sales@emi-ic.com
Contact Us >>
sales@emi-ic.com
Contact Us >>
Means of Payment
For your convenience, we accept multiple payment methods in USD, including PayPal, Credit Card, and wire transfer.
RFQ (Request for Quotations)
It is recommended to request for quotations to get the latest prices and inventories about the part.
Our sales will reply to your request by email within 24 hours.
IMPORTANT NOTICE
1. You’ll receive an order information email in your inbox. (Please remember to check the spam folder if you didn’t hear from us).
2. Since inventories and prices may fluctuate to some extent, the sales manager is going to reconfirm the order and let you know if there are any updates.
| Category | Specification | Details |
|---|---|---|
| Model | Product Line | Optaneia Compute Fabric Series |
| Model Number | R2496 | |
| Chassis | Form Factor | 3U Compute Array |
| Architecture | Non-von Neumann; Data-Centric “Data Fabric” | |
| Compute Fabric | Technology | Photonic Tensor Cores integrated with FeRAM/MRAM |
| Compute Principle | In-Memory Computing (IMC) & Photonic Processing | |
| Precision | Configurable (1-8 bit for AI, 16-32 bit for HPC) | |
| Operations | Matrix multiplications, convolutions, graph operations | |
| Memory | Technology | Non-Volatile Memory (FeRAM/MRAM) |
| Capacity per Array | 512 TB (effective “active” memory) | |
| Bandwidth | > 100 TB/s (within the compute fabric) | |
| Classical Interface | Host Processors | 2 x 6th Gen Intel® Xeon® CPUs (for control flow) |
| Host Interface | CXL 4.0 over PCIe 7.0 | |
| Host Memory | 1 TB DDR6 | |
| Networking | Inter-Array Fabric | Optical I/O (1.6 Tb/s per port) |
| Protocol | Proprietary low-latency protocol for model parallelism | |
| Performance | Throughput | > 100 PetaOPS (INT8) |
| Energy Efficiency | ~100 TOPs/Watt | |
| Latency | Nanosecond-scale for in-fabric operations | |
| Power | Power Supply | 1600W (mostly for host CPUs and I/O) |
| Compute Fabric Power | < 5 pJ/op | |
| Cooling | Cooling System | Liquid Cold Plate (targeted for photonic components) |
| Physical | Dimensions | (H x W x D) 131 mm x 447 mm x 800 mm |
| Weight | ~40 kg (88 lbs) | |
| Programming | Model | Dataflow Programming Interface |
| Framework Support | PyTorch, TensorFlow plugins; C++ API |
Shipping Method
Currently, our products are shipped through DHL, FedEx, SF, and UPS.
Delivery Time
Once the goods are shipped, estimated delivery time depends on the shipping methods you chose:
FedEx International, 5-7 business days.
The following are some common countries’ logistic time.
Mon–Fri 8:00 AM–05:00 PM
Sat–Sun Closed