Altair 3G16SN
3U GPU platform with sixteen hot-swap bays, built for accelerated work that has to keep its data close — inference, analytics and visual computing.
Overview
The Altair 3G16SN is a 3U accelerated compute platform with sixteen hot-swap bays, for work where the accelerator and the dataset belong in the same chassis. The 3U body is what makes that combination possible: enough vertical room for full-height dual-slot cards with the airflow to keep them fed, and enough bays that the data does not have to arrive over the network.
It is specified for inference serving, analytics and visual computing — the accelerated jobs that read a great deal and are bounded as much by how fast data reaches the card as by the card itself. Sixteen tri-mode bays take SAS, SATA and NVMe, so a fast scratch tier and a capacity tier can sit behind the accelerator in one machine.
On card fitment we are deliberately precise, and we separate two things that are often run together. Cards within the physical envelope include the NVIDIA RTX 4000 Ada Generation (111.8 x 241.3 mm, single slot, 130 W, active blower), the RTX 4000 SFF Ada Generation and RTX 2000 Ada Generation (both 68.6 x 167.6 mm, dual slot, 70 W, active) and the NVIDIA L4 (69 x 169 mm, single slot, 72 W, passive). That is a statement about dimensions, taken from each card's published specification. Final fitment, power and thermal validation is confirmed at quotation against the specific build. We do not describe a card as tested or validated in this platform unless we have tested it ourselves.
Why choose the Altair 3G16SN
- Full-height ATX accelerators up to 111.8 x 241.3 mm — the RTX 4000 Ada's own dimensions
- Cards that fit by dimension: RTX 4000 Ada, RTX 4000 SFF Ada, RTX 2000 Ada and NVIDIA L4
- Passive server-cooled and self-fanned cards both accepted
- Sixteen tri-mode bays, so the dataset sits in the same chassis as the accelerator
- Fitment by dimension is published; power and thermal validation is confirmed at quotation
Technical specifications
- Form factor
- 3U rackmount
- Accelerator envelope
- Full-height ATX, up to 111.8 mm high and 241.3 mm long — the dimensions of the NVIDIA RTX 4000 Ada Generation, which is the tallest card this 3U chassis takes. Passive server-cooled and self-fanned cards are both accepted.
- Cards that fit by dimension
- Cards within the physical envelope include the NVIDIA RTX 4000 Ada Generation (111.8 x 241.3 mm, single slot, 130 W, active blower), the RTX 4000 SFF Ada Generation and RTX 2000 Ada Generation (both 68.6 x 167.6 mm, dual slot, 70 W, active) and the NVIDIA L4 (69 x 169 mm, single slot, 72 W, passive). That is a statement about dimensions, taken from each card's published specification. Final fitment, power and thermal validation is confirmed at quotation against the specific build.
- Drive bays
- 16 hot-swap, tri-mode backplane
- Drive support
- SAS, SATA and NVMe
- Processor
- Dual-socket Intel Xeon Scalable
- Cooling
- Front-to-back airflow sized for accelerator load
- Power
- Redundant hot-swap power supplies, sized with the card count
- Management
- Dedicated out-of-band management port
- Warranty
- 1 to 5 years as per terms
Why 3U, and which cards fit it
The height is the specification. This 3U chassis accepts a full-height ATX card up to 111.8 mm high and 241.3 mm long — the dimensions of the NVIDIA RTX 4000 Ada Generation, which is the tallest card it takes. Cards within the physical envelope include the NVIDIA RTX 4000 Ada Generation (111.8 x 241.3 mm, single slot, 130 W, active blower), the RTX 4000 SFF Ada Generation and RTX 2000 Ada Generation (both 68.6 x 167.6 mm, dual slot, 70 W, active) and the NVIDIA L4 (69 x 169 mm, single slot, 72 W, passive). Both cooling arrangements are accepted. A passive card such as the L4 has no fan of its own and is cooled by the chassis airflow; a self-fanned card such as the RTX 4000 Ada or RTX 4000 SFF Ada carries its own blower. The two need different things from the fan curve, and many chassis quietly suit only one.
Accelerator and data in one machine
An accelerated job that reads constantly is often limited by how fast data reaches the card. Sixteen tri-mode bays let a fast NVMe scratch tier and a capacity tier sit behind the accelerator in the same chassis, which removes the network from the inner loop for the workloads where that is the bottleneck.
What is published, and what is confirmed at quotation
Two different statements, and we keep them apart. WHAT FITS BY DIMENSION is a fact you can check against any card's published specification, so it is published here: height, length and slot width against the envelope this chassis provides. WHETHER A GIVEN CARD IS VALIDATED IN A GIVEN BUILD is a different question, and it depends on the power budget, the airflow at your ambient, and the rest of the configuration. That is confirmed at quotation against the actual build. We do not describe a card as tested or validated in this platform unless we have tested it ourselves.