TORmem
Technology Products Applications Team Blog Careers
Partner With Us

OUR PRODUCTS

TORmem delivers patented, memory-centric infrastructure built for AI factories, hyperscale data centers, and HPC. Our platform transforms memory into a shared, scalable, high-performance fabric resource for next-generation AI systems.

2G

SERIES: AI EDGE
TOR-AI-Edge-2G server

High-performance AI inference

TOR-AI-Edge-2G

Enterprise-grade AI edge server engineered for demanding AI inference, large language model deployment, computer vision, and high-performance computing workloads.

WORKLOAD
AI Inference / LLMs
DEPLOYMENT
Edge / On-Prem
Processor
AMD Threadripper PRO 9975WX, 32 cores
Graphics
Dual NVIDIA RTX 5090 32GB GPUs
Memory
256GB DDR5-6400
Networking
ConnectX-7, 200GbE / NDR200 InfiniBand
  • Dual RTX 5090 acceleration for AI and GPU computing
  • Dual 3.84TB Samsung NVMe SSD storage
  • Optimized GPU thermal and performance management
  • Shipment records for firmware, kernel modules, OS, and packages
P/N: TME-2G-AMD-256-4T-R5090

ODM / OEM availability

1U1G2E-TURIN

1U1G2E-TURIN

GPU server

Complete DVT (design validation test)

NVIDIA RTX PRO™ 6000 Blackwell Server Edition NVIDIA RTX PRO™ 6000 Blackwell Max-Q Workstation Edition able to contain 1 GPU + 1 CX7

1U Rackmount with 1+1, 80-PLUS Titanium, 2000W CRPS Single Socket SP5 (LGA 6096), supports AMD EPYC™ 9005/9004 (with AMD 3D V Cache™ Technology) and 97x4 series processors 8 DIMM slots (1DPC), supports DDR5 RDIMM, RDIMM-3DS 2 Hot-swap 2.5" NVMe (PCIe4.0 x4) drive bays 1 FHFL dual-slot PCIe5.0 x16, 1 HHHL single-slot PCIe5.0 x16 2 M.2 (PCIe5.0 x4 or SATA 6Gb/s) 1 OCP NIC 3.0 (PCIe5.0 x16) Remote management (IPMI)

MOQ: Standard Model: No minimum order quantity applies. Availability will be confirmed based on real-time stock status.Std lead: 16-20 wks
2U2G-TURIN/HPR

2U2G-TURIN/HPR

GPU server

Mass production

NVIDIA RTX PRO™ 6000 Blackwell Server Edition NVIDIA RTX PRO™ 6000 Blackwell Max-Q Workstation Edition NVIDIA H200 NVL able to contain 2 GPU + 1 CX7

2U Rackmount with 1+1 redundant 2700W CRPS, 80-PLUS Titanium Single Socket SP5 (LGA 6096P), supports AMD EPYC™ 9005/9004 (with AMD 3D V-Cache™ Technology) and 97x4 series processors 12 DIMM slots (1DPC), supports DDR5 RDIMM, RDIMM-3DS 2 Hot-swap 2.5" NVMe (PCIe4.0 x4) drive bays 2 FHFL dual-slot PCIe5.0 x16, supports GPU and accelerator cards of up to 600W TDP each (25°C ambient temp.) 1 HHHL PCIe5.0 x16 1 OCP NIC 3.0 (PCIe5.0 x16) 2 M.2 (PCIe5.0 x4 or SATA 6Gb/s) Remote management (IPMI)

MOQ: Standard Model: No minimum order quantity applies. Availability will be confirmed based on real-time stock status.Std lead: 16-20 wks
4U4G-TURIN/HPR

4U4G-TURIN/HPR

GPU server

Complete DVT (design validation test)

4U rackmount GPU server. Designed for AI / HPC / inference clusters

Ready for PVT. Will be released

MOQ: Standard Model: No minimum order quantity applies. Availability will be confirmed based on real-time stock status.Std lead: 16-20 wks
4U8G-TURIN2/RF+

