For more performance results, visit our Performance Results page
Explore
IPU-POD
Ideal for exploration, the IPU-POD16 gives you all the power, performance and flexibility you need to fast track your IPU prototypes and speed from pilot to production. IPU-POD16 is your easy-to-use starting point for building better, more innovative AI solutions with IPUs whether you're focused on language and vision, exploring GNNs and LSTMs or creating something entirely new.
- Fast & efficient for dense matmul models
- Excels at sparse & fine-grained computation
- Expert support to get you up and running quickly
IPU-POD16 opens up a new world of machine intelligence innovation.
Built from 4 inter-connected IPU-M2000s and a pre-qualified host server from your choice of leading technology brands, IPU-POD16 is available to purchase today in the cloud or for your datacenter from our global network of channel partners and systems integrators.
We saw a roughly 5X speed gain. That means a researcher can now run potentially five times more experiments, which means we accelerate the whole research and development process and ultimately end up with better models in our products.
Razvan Ranca
Tractable
Performance
World-class results whether you want to explore innovative models and new possibilities, faster time to train, higher throughput or performance per TCO dollar.
ECOSYSTEM
Software tools and integrations to support every step of the AI lifecycle from development to deployment to improve productivity and AI infrastructure efficiency. And just make it easier to use.
| IPUs | 16x GC200 IPUs |
| IPU-M2000s | 4x IPU-M2000s |
| Memory | 14.4GB In-Processor-Memory™ and up to 1024GB Streaming Memory |
| Performance | 4 petaFLOPS FP16.16 1 petaFLOPS FP32 |
| IPU Cores | 23,552 |
| Threads | 141,312 |
| IPU-Fabric | 2.8Tbps |
| Host-Link | 100 GE RoCEv2 |
| Software |
Poplar TensorFlow, PyTorch, PyTorch Lightning, Keras, Paddle Paddle, Hugging Face, ONNX, HALO OpenBMC, Redfish DTMF, IPMI over LAN, Prometheus, and Grafana Slurm, Kubernetes OpenStack, VMware ESG |
| System Weight | 66kg + Host server |
| System Dimensions | 4U + Host servers and switches |
| Host Server | Selection of approved host servers from Graphcore partners |
| Thermal | Air-Cooled |
| Optional Switched Version | Contact Graphcore Sales |
Build
IPU-POD
Ramp up your AI projects, speed up production and see faster time to business value. IPU-POD64 is the powerful, flexible building block for world-leading AI performance in your enterprise datacenter, private or public cloud. Whether you're running large language models or rely on fast and accurate vision models, IPU-POD64 will deliver the results you need today as well as giving you the opportunity to explore innovative AI solutions for tomorrow.
View Product
Grow
IPU-POD
When you're ready to scale, choose IPU-POD128 for production deployment in your enterprise datacenter, private or public cloud. Experience massive efficiency and productivity gains when large language training runs are completed in hours or minutes instead of months and weeks. IPU-POD128 delivers for AI at scale.
View Product
Grow
IPU-POD
When you're ready to explore AI compute at supercomputing scale, choose IPU-POD256 for production deployment in your enterprise datacenter, private or public cloud. Experience massive efficiency and productivity gains when large language training runs are completed in hours or minutes instead of months and weeks. IPU-POD256 delivers AI at scale.
View Product