Intel Gaudi 3 128GB
Performance
Peak theoretical throughput by precision type
| Precision | Peak |
|---|---|
FP64 | No verified data available |
FP32 32-bit floating point | 229TFLOPS |
TF32 TensorFloat-32 | 459TFLOPS |
BF16 Brain Float 16 | 1,678TFLOPS |
FP16 16-bit floating point | 459TFLOPS |
FP8 8-bit floating point | 1,678TFLOPS |
FP6 | No verified data available |
FP4 | No verified data available |
INT8 | No verified data available |
Every figure here is dense. The vendor documents no structured (2:4) sparsity mode for this part, so there is no second number to quote.
Dense peak divided by accelerator TDP. Board power only: excludes host CPUs, networking, cooling and facility overhead. TDP for this part: 900 W.
The Gaudi 3 128GB in the GPU landscape
Peak FP16 TFLOPS (dense) against TDP, single-GPU parts tracked by Flopper
Higher and further left is better: more half-precision throughput for less power.
Specifications
Architecture
Gaudi 3
Form Factor
OAM
Launch Year
2024
Process Node
5nm
Memory
128 GB HBM2E
Bandwidth
3,700 GB/s
TDP
900 W
Max power (Flopper estimate)
~1.0 kW est. Flopper estimate: 900 W TDP x 1.15. The vendor publishes no maximum board power for this part.
Interconnect
1.2 TB/s RoCE v2 RDMA (24x 200GbE), 1200 GB/s bidirectional
direction and scope not stated by vendor
Spec Confidence
Official
Full Specifications
| Memory | |
|---|---|
| Memory | 128 GB |
| Memory Type | HBM2E |
| Bandwidth | 3.7 TB/s |
| Interface Width | No verified data available |
| Interconnect & I/O | |
| GPU-to-GPU | RoCE v2 RDMA (24x 200GbE), 1200 GB/s bidirectional |
| Interconnect Bandwidth | 1.2 TB/s direction and scope not stated by vendor |
| Cache | |
| L2 Cache | 96 MB |
| Power & Thermal | |
| TDP | 900 W |
| Max power (Flopper estimate) | ~1.0 kW est. |
| Enterprise Features | |
| Sparsity | No |
| General | |
| Form Factor | OAM |
| Architecture | Gaudi 3 |
| Process Node | 5nm |
| Launch Year | 2024 |
Datasheet & Resources
Flopper Datasheet
Intel Gaudi 3 128GB specifications, generated from our database. Printable.
Vendor product page
Intel documentation for the Gaudi 3 128GB
Intel Gaudi 3 AI Accelerator
Intel · 2024-04-09
Intel Data Center GPU Documentation
oneAPI guides, Xe architecture
Data Provenance
Every figure traced to a source
Primary Source
- Document
- Intel Gaudi 3 AI Accelerator
- Publisher
- Intel
- Published
- 2024-04-09
Data Quality
- Spec confidence
- Official
- Clock basis
- Boost
- Core precisions with figures
- 5 of 9
- Normalization
- All values in TFLOPS
Compare cloud providers offering on-demand GPU instances for AI training, inference, and HPC workloads.
Browse GPU Cloud ProvidersSimilar GPUs
Systems Using This GPU
Pre-configured systems featuring the Intel Gaudi 3
| System | GPU Count | Peak Performance | Total Power | |
|---|---|---|---|---|
Intel Gaudi 3 HLB-325 Baseboard Gaudi · baseboard · 2024 | 8x Gaudi 3 | No verified data available | 7.6 kW | View System |
Supermicro X14 Gaudi 3 AI Training Platform X14 · rack · 2024 | 8x Gaudi 3 | No verified data available | No verified data available | View System |
Ingrasys GB6181I GB · rack · | 8x Gaudi 3 | No verified data available | No verified data available | View System |
Frequently Asked Questions
How many TFLOPS does the Intel Gaudi 3 have?
The Intel Gaudi 3 delivers 229 TFLOPS FP32, 459 TFLOPS FP16 and 1,678 TFLOPS FP8 at peak.
What is the power consumption of the Intel Gaudi 3?
The Intel Gaudi 3 has a TDP (Thermal Design Power) rating of 900 watts.
How much memory does the Intel Gaudi 3 have?
The Intel Gaudi 3 is equipped with 128 GB of memory with 3,700 GB/s of memory bandwidth.
What architecture is the Intel Gaudi 3 based on?
The Intel Gaudi 3 is based on the Gaudi 3 architecture, launched in 2024.
How much does the Intel Gaudi 3 cost to rent?
Flopper is not tracking any live rental listings for the Intel Gaudi 3. The provider directory lists the clouds that rent comparable accelerators on demand.
When did the Intel Gaudi 3 come out?
The Intel Gaudi 3 launched in 2024 on the Gaudi 3 architecture.
Stay Updated on GPU Releases
Get notified when new GPUs are added or specifications are updated.
No spam, unsubscribe anytime.