) https://taknet.co.th/qsan/qsm4/user-guide/
- XCubeFAS — all-NVMe flash array
- XCubeSAN 5300 / 3300 — hybrid flash SAN (block: iSCSI, FC, NVMe-oF) with XEVO 3; dual-active controllers, 99.9999% availability, up to 650 drives; focus: large-scale surveillance, smart city, virtualization — https://taknet.co.th/qsan/
  - Model pages (D = dual-active, S = single-upgradable): XS5312 https://taknet.co.th/qsan/xs5312/ · XS5316 https://taknet.co.th/qsan/xs5316/ · XS5324 https://taknet.co.th/qsan/xs5324/ · XS5326 https://taknet.co.th/qsan/xs5326/ · XS3312 https://taknet.co.th/qsan/xs3312/ · XS3316 https://taknet.co.th/qsan/xs3316/ · XS3324 https://taknet.co.th/qsan/xs3324/ · XS3326 https://taknet.co.th/qsan/xs3326/

### Storage (TerraMaster Enterprise NAS)
- Supreme: U16-725 Plus (3U 16-bay), U12-725 Plus, U12-725, U8-725 — Intel Xeon D, ECC DDR4 up to 1 TB, 3× M.2 NVMe, 4× 10GbE SFP+ + 2× 5GbE (25GbE option); 3-year warranty per TerraMaster datasheet
- Advanced: U12-500 Plus (3-year warranty per datasheet), U8-500 Plus, U8-450 · Entry: U4-500, U8-423, U4-423 (2-year warranty)
- TOS 7 with Business Backup Suite (Centralized Backup, Duple Backup, TerraSync, CloudSync), Btrfs snapshot, HyperLock-WORM; Taknet is a TerraMaster reseller in Thailand handling sizing, installation and RMA — https://taknet.co.th/terra-master/ (per-model spec pages and datasheets linked there); TOS 7 backup & business continuity guide — https://taknet.co.th/terra-master/tos7/

### Virtualization & HCI
- Proxmox VE, Sangfor HCI, VMware vSphere/ESXi, OpenStack and Kubernetes, high-availability clusters and software-defined storage

### AI infrastructure (NVIDIA)
- NVIDIA GB300 NVL72, HGX B300, RTX PRO 6000 Blackwell clusters, AI data platform with QSAN storage, turnkey AI factory
- NVIDIA DGX Spark — MSI EdgeXpert MS-C931 (GB10, 128 GB unified memory)
- NVIDIA DGX Station — MSI XpertStation WS300 (GB300, up to 748 GB coherent memory)

### Government ICT Spec 2569
- Standard server classes and SAN storage aligned with the Thai Digital Government ICT price/spec standard (B.E. 2569, revised May 2026, items 1–11)

### CCTV storage
- Supermicro SuperStorage (PB-class), QSAN XCubeNXT / XCubeNAS for VMS platforms: Milestone XProtect, Genetec Security Center, Avigilon Alta, Hanwha Wisenet WAVE, Bosch BVMS, Dahua DSS/IVSS

## Supermicro NVIDIA Vera Rubin NVL72 — key facts
- Page: https://taknet.co.th/supermicro/gpu/nvidia-vera-rubin-nvl72/
- Model: Supermicro SRS-VR-NVL72, liquid-cooled 48U rack-scale system; Supermicro began shipping on 23 Sep 2026
- Compute: 72 NVIDIA Rubin GPUs + 36 NVIDIA Vera CPUs in one NVLink domain; 18 compute trays (4 GPU + 2 CPU each)
- Memory: 20.7 TB HBM4 (288 GB per GPU), 1.6 PB/s; up to 54 TB LPDDR5X
- Interconnect: NVLink 6, 216 TB/s (3 TB/s per GPU, 9 switch trays); ConnectX-9 1.6 Tb/s per GPU; BlueField-4 800 Gb/s; Spectrum-X Ethernet or Quantum-X800 InfiniBand
- Power & cooling: 4 × 110 kW power shelves; Supermicro DLC-2 direct liquid cooling with in-rack or in-row CDU (up to 2.2 MW per Supermicro datasheet), RDHx, L2A sidecar
- Performance target: up to 10x throughput per watt vs NVIDIA Blackwell (vendor-stated target)
- DCBBS scalable unit: 1,152 GPUs, 331 TB HBM4, 864 TB LPDDR5X, 5 MW envelope, cooling sized at 227 kW per rack, scalable to gigawatt-class AI factories
- Vera Rubin family also offered: HGX Rubin NVL8 (2U, 8 GPUs, up to 72 GPUs per rack, Vera or x86 CPU), NVIDIA Vera CPU rack (up to 256 Vera CPUs / 22,528 cores, 300 TB/s), Vera Rubin NVL4 DCBBS blueprint for FP64 HPC + AI (360 kW per rack, 1.8 MW in-row CDUs 2+1)
- Taknet role: Supermicro Authorized Distributor in Thailand — sizing, TOR, project registration and authorization letters, site survey, L12 rack delivery, on-site support and RMA. Contact sales@taknet.co.th