4U8G-TURIN2/RF+

GPU server

Complete DVT (design validation test)

NVIDIA RTX PRO™ 6000 Blackwell Server Edition NVIDIA RTX PRO™ 6000 Blackwell Max-Q Workstation Edition NVIDIA H200 NVL

Enhanced airflow for reliable cooling of 600W GPUs and accelerator cards 4U Rackmount with 3+1, 80-PLUS Titanium, 2700W CRPS Dual Socket SP5 (LGA 6096), supports AMD EPYC™ 9005/9004 (with AMD 3D V-Cache™ Technology) and 97x4 series processors 12+12 DIMM slots (1DPC), supports DDR5 RDIMM, RDIMM-3DS 4 hot-swap 2.5" NVMe (PCIe5.0 x4) drive bays, 20 Hot-swap 2.5" SATA drive bays 8 FHFL dual-slot PCIe5.0 x16, 1 FHHL PCIe5.0 x8 Supports 2 M.2 (PCIe3.0 x4 or SATA 6Gb/s) 1 OCP NIC 3.0 (PCIe5.0 x8) 2 RJ45 (1GbE) by Intel® i350 Remote Management (IPMI)

MOQ: Standard Model: No minimum order quantity applies. Availability will be confirmed based on real-time stock status.Std lead: 16-20 wks
1U12E-GENOA/2L2T

1U12E-GENOA/2L2T

Compute server /Storage server

Mass production

all flash storage, for compute server & storage server & memory appliance

1U Rackmount with 1+1, 80-PLUS Platinum, 1000W Slim PSU Single Socket SP5 (LGA 6096), supports AMD EPYC™ 9005*/9004 (with AMD 3D V-Cache™ Technology) and 97x4 series processors 24 DIMM slots (2DPC), supports DDR5 RDIMM, RDIMM-3DS 12 hot-swap 2.5" NVMe (PCIe5.0 x4)/SATA drive bays 2 FHHL PCIe5.0 x16 Supports 2 M.2 (PCIe5.0 x4) 1 OCP NIC 3.0 (PCIe5.0 x16) 2 RJ45 (10GbE) by Broadcom BCM57416 2 RJ45 (1GbE) by Intel® i210 Remote Management (IPMI)

MOQ: Standard Model: No minimum order quantity applies. Availability will be confirmed based on real-time stock status.Std lead: 16-20 wks
1U12E-TURIN/EVAC

1U12E-TURIN/EVAC

Storage server

Mass production

all flash storage, single socket

1U Rackmount with 1+1 hot-swap 1600W slim RPSU Single Socket SP5 (LGA 6096), supports AMD EPYC™ 9005/9004 (with AMD 3D V-Cache™ Technology) and 97x4 series processors 12 DIMM slots (1DPC), supports DDR5 RDIMM, RDIMM-3DS 12 hot-swap 2.5" NVMe (PCIe5.0 x4) drive bays 1 FHHL PCIe5.0 / CXL2.0 x16 Supports 2 M.2 (PCIe5.0 x4 or SATA 6Gb/s) 1 OCP NIC 3.0 (PCIe5.0 / CXL2.0 x16) Remote management (IPMI)

MOQ: Standard Model: No minimum order quantity applies. Availability will be confirmed based on real-time stock status.Std lead: 16-20 wks
1U12E-TURIN2/EVAC

1U12E-TURIN2/EVAC

Storage server

Mass production

all flash storage, dual socket

1U Rackmount with 1+1, 80-PLUS Titanium, 2000W CRPS Dual Socket SP5 (LGA 6096), supports AMD EPYC™ 9005/9004 (with AMD 3D V-Cache™ Technology) and 97x4 series processors 12+12 DIMM slots (1DPC), supports DDR5 RDIMM, RDIMM-3DS 12 hot-swap 2.5" NVMe (PCIe5.0 x4) drive bays 2 FHHL PCIe5.0 x16 Supports 2 M.2 (PCIe3.0 x4) 1 OCP NIC 3.0 (PCIe5.0 x16) 2 RJ45 (1GbE) by Intel® i350 Remote Management (IPMI)

