Graphcore IPU (Intelligence Processing Unit), powered by AMAX’s Graphcore certified server solutions, have been built to support new breakthroughs in machine intelligence, and are optimized for industries such as Finance, Healthcare, Scientific Research, Telecoms, and Universities.
Graphcore IPU Advantages
Massively parallel MIMD. High performance/efficiency for future ML trends
Model & Data in tightly coupled large locally distributed SRAM
Delivering impressive performance for NLP, including BERT-Large training, cutting hours from AI development cycles.
Delivering significant performance advantages for AI applications, and running EfficientNet on the IPU requires extra INT8 quantization.
Graphcore IPU Products
- 1 Petaflop Compute system
- For Proof of Concept using IPU
- Designed for AI training & inference
- 4 petaFLOP AI system
- Dedicated, Powerful AI Compute
- Turnkey, ready for installation
- 16 petaFLOPS of AI training or inference
- Simple, Powerful Built in Networking
- For production deployment workloads
- 32 petaFLOPS of AI training or inference
- Superior scaling & blazing fast performance
- Full systems integration support for datacenter installation
Poplar® SDK Software
The Poplar SDK is a complete software stack, which was co-designed from scratch with the IPU, to implement our graph toolchain in an easy to use and flexible software development environment.
At a high level, Poplar is fully integrated with standard machine learning frameworks so developers can port existing models easily, and get up and running out-of-the-box with new applications in a familiar environment.
Below these frameworks sits Poplar. For developers who want full control to exploit maximum performance from the IPU, Poplar enables direct IPU programming in Python and C++.
AMAX Graphcore Approved Server
AMAX ServMax® A-114NV server have been approved by Graphcore for use with the IPU-POD switched systems.
The A-114NV server is a dual-socket 1U server and, when configured as described below, is approved and supported for use with both IPU-POD DA and IPU-POD switched systems. Note for larger rack systems (IPU‑POD128 and onwards) you will need one server per IPU‑POD64 rack.
System Configuration
| CPU | 2 | Dual AMD 7742 2.25 GHz, 64C/128T, 256 M, 225 W, 3200 |
|---|---|---|
| Memory | 16 | 32Gb RDIMM: DDR4, 3200 MHz, registered, ECC, 288-pin gold-plated DIMMs |
| Storage | 2 | 960 GB NVMe |
| 2 | 4 TB NVMe | |
| Networking | 2 | Mellanox ConnectX-5 EX Dual Port 40/100GbE QSFP28 Adapter |