Supermicro AS-3015TR-I4 Edge GPU Tower

A quiet 9-litre slim tower that takes a single GPU, so a branch or an office can run accelerated AI inference locally without a server room or its noise.

Price on request · Talk to a Sirap engineer or call (+356) 2138 5911.

Sirap, your Malta partner for Supermicro, with local specification, installation, service and support.

What the AS-3015TR-i4 is for

The AS-3015TR-i4 is the GPU member of Supermicro’s AMD EPYC 4005 edge family: a slim-tower system in a 9-litre chassis, built quiet enough to sit in a working office and able to take one dual-slot GPU for local acceleration. It is the box for an edge site whose AI workload is heavier than a CPU can comfortably serve, computer vision, larger local models, real-time inference, but which has nowhere to put a rack server and no tolerance for data-centre noise. With a GPU such as an NVIDIA RTX PRO 2000 Blackwell fitted, it runs accelerated inference next to the data that feeds it. It is the local-acceleration option in our guide to on-premise AI compute.

Who it is for

It suits finance and iGaming sites that need real-time, GPU-accelerated inference held locally, plus offices doing vision, analytics or model work where a rack is not an option. For sites that only need CPU inference or a POS/gateway role, the smaller AS-E300-14GR is more proportionate; for rack-mounted back-office consolidation, see the AS-1116R-FN4. When the GPU workload outgrows a single accelerator, it is time to move up to a core GPU server.

Key features

Takes a real GPU. The chassis houses one dual-slot GPU card (up to roughly 2.7in high by 6.6in long), for example an NVIDIA RTX PRO 2000 Blackwell, so local inference is genuinely accelerated rather than CPU-bound.

Office-quiet, office-sized. A 9-litre slim tower designed for low-noise operation sits beside a desk or in a small office, not only in a comms room.

Flexible local storage. Optional slim optical and a 3.5in disk drive give the box room for local data and media alongside its primary storage.

Secure and remotely managed. TPM 2.0 and AMD Secure Encrypted Virtualization protect the data, with IPMI 2.0 out-of-band management for remote administration, the same security footing as the rest of the EPYC 4005 edge family.

Technical specifications

Specifications confirmed from the Supermicro AMD EPYC 4005 launch and the EPYC 4000 series product page. The GPU model is an example of a supported card, not the only option; Sirap confirms GPU choice and the full build against the current datasheet and your workload as part of the quote.

What is in scope from Sirap

For the AS-3015TR-i4 the decision that matters is the GPU, and getting it right is where we earn our place: we match the accelerator to the actual inference workload, neither starving it nor paying for a card the model never uses, then build, thermally validate for quiet office operation, and support the unit from Malta. Where the workload is borderline, we will tell you honestly whether this tower or a proper GPU server is the right home for it.

Often paired with

• The CPU-only AS-E300-14GR and rack AS-1116R-FN4 across a mixed-site estate, and a core GPU server when one card is no longer enough.
• An Eaton UPS sized for the GPU’s real draw, because an accelerated box has a peakier load than a plain office PC.
• Central DataCore storage the edge inference reports back to.

Could this be grant-funded?

Bringing AI acceleration into local sites is a digitisation investment that can align with Malta Enterprise schemes such as the Digitalise Your SME grant or the Smart & Sustainable Investment Grant. Fit depends on the project. Read our guide and see if your project qualifies, or talk to a Sirap engineer.

Frequently asked questions

Which GPU does it take?
One dual-slot card up to roughly 2.7in by 6.6in, such as an NVIDIA RTX PRO 2000 Blackwell. The right card depends on your inference workload, which we size with you rather than defaulting to the biggest option.

Is it quiet enough for an office?
Yes, it is a low-noise slim tower designed to run outside a server room. We thermally validate the GPU configuration so it stays quiet under sustained load rather than only at idle.

When should I choose a GPU server instead?
When one GPU is no longer enough, or when several sites’ inference is better consolidated centrally. At that point a 4U Universal GPU server or the X14 5U NVIDIA GPU server is the right step, and we will say so.

Talk to a Sirap engineer

Tell us the inference workload and where the box has to live, and we will match the GPU, build and validate it, and support it from here in Malta.

Request a quote · Price on request