MOQ: Standard Model: No minimum order quantity applies. Availability will be confirmed based on real-time stock status.Std lead: 16-20 wks
4U60L8E-GENOA/2T

4U60L8E-GENOA/2T

Storage server

Mass production

Cold storage and backup solution

4U Rackmount with 2+2, 80-PLUS Platinum, 800W CRPS Single Socket SP5 (LGA 6096), supports AMD EPYC™ 9005*/9004 (with AMD 3D V-Cache™ Technology) and 97x4 series processors 8 DIMM slots (1DPC), supports DDR5 RDIMM, RDIMM-3DS Up to 60 Hot-swap 3.5" SATA/SAS drive bays 8 Hot-swap 2.5" NVMe (PCIe5.0 x4) drive bays 4 FHHL PCIe5.0 x16 Supports 2 M.2 (PCIe5.0 x4) 2 RJ45 (10GbE) by Intel® X550-AT2 Remote management (IPMI)

MOQ: Standard Model: No minimum order quantity applies. Availability will be confirmed based on real-time stock status.Std lead: 16-20 wks
2U12L8E-GENOA2

2U12L8E-GENOA2

Replacement Dell PowerEdge R7625

Mass production

Replacement Dell PowerEdge R7625

2U Rackmount with 1+1, 80-PLUS Platinum, 2000W CRPS Dual Socket SP5 (LGA 6096), supports AMD EPYC™ 9005*/9004 (with AMD 3D V-Cache™ Technology) and 97x4 series processors 12+12 DIMM slots (1DPC), supports DDR5 RDIMM, RDIMM-3DS 8 hot-swap 3.5"/2.5" NVMe (PCIe4.0 x4)/SATA drive bays 4 hot-swap 3.5"/2.5" SATA drive bays 2 fixed 2.5" NVMe (PCIe5.0 x4) drive bays 2 FHHL PCIe5.0 x16 or 1 FHHL PCIe5.0 x16, 2 FHHL PCIe5.0 x8 2 FHHL PCIe5.0 x16 or 1 FHHL PCIe5.0 x16, 2 FHHL PCIe5.0 x8 1 HHHL PCIe5.0 x16 or 2 HHHL PCIe5.0 x8 Supports 2 M.2 (PCIe3.0 x4) 2 RJ45 (1GbE) by Intel® i350 1 OCP NIC 3.0 (PCIe5.0 x16) Remote Management (IPMI)

MOQ: Standard Model: No minimum order quantity applies. Availability will be confirmed based on real-time stock status.Std lead: 16-20 wks

Memory Appliances

Multi-terabyte, RDMA-powered memory systems operating as rack-level and fabric-level memory pools for CPU and GPU clusters — enabling true disaggregated memory, dynamic expansion, and shared memory services.

AI Infrastructure Platforms

Integrated solutions combining TORmem memory, GPU/CPU servers, and high-speed networking to deliver turnkey platforms for AI factories, large-scale inference/training, and HPC.

Fabric Software Platform

A software-defined control layer that orchestrates disaggregated memory across CPU, GPU, and AI workloads — enabling pooling, telemetry, and memory-centric operations.

DESIGNED FOR MODERN DATA CENTERS

Vendor-Agnostic

x86, ARM, GPU, heterogeneous accelerators

Deployable Today

Production-grade RDMA architecture

Scalable

From a single rack to hyperscale AI factories

M1000

SERIES: M1000
M1000 System

M1000 Memory Appliance

The TORmem M1000 is a production-ready memory disaggregation appliance built on AMD Turin processors and high-capacity DDR5 memory.

