Enterprise Solutions · Data & AI
NVIDIA DGX Spark in Malaysia — every OEM build, quoted in RM
A petaFLOP of AI compute and 128 GB of unified memory in a 1.2 kg box that runs off a wall socket. We supply the NVIDIA Founders Edition and every OEM version — ASUS, Dell, HP, Lenovo, Acer, Gigabyte and MSI — through authorised Malaysian distribution, with SST and MyInvois handled.
On promotion now (August 2026)
The OEM builds from Dell, Lenovo, ASUS, Acer and HP are on promotional pricing through our distribution channel this period. Promotional windows on GB10 systems are short and allocation-dependent — ask for current numbers before committing to a budget.
What DGX Spark actually is
A desktop AI supercomputer built on the NVIDIA GB10 Grace Blackwell Superchip. The point of it is the 128 GB of memory shared coherently between CPU and GPU: it lets you hold models on your own desk that would otherwise need rented data-centre time. These specifications are fixed by NVIDIA and identical in every OEM build.
- Superchip
- NVIDIA GB10 Grace Blackwell
- CPU
- 20-core Arm — 10× Cortex-X925 + 10× Cortex-A725
- GPU
- Blackwell architecture, 5th-gen Tensor Cores
- AI performance
- Up to 1 petaFLOP at FP4
- Unified memory
- 128 GB LPDDR5x, CPU/GPU coherent
- Memory bandwidth
- 273 GB/s
- Networking
- ConnectX-7 at 200 Gbps, plus 10 GbE
- Power
- 240 W PSU (GB10 TDP 140 W)
- Size and weight
- 150 × 150 × 50.5 mm, 1.2 kg
- Operating system
- NVIDIA DGX OS
What 128 GB buys you
Inference, one unit
Models up to 200 billion parameters
Fine-tuning, one unit
Models up to 70 billion parameters
Two units linked
Models up to 405 billion parameters
Figures published by NVIDIA for the GB10 platform.
Every DGX Spark model we can quote
NVIDIA fixes the superchip, the 128 GB of unified memory and the ConnectX-7 networking across all of these, so raw compute does not change with the badge. OEMs differ on storage capacity, cooling and acoustics, chassis, warranty terms and remote management.
NVIDIA
DGX Spark (Founders Edition)
The reference build, with 4 TB self-encrypting NVMe. The baseline every OEM version is measured against.
ASUS
Ascent GX10
ASUS's build of the same GB10 platform, offered in more than one storage capacity.
Dell Technologies
Pro Max with GB10
Dell's build, sold into the same fleet and support arrangements as the rest of a Dell estate.
HP
ZGX Nano AI Station (G1n)
HP's Z-series take on the platform, aimed at workstation buyers already standardised on Z.
Lenovo
ThinkStation PGX
Lenovo's first GB10 workstation, carrying ThinkStation warranty and service options.
Acer
Veriton GN100 AI Mini Workstation
Acer's compact build of the same platform.
Gigabyte
AI TOP ATOM
Gigabyte's GB10 build, listed by NVIDIA alongside the others.
MSI
EdgeXpert (MS-C931)
MSI's GB10 build, sold into industrial and edge deployments as well as the desk.
When DGX Spark is the wrong purchase
Memory bandwidth is 273 GB/s. That is generous for the size and price class and it is why the machine is excellent for prototyping, for fine-tuning up to around 70 billion parameters, and for keeping data on your own premises. But bandwidth, not capacity, is what governs inference throughput once a model is loaded — and a data-centre GPU has several times more of it.
So if the plan is high-volume production inference for many concurrent users, this is the wrong box and we will say so before you buy it rather than after. We would rather quote you the right thing than the expensive thing.
How buying works in Malaysia
Company purchase, not a consumer checkout — the paperwork a finance department needs is part of the quote.
- 01
Tell us the model
Which OEM build, how many, and whether you need two linked.
- 02
Quotation in RM
Priced locally, usually within two working days, SST stated.
- 03
Purchase order
Standard PO terms; MyInvois e-invoice issued.
- 04
Authorised supply
Sourced through authorised Malaysian distribution.
- 05
Delivered and set up
Racked or on a desk, on your network, DGX OS ready.
- 06
Put to work
Optional: we get a model serving something useful behind it.
Buying the box is the easy half
Most DGX Spark units sold in Malaysia will spend their first six months running a demo. We are a member of the Anthropic Claude Partner Network and we build agentic AI systems that run in production, so we can quote the hardware and then make it earn its keep — a model serving a real workflow behind your own network, not a benchmark screenshot.
