Supermicro GPU SuperServer 6049GP-TRT Review
July 3, 2025We finally have the Supermicro GPU SuperServer 6049GP-TRT for you! This system is powered by dual 1st or 2nd generation intel Xeon Scalable CPUs with up to 28 cores and 56 virtual threads, each. This system can support up to 10x double-width or up to 20x single-width PCIe-based GPUs.
This 4U platform is designed for High-Performance Computing (HPC), Research Laboratories, Video Transcoding, plus the all-important AI, Machine Learning, Deep Learning and all those other AI-based workloads, to name a few. With support for dual processors and up to 20x single-wide or 10x double-wide GPUs, the Supermicro GPU SuperServer 6049GP-TRT certainly has the goods to deliver on those high-performance workloads.
There are 24x 3.5” hot-swap drive bays or up to 22x hot swap 3.5” bays with 2x 2.5” U.2 NVMe bays. An optional configuration includes up to 8x U.2 NVMe drives supported by a controller. A small control panel on the left-hand side of the chassis includes a power ON and Reset button. Below those, LED status lights for Power, HDDs NIC 1, NIC 2, Universal Information, and Power Fail.
On the back of the Supermicro GPU SuperServer 6049GP-TRT are 21x PCIe slots along the top of the chassis. Below are dual 2000W Titanium-level power supplies to either side of the chassis for 4x total. With 4x PSUs, this chassis supports 2+2 redundancy such that if 2x PSUs are disabled, the system will still remain operational allowing an Administrator to replace the failed PSUs in a timely manner.
An I/O panel supports a default configuration of a VGA port, 2x 10GbE RJ45 ports for LAN 1 and LAN2, dual USB 3.0 ports, then 2x more, and a dedicated Platform Management Interface or IPMI LAN port.
That IPMI LAN port provides remote access to the system. Supermicro provides a number of management tools to manage the system remotely including Supermicro Server Manager (SSM), Supermicro Power Manager (SPM), the Supermicro Update Manager or SUM and SuperDoctor 5 also referred to as SD5. Supermicro Server Manager is an all-in-one application for monitoring and maintaining Supermicro Servers in a data center environment. It does require licensing for each server node, and can manage up to 10,000 nodes with one SSM server. Operating systems supported include Red Hat and SUSE Linux, plus Windows Server 2012 – 2022. SuperDoctor 5 Also allows Administrators to manage and monitor systems in data centers in real-time but is used as a one-to-one option, whereas the Supermicro Server Manager is a one-to-many application.
Once we pry off the cover panel you can see the motherboard with dual sockets surrounded by 24x DDR4 DIMM slots. CPUs supported include the 1st and 2nd generation Intel Xeon Scalable processors with a Thermal Design Power (TDP) rating of up to 205W. Registered DIMMs and Load-Reduced DIMM, are supported using either those Skylake or Cascade Lake CPUs. Both 1st and 2nd generation Intel Xeon Scalable processors have 6-channel memory architecture, and with 12x slots per socket you can install 2x DIMMS per channel for up to 6TB of RDIMMMs or LRDIMMs. Only 2nd generation Intel Xeon Scalable processors, codename “Cascade Lake”, are compatible with Intel’s first generation Optane Data Centric Persistent Memory Modules or DCPMM, codename “Apache Pass, which are later called PMem 100 series. This is beginning to sound like a spy novel… Code name SuperStupid.
Intel Optane PMem modules are installed along with DRAMs or Dynamic Random Accessed Memory, either Registered or Load-Reduced. For this platform you could install 1 – 12 Optane PMem 100, paired with DRAM. For PMem 100 series, speeds of 2666MHz are supported with capacities of 128GB, 256GB, and 512GB. Using 512GB PMem modules you can install up to 6TB of Optane along with up to 3TB of DRAM using 128GB modules. Intel Optane PMem has been discontinued. They produced a PMem 200 series code named “Barlow Pass” and a PMem 300 series, codenamed “Crow Pass” but that would be the end of that experiment. We can only assume that all of these “Passes” lead to the lakes… You know “Cascade Lake”, “Sky Lake”…
Optane PMem offers three distinct modes of operation. Memory Mode which uses DRAM DIMMs as a cache for the Optane module. Memory Mode is not persistent as the DRAM modules are only used for cache and the PMem modules for memory as well, and treated as DRAM by the OS. App Direct Mode uses the DRAMs as memory and uses the Optane memory like persistent flash storage with low-latency and high-bandwidth. Mixed Mode provides a little of both, with Administrators using BIOS to determine the ratio or Optane devoted to Memory Mode and that which will be assigned to AppDirect Mode thereby gaining the best of both worlds, for frequently accessed data.
There are 20x PCIe 3.0 x16 slots for GPUs, plus another full-height, full-length x8 slot right in the middle that can be used for a MegaRAID controller. The two slots on the far right, if you are looking at the system from the front, can also be used for two separate controllers in the event you want to install the storage option with up to 8x U.2 NVME drives. Those slots could also be used to support additional NICs for faster network communications, or any of those other slots for that matter. The PCIe board or daughter board PCIe lanes, are connected to the motherboard via 4 PCIe 3.0 x16 PEX switch modules for the GPU to CPU interconnect.
Additionally, there is a single SATA DOM slot and PCIe 3.0 x4 slot on the system board for an M.2 storage device that can be used to support the OS or for additional storage.
GPUs supported on the daughter board include the NVIDIA RTX 5000, RTX 6000 or the NVIDIA T4 GPUs. The first two are double-wide cards, while that last one is a single-wide card ideal for distributed workloads, virtualization. For graphics applications, AI applications with Deep Learning or Machine Learning or even data analytics, the RTX 5000 and RTX 6000 are a good choice.
RTX refers to “Ray Tracing Texel extreme” providing real-time ray tracing. Yes, we know what you’re thinking, “what is a texel?” A texel is short for a texture element and is the smallest addressable unit of a texture map. Anyways, the RTX 6000 supports a PCIe 4.0 x16 interface, and as we have said this platform supports a PCIe 3.0 interface for support of the GPUs. With that, the GPU to CPU switch supports a queuing structure that maximizes the available bandwidth and minimizes latency.
While this is not the latest GPU server in Supermicro’s arsenal of high-performance GPU servers, it’s still no slouch and potentially a deal to be had on this system. If you’re looking for more information on this system or any other system, check out our website.