M1000 delivers shared memory to multiple servers over fast Ethernet fabrics using RDMA and RoCEv2 lossless networking, enabling remote memory to behave as closely as possible to local system memory.

Designed for modern AI and in-memory workloads, M1000 allows enterprises to expand memory capacity, increase utilization, and eliminate the cost and inefficiency of over-provisioned servers.

The M1000 is the foundation of TORmem’s memory-centric architecture—unlocking larger models, faster in-memory workloads, and a more efficient path to scaling next-generation data-center infrastructure.

SWITCH

TORswitch 100G-800G

TORswitch

High-Performance. Open. Enterprise-Ready. 32-Port 100G Switch Optimized for AI, Storage, and Cloud Networking.

  • High-speed disaggregated memory bank
  • Custom FPGA acceleration
  • Low-latency interconnect
TORswitch Front

M4000 PROTOTYPE

M4000 Front View
FRONT VIEW
M4000 Rear View
REAR VIEW

These real images of our product show the standard 19" 2 RU form factor which will be familiar to anyone who has deployed equipment in a data center. The backplane of the device hosts the high-speed connectivity to the rest of the rack, and each memory bank can be swapped out at the front. From a hardware perspective, dual hot-swappable power supplies are available.

L11 RACK

SYSTEM: L11 INTEGRATION
Full L11 AI Rack Integration

Full L11 AI Rack Integration

Built, Validated, and Ready for Deployment.

TORmem offers complete rack-level system integration, incorporating GPU servers, storage server clusters, high-speed networking fabrics, and our patented memory-centric disaggregation hardware into a unified compute engine.

Every L11 system undergoes rigorous thermal profiling, packet-loss latency testing under deep queues, and synthetic AI training workloads before being shipped directly to your data center.

  • Turnkey L11 rack-scale integration
  • Heterogeneous computing nodes support
  • Complete cable diagnostic validation

CERTIFIED

SERIES: CERTIFIED PLATFORMS

TORmem Certified Platforms

TORmem Certified Platforms are systems built with our ODM partners and validated by TORmem to meet performance, reliability, and compatibility standards for AI, HPC, and memory-centric workloads.

4U AI & HPC Server
4U AI & HPC Server

4U AI & HPC Server

  • Enhanced airflow for reliable cooling of 600W GPUs and accelerator cards
  • 4U Rackmount with 3+1 redundant 3200W CRPS
  • Dual Socket SP5 (LGA 6096), supports AMD EPYC™ 9005/9004 (with AMD 3D V-Cache™ Technology) and 97x4 series processors
  • 12+12 DIMM slots (1DPC), supports DDR5 RDIMM, RDIMM-3DS
  • 16 hot-swap 2.5" NVMe (PCIe5.0 x4) drive bays or 24 Hot-swap 2.5" SATA/SAS* drive bays *Additional RAID/HBA card required
  • 10 FHFL dual-slot PCIe5.0 x16 + 2 FHFL single-slot PCIe5.0 x8 shared space with adjacent dual-slot (with PEX89104 PCIe Switch)
  • 1 FHFL single-slot PCIe5.0 x16 (CPU direct link), 2 FHHL PCIe5.0 x16
  • Supports 2 M.2 (PCIe3.0 x4 or SATA 6Gb/s)
  • 1 OCP NIC 3.0 (PCIe5.0 x16), 2 RJ45 (1GbE) by Intel® i350
  • Remote Management (IPMI)

1U High-Performance Server

  • 1U Rackmount with 1+1, 80-PLUS Platinum, 2000W CRPS
  • Dual Socket SP5 (LGA 6096), supports AMD EPYC™ 9005/9004 (with AMD 3D V-Cache™ Technology) and 97x4 series processors
  • 12+12 DIMM slots (1DPC), supports DDR5 RDIMM, RDIMM-3DS
  • 12 hot-swap 2.5" NVMe (PCIe5.0 x4) drive bays
  • 2 FHHL PCIe5.0 x16
  • Supports 2 M.2 (PCIe3.0 x4)
  • 1 OCP NIC 3.0 (PCIe5.0 x16)
  • 2 RJ45 (1GbE) by Intel® i350
  • Remote Management (IPMI)

