Skip to content
EMARQUE AI

Built by EMARQUE

EMARQUE AI Server

An AI server with six or eight NVIDIA RTX PRO 6000 Blackwell GPUs, and 576 GB or 768 GB of GPU memory, in one 6.5U rackmount. Built and supported by us in Malaysia.
  • 6 or 8 × NVIDIA RTX PRO 6000 Blackwell Server Edition, 96 GB GDDR7 with ECC each
  • Up to 768 GB of GPU memory in one server
  • Dual-socket platform; up to 8 TB ECC DDR5 in the eight-GPU size
  • Hot-swap storage bays, NVMe in the eight-GPU size
  • NVIDIA ConnectX SuperNIC, up to 400 / 800 Gb/s
  • Air-cooled 6.5U rackmount, front-to-back airflow

The price depends on the number of GPUs, the memory, storage and networking you choose.

EMARQUE AI Server in the In Win IW-RG658-PRO 6.5U chassis, front three-quarter view

Product overview

Front of the IW-RG658-PRO chassis
Front: hot-swap fans over the I/O strip
Rear of the IW-RG658-PRO chassis
Rear: power supplies and the PCIe slot bank
Inside the IW-RG658-PRO chassis, lid off
Inside: the system chamber, lid off

Two sizes

The same GPUs and the same build quality; the size sets the GPU count, memory and power.

The eight-GPU AI Server in the In Win IW-RG658-PRO chassis

Eight GPUs

GPUs
8 × RTX PRO 6000 Blackwell Server Edition
GPU memory
768 GB
CPU and memory
Dual 5th-Gen Intel Xeon Scalable, up to 8 TB ECC DDR5
Storage
8 × E1.S NVMe + 4 × 2.5" hot-swap
Power
9,600 W, 3+1 redundant
Chassis
In Win IW-RG658-PRO, 6.5U
The six-GPU AI Server in the In Win IW-RG650A chassis

Six GPUs

GPUs
6 × RTX PRO 6000 Blackwell Server Edition
GPU memory
576 GB
CPU and memory
Dual-socket platform, ECC DDR5, set in your quote
Storage
Up to 4 × 2.5" SSD
Power
Confirmed in your quote
Chassis
In Win IW-RG650A, 6.5U

Key features

The IW-RG658-PRO GPU chamber, showing its bank of full-height PCIe slots

Six or eight GPUs in one server

Six or eight RTX PRO 6000 Blackwell Server Edition cards with 96 GB each, up to 768 GB in total, on PCIe Gen5. Enough GPU memory to run large models for a whole department from one machine.
Hot-swap 2.5-inch drive trays drawn out of the IW-RG658-PRO

Storage and networking to match

Eight E1.S NVMe and four 2.5-inch hot-swap bays in the eight-GPU size, up to four 2.5-inch bays in the six, and an NVIDIA ConnectX SuperNIC for linking servers, with a separate 10 GbE management port.
CRPS power supply modules part-withdrawn from the rear of the IW-RG658-PRO

Redundant power

In the eight-GPU size, 9,600 W from four supplies, one of them spare, so the server keeps running if a supply fails.
  • Separate GPUs for separate work

    There is no NVLink between the cards, so each one works on its own. Different teams and models can run on their own GPUs in the same server.

  • Configured in your quote

    Six or eight GPUs, the memory, storage and networking are set to your workload.

Use cases

  • Document search

    A private assistant for several departments

    A chat assistant that answers from your own documents, shared by finance, legal, engineering and operations, with nothing sent outside.

  • Model serving

    Several models side by side

    Run two or three models at once, each on its own pair of cards, for teams that share the server.

  • Vision and voice

    Camera and audio on site

    Video and audio streams processed in your building, spread across the cards.

  • Fine-tuning

    Serve by day, fine-tune by night

    Run your models on some cards during working hours, then use all of them for LoRA or QLoRA fine-tuning overnight.

Specifications

The final parts list is confirmed in your quote.

EMARQUE AI Server, front three-quarter view
GPU
6 or 8 × NVIDIA RTX PRO 6000 Blackwell Server Edition · 96 GB GDDR7 with ECC each (576 GB or 768 GB total)
AI compute
4 PFLOPS FP4 with sparsity per GPU, six or eight GPUs (NVIDIA figure). The cards are not linked
GPU memory bandwidth
1.6 TB/s per card
GPU interconnect
PCIe Gen5. No NVLink: each card works on its own
CPU
Eight GPUs: dual 5th-Gen Intel Xeon Scalable (up to 350 W each). Six GPUs: dual-socket platform, selected in your quote
Memory
Eight GPUs: up to 8 TB ECC DDR5 (32 DIMMs, up to 4800 MHz). Six GPUs: ECC DDR5, set in your quote
Storage
Eight GPUs: 8 × E1.S NVMe hot-swap + 4 × 2.5" hot-swap bays. Six GPUs: up to 4 × 2.5" SSD bays
Networking
NVIDIA ConnectX SuperNIC up to 400 / 800 Gb/s · 10 GbE management
Power
Eight GPUs: 9,600 W, 3+1 redundant CRPS, 80 PLUS Titanium. Six GPUs: confirmed in your quote
Cooling
Air-cooled: hot-swap fans, front-to-back airflow
Chassis
In Win 6.5U rackmount (4.5U GPU + 2U system): IW-RG658-PRO for eight GPUs, IW-RG650A for six
Software
Ubuntu LTS. NVIDIA AI Enterprise is licensed separately
Warranty
3-year EMARQUE system warranty · 90-day exchange · lifetime labour

Questions about the AI Server

  • AI server or GPU server: what is the difference?
    In practice they are the same machine: a rackmount server whose AI work runs on its GPUs. This one carries six or eight NVIDIA RTX PRO 6000 Blackwell GPUs with 96 GB each, so several models can run side by side, each on its own GPUs.
  • How much does an AI server cost in Malaysia?
    It depends on the number of GPUs, the memory, storage and networking you choose. We quote every AI Server in writing, in MYR, once we know what you want to run.
  • What is the RTX PRO 6000 Blackwell Server Edition?
    The data-centre version of NVIDIA's RTX PRO 6000 Blackwell: 96 GB of GDDR7 memory with ECC. This server uses the air-cooled version, cooled by the server's own airflow. Its power is configurable up to 600 W, so it can be matched to the server's power supply and cooling.
  • Six GPUs or eight?
    Both sizes use the same RTX PRO 6000 Server Edition cards on PCIe Gen5 in a 6.5U In Win chassis. Eight GPUs give 768 GB of GPU memory, more storage bays and 9,600 W of redundant power; six give 576 GB. We size it to your models and users in the quote.
  • How is it different from the AI PRO 500?
    The AI PRO 500 is a four-GPU workstation: Workstation Edition cards on a Threadripper PRO platform in 5U. The AI Server takes six or eight Server Edition cards on a dual-socket server platform in 6.5U.
  • Can it be configured?
    Yes. The GPU count, memory, storage and networking are set in your quote.
  • What is the warranty?
    Three years from us, covering assembly, configuration and thermal performance, with a 90-day exchange on failed parts and lifetime labour. The component makers' warranties apply on top, and we handle every claim in Malaysia.
  • What if we need the GPUs linked by NVLink?
    Look at the NVIDIA DGX B200 or DGX B300, or our AI HGX B300. Their eight GPUs are linked by NVLink, which suits training large models across all of them at once.

Get a quote for the AI Server.

Tell us what it needs to run and who will use it. We reply with a written quote in MYR.