## GPU for Ansys Speos — key facts
- Page: https://taknet.co.th/projects/ansys/speos/ (Markdown: https://taknet.co.th/projects/ansys/speos/index.md)
- Principle: Speos GPU Compute ray-traces on NVIDIA OptiX; each GPU holds a full copy of the scene, so the largest scene = VRAM of one card. Adding GPUs scales speed (Ansys: near-linear), not memory. RTX PRO Blackwell has no NVLink.
- Cards: RTX PRO 4000 (24 GB, 70 RT cores, 145 W), RTX PRO 5000 (48 GB, 110 RT cores, 300 W), RTX PRO 6000 (96 GB, 188 RT cores, 600 W; Max-Q 300 W). Recommended usable scene ≈ 90% of VRAM (~21 / ~43 / ~86 GB)
- Speed proxy (Blender Open Data, Cycles OptiX, Blender 4.5 median, retrieved 15 Oct 2026): RTX PRO 6000 16,737; 6000 Max-Q 14,746; 5000 10,929; RTX 6000 Ada 10,655; RTX PRO 4000 6,247; RTX A5000 4,390
- Configurations (speed index vs 1× RTX PRO 4000, 90% multi-GPU efficiency): A 3× RTX PRO 5000 = 144 GB, 4.9×; B 2× RTX PRO 6000 Max-Q = 192 GB, 4.5× (recommended; ~86 GB scenes; expandable to 4× = 384 GB, 8.7×); C 4× RTX PRO 5000 = 192 GB, 6.5×
- Requirements: Speos HPC on Linux needs at least NVIDIA A5000-class GPU, driver 580.95.05 and libnvoptix; system RAM ≥ 2× per-card VRAM suggested
- Taknet role: Supermicro Authorized Distributor in Thailand — sizing, supply on Supermicro GPU platforms, burn-in, installation and after-sales. Contact sales@taknet.co.th

## MSI XpertStation WS300 — key facts

- Model: WS300T60L (MS-S429), built on NVIDIA DGX Station architecture
- Superchip: NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip
- CPU: NVIDIA Grace, 72 Arm Neoverse V2 cores, up to 496 GB LPDDR5X (396 GB/s)
- GPU: NVIDIA Blackwell Ultra, up to 252 GB HBM3e (7.1 TB/s); coherent memory up to 748 GB via NVLink-C2C
- Networking: 2x QSFP112 400 GbE (NVIDIA ConnectX-8 SuperNIC), 10GBASE-T, 1GbE dedicated BMC port
- Storage: 2x M.2 PCIe 5.0 boot drives (2x 2 TB, RAID 1), 2x M.2 PCIe 6.0 slots for data (via ConnectX-8)
- Management: ASPEED AST2600 BMC, AMI MegaRAC firmware, IPMI 2.0, DMTF Redfish
- Power: 1600 W 80 PLUS Platinum; requires a 20 A circuit. Operating temperature 10–30 °C
- Operating system: NVIDIA DGX OS 7 (Ubuntu 24.04 LTS) pre-installed
- Price in Thailand: launch promotion THB 3.89 million including VAT via Taknet (final quote depends on add-in cards, warranty and installation). Overseas retail reported at about US$85,000–99,900 (Mar–Aug 2026)

### Independent reviews (summary: https://taknet.co.th/msi-ws300/#reviews)

- StorageReview (24 Aug 2026), tested the MSI WS300 directly: "the most capable workstation we have ever reviewed". GPT-OSS-120B peak 9,294 tok/s vs 3,384 (RTX PRO 6000 workstation) vs 500 (DGX Spark); NVFP4 matmul 6,134 vs 1,449 vs 364 TFLOPS; 2.3–3x RTX PRO 6000 and 14–19x DGX Spark when serving 5 models at full concurrency; GPU peak 71 °C under liquid cooling. Nemotron-3-Ultra 550B (NVFP4, 114 GB offloaded to LPDDR5X): 43 tok/s single user. Considerations: Arm64 host, dedicated 20 A circuit, no display output without an add-in RTX PRO card, Grace-to-GPU offload bandwidth ~390 GB/s. https://www.storagereview.com/review/msi-xpertstation-ws300-review-748gb-of-coherent-memory-and-20-petaflops-on-a-desk
- ServeTheHome (6 Oct 2026), Gigabyte DGX Station GB300 (same platform): GPU-to-GPU RDMA 784 Gb/s over 2x 400G ConnectX-8; Grace CPU path ~229 Gb/s (PCIe Gen5 x8). https://www.servethehome.com/gigabyte-w775-v10-l01-hands-on-bringing-nvidia-gb300-deskside/