1U Intel Xeon 6 Server

  • 1U Rackmount with 1+1, 80-PLUS Platinum, 1600W RPSU
  • Single Socket BR (LGA 7529), supports Intel Xeon 6900P-series processors
  • 12 DIMM slots (1DPC), supports DDR5 RDIMM, RDIMM-3DS, MCR DIMM
  • 8 Hot-swap E1.S (PCIe5.0 x4) drive bays
  • 1 FHHL PCIe5.0 x16, 1 HHHL PCIe5.0 x16
  • 1 OCP NIC 3.0 (PCIe5.0 x16)
  • Supports 2 M.2 (PCIe5.0 x4)
  • 1 DC-SCM 2.0 module
  • Remote management (IPMI)

4U Multi-Node Server

  • 4U chassis with 18 nodes support 36 x 2.5'' NVMe SSD, 2000W Redundant (2+2) PSU
  • Supports AMD EPYC™ 4005/4004 and AMD Ryzen™ 9000/8000/7000 Series Processors (65W)
  • 4 DDR5 ECC/ non-ECC UDIMM, up to 48GB each
  • Supports 2 x 2.5" NVMe SSD per node
  • Supports 1 M.2 per node
  • Integrated IPMI 2.0 and KVM with Dedicated LAN
  • 2x RJ45 (10GbE) by Broadcom BCM57416)

EXECUTION ROADMAP (2026)

Focused delivery milestones supporting PoC readiness and production deployment.

All milestones are backed by active development, system integration, and ongoing customer evaluation.

Q1 2026 — Platform Availability (PoC)

AI Edge Pro Server

Under active customer PoC and evaluation for real-world AI inference workloads.

SONiC Software

Available for deployment on TORmem networking platforms.

100G / 400G Ethernet

Production-ready hardware platform validated for high-performance data and memory traffic.

RDMA / RoCEv2 Networking Stack

Integrated and validated for high-throughput, low-latency memory traffic.

Memory Disaggregation Platform

Available for customer PoCs, enabling scalable memory expansion over Ethernet.

Q2 2026 — Capacity Expansion

48TB Appliance

Platform design initiated; development milestone in progress to support next-generation large-memory workloads.

Roadmap Planning 1

48TB Appliance

Roadmap Planning 2

48TB Appliance

Roadmap reflects current execution targets and may evolve based on validation results, supply-chain readiness, and customer feedback.

SOFTWARE INTELLIGENCE

TORmem CONTROL CENTER

SERVER STATUS

Server-01
ONLINE
Server-02
ONLINE
Server-03
ALLOCATING
Server-04IDLE

MEMORY POOL

12TBACTIVE
ALLOCATED
8.2 TB
AVAILABLE
3.8 TB
Client Allocation
ALLOCATION MANAGER
Client Screen
CLIENT METRICS

Our product is supported by a robust memory manager which allows the administrator to monitor, control and optimize the memory allocation given to any particular client server in the rack in real-time. We believe that visibility for the administrator is a focus of our development.

WHY TORmem

  • Patented memory disaggregation technology
  • Real, production-grade RDMA platform
  • Built for AI, hyperscale, and HPC
  • Designed by data-center and hardware veterans
  • Optimized for performance and long-term TCO
OUR VISION

TORmem’s Mission

TORmem is building the memory foundation of the AI era.

"We exist to ensure that memory never limits intelligence — bringing AI infrastructure to everyone."

BENCHMARKS & COMPARISON

"We believe that we are the farthest along in solving the challenge of one memory for all. As we refine our design and product capabilities, we will share performance figures with the industry that validate our unique approach."

STATUS: IN VALIDATION

WHY DISAGGREGATED MEMORY?