Chip comparison
Intel Gaudi 3 vs Tenstorrent Blackhole
Side-by-side specs
Straight from each vendor's datasheet. Winner in bold: higher throughput, memory, or transistor count; lower power draw or price. Sparse (2:4) throughput roughly doubles the FP8/INT8 numbers shown.
| Field | Intel Gaudi 3 | Tenstorrent Blackhole |
|---|---|---|
| Process node | TSMC N5 | TSMC N6 |
| TDP | 900 W | 300 W |
| Memory | 128 GB HBM2e | 32 GB GDDR6 |
| Memory bandwidth | 3,675 GB/s | 512 GB/s |
| FP16 (dense) | 1,835 TFLOPS | 745 TFLOPS |
| BF16 (dense) | 1,835 TFLOPS | — |
| FP8 (dense) | 1,835 TFLOPS | — |
| INT8 (dense) | — | 2,980 TOPS |
| Form factor | OAM | PCIe |
| Announced | 2024-04-09 | 2024-07-01 |
| Released | 2024-09-01 | 2025-01-01 |
Robots running Intel Gaudi 3
No robots publicly running it yet.
Robots running Tenstorrent Blackhole
No robots publicly running it yet.
Data centers with Intel Gaudi 3
No data centers publicly running it yet.
Data centers with Tenstorrent Blackhole
No data centers publicly running it yet.
Common questions
Intel Gaudi 3 vs Tenstorrent Blackhole: which is faster for training?
Intel Gaudi 3 has 2.46x the dense FP16/BF16 throughput of the other (Intel Gaudi 3: 1,835 TFLOPS; Tenstorrent Blackhole: 745 TFLOPS). Real-world training speed also depends on memory bandwidth, interconnect, and software stack: see the spec diff for those inputs.
Intel Gaudi 3 vs Tenstorrent Blackhole: which has more memory?
Intel Gaudi 3 carries more HBM (Intel Gaudi 3: 128 GB HBM2e; Tenstorrent Blackhole: 32 GB GDDR6). Memory capacity sets the largest model that fits on one chip without partitioning; memory bandwidth sets inference throughput.
Intel Gaudi 3 vs Tenstorrent Blackhole: which is more power-efficient?
Dense FP16/BF16 TFLOPS per watt: Intel Gaudi 3 2.04 (1,835 ÷ 900 W); Tenstorrent Blackhole 2.48 (745 ÷ 300 W). Tenstorrent Blackhole wins at the die level; system-level efficiency also depends on cooling and interconnect.
See also: every chip comparison · Intel Gaudi 3 full page · Tenstorrent Blackhole full page.