### News timeline (https://taknet.co.th/msi-ws300/#news)

- 2026-10-07: DGX Station for Windows shown again at Microsoft event; Windows + Linux via WSL on the same GB300 hardware; expected Q4 2026
- 2026-09-14: DGX OS 7.6.0 released (Ubuntu 24.04.5, CUDA 13.0 U3, driver 580.178); upgrades to DGX OS 8.x not yet supported on GB300 platforms
- 2026-08-27: MSI announces WS300 availability; NemoClaw + OpenShell support; two units can be clustered
- 2026-08-12: MSI adds NVIDIA Nemotron 3.5 Lightning and NeMo Switchyard model routing support
- 2026-06-01: NVIDIA announces DGX Station for Windows at GTC Taipei; MSI among OEM partners
- 2026-03-20: DGX Station GB300 orders open at GTC 2026; shipping spec 252 GB HBM3e (7.1 TB/s)

## FAQ

Q: How is the MSI WS300 different from NVIDIA DGX Spark?
A: Both use Grace Blackwell and DGX OS. DGX Spark (GB10) has 128 GB unified memory and ConnectX-7 200 GbE for individual developers. WS300 (GB300) has up to 748 GB coherent memory, 72 CPU cores, 2x 400 GbE ConnectX-8 and a BMC, suited to teams and very large models. Projects built on DGX Spark can move to WS300 because both are Arm64 with DGX OS.

Q: Can WS300 be upgraded to DGX OS 8?
A: DGX OS 8.0.0 (Ubuntu 26.04) was released on 31 Aug 2026 and DGX OS 7.6.0 (14 Sep 2026) is the upgrade path to 8.x, but NVIDIA release notes state that upgrades to 8.x are not yet supported on GB300 platforms. Taknet recommends staying on DGX OS 7 until NVIDIA announces GB300 support.

Q: How fast is the MSI WS300 in independent reviews?
A: StorageReview (Aug 2026) measured GPT-OSS-120B at 9,294 tok/s peak vs 3,384 on an RTX PRO 6000 workstation and 500 on DGX Spark, and ran Nemotron-3-Ultra 550B (NVFP4) at about 43 tok/s per user. Summary: https://taknet.co.th/msi-ws300/#reviews

Q: Can the MSI WS300 run Windows?
A: The WS300 currently ships with DGX OS 7 (Linux). NVIDIA announced DGX Station for Windows on 1 Jun 2026 with MSI among the OEMs, expected in Q4 2026, on the same GB300 hardware with Linux tools via WSL. Whether existing Linux units can be converted has not been announced.

Q: How much does the MSI WS300 cost in Thailand?
A: Launch promotion price THB 3.89 million including VAT via Taknet; the final quote depends on add-in cards, warranty and installation. Overseas retail has been reported at about US$85,000–99,900. Quotation: sales@taknet.co.th · +66 2 737 5508

Q: Can NVIDIA AI Workbench run on WS300?
A: Yes. AI Workbench supports Ubuntu 22.04/24.04/26.04 and runs on Arm64 DGX Spark; on WS300 the recommended setup is a Remote Location (Desktop App on a laptop, workloads on WS300 over SSH). Guide: https://taknet.co.th/msi-ws300/AI-workbench/

Q: How do I set up the WS300 BMC for the first time?
A: Connect the rear 1GbE dedicated management port, power on and wait about 5 minutes, find the BMC IP via DHCP or `ipmitool lan print`, then change the factory default password (a change is enforced) and configure static IP, NTP and user accounts. Guide: https://taknet.co.th/msi-ws300/manuals/#bmc-access

Q: Which AI applications can I run on WS300?
A: NVIDIA publishes DGX Station playbooks for LLM serving (vLLM, SGLang), NVFP4 quantization and pretraining, local coding agents (Claude Code with Ollama), secure agents (OpenShell, NemoClaw), knowledge graphs (txt2kg), GPU topic modeling, robotics (Isaac GR00T), image/video generation (ComfyUI) and remote access (Brev). Thai summaries: https://taknet.co.th/msi-ws300/AI-playbooks/

## Contact

For quotations, TOR support or AI solution consulting: sales@taknet.co.th · +66 2 737 5508 · https://taknet.co.th/#contact
For after-sales support: support@taknet.co.th · hotline +66 95 759 7557