Common questions
- Where can I buy NVIDIA DGX Spark in Malaysia?
- Anchor Sprint supplies DGX Spark in Malaysia through authorised distribution, in both the NVIDIA Founders Edition and every OEM build — ASUS Ascent GX10, Dell Pro Max with GB10, HP ZGX Nano, Lenovo ThinkStation PGX, Acer Veriton GN100, Gigabyte AI TOP ATOM and MSI. Send the model and quantity through the enquiry form and we return a quotation in Ringgit with SST treatment stated and a MyInvois e-invoice on purchase.
- How much does a DGX Spark cost in Malaysia?
- We do not publish prices. DGX Spark pricing moves with the OEM build, storage capacity and the promotion running at the time, so it is quotation-based. Send us the model you want and we will come back in Ringgit, usually within two working days.
- What is the difference between the OEM versions of DGX Spark?
- Less than most buyers expect. NVIDIA fixes the GB10 superchip, the 128 GB of unified memory and the ConnectX-7 networking across every build, so raw compute is the same whichever badge is on the front. OEMs differentiate on storage capacity, cooling and acoustics, chassis design, warranty terms and remote management. In practice the right choice is usually whichever vendor your existing support and procurement arrangements already cover.
- Is DGX Spark the same thing as RTX Spark?
- No — they are two different NVIDIA products with confusingly similar names, and buyers mix them up constantly. DGX Spark is the GB10 Grace Blackwell developer workstation on this page: it runs NVIDIA DGX OS, which is Linux, and it exists to hold and fine-tune large models locally. RTX Spark is NVIDIA's SoC family for Windows-on-Arm AI PCs, announced at Computex 2026 and built into laptops and desktops by the usual PC makers. The silicon is related, but one is a developer workstation and the other is a personal computer. If you are shopping for a laptop, you want RTX Spark; if you want to run a 200-billion-parameter model on your own desk, you want DGX Spark. Tell us the workload and we will tell you which.
- Is DGX Spark a replacement for a data-centre GPU server?
- No, and it is worth being clear about that before you buy. DGX Spark has 128 GB of unified memory at 273 GB/s. That is a lot of capacity for the size and price class, which is what makes it excellent for prototyping, fine-tuning up to about 70 billion parameters and running models locally. But memory bandwidth, not capacity, is what governs large-model inference throughput, and a data-centre GPU has several times more. If your goal is high-volume production inference for many concurrent users, a Spark is the wrong tool and we will tell you so.
- Can I link two DGX Spark units together?
- Yes. The ConnectX-7 networking at 200 Gbps exists for exactly this, and NVIDIA states that two connected units work with models up to 405 billion parameters. If that is your plan, tell us at enquiry stage so the quotation includes the cabling and the second unit is matched.
- Do I need a DGX Spark, or should I just use cloud GPUs?
- It depends on utilisation and data residency, not on preference. A Spark is capital you own and can run at full utilisation without metering, which suits teams developing continuously or handling data that should not leave the building under PDPA. Cloud suits bursty workloads and anything needing more bandwidth than a Spark has. We build and operate AI systems in production, so we can tell you which case you are in before you commit to hardware.
- Does the quotation include SST and a MyInvois e-invoice?
- Yes. Quotations are issued in Ringgit with SST treatment stated, purchase orders run on standard local terms, and invoices are issued through MyInvois in line with Malaysia's e-invoicing requirements.
- Can Anchor Sprint help set it up and actually use it?
- Yes, and that is the part most resellers cannot do. We are a member of the Anthropic Claude Partner Network and we build agentic AI systems in production. Supplying the box is the easy half; getting a model serving something useful behind your own network is the half that decides whether the purchase pays for itself.
Get a DGX Spark quotation in Ringgit
Tell us the model and quantity. We come back in RM with SST stated, usually within two working days — and we will tell you if a Spark is not the right machine for what you described.
- Every OEM build quotable
- Quoted in RM, SST stated
- MyInvois e-invoice on purchase
- Authorised Malaysian distribution
NVIDIA, DGX, DGX Spark, Grace, Blackwell and ConnectX are trademarks of NVIDIA Corporation. ThinkStation is a trademark of Lenovo; Ascent of ASUSTeK; Pro Max of Dell Technologies; ZGX of HP Inc.; Veriton of Acer; AI TOP of GIGABYTE. Anchor Sprint is an independent reseller sourcing through authorised Malaysian distribution and is not affiliated with, endorsed by or a partner of NVIDIA Corporation or these manufacturers.