Skip to content

Commit

Permalink
Spec Updates
Browse files Browse the repository at this point in the history 
Fixed specifications to read "TeraFLOPS" instead of "TeraFLOPs", all instances of "BFP" replaced with "BLOCKFP"
  • Loading branch information
@dsklavos
dsklavos committed Jan 7, 2025
1 parent 77d437e commit 5ddb5c5
Show file tree
Hide file tree
Showing 2 changed files with 12 additions and 12 deletions.
16 changes: 8 additions & 8 deletions core/aibs/grayskull/specifications.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -21,9 +21,9 @@ The Grayskull™ e75 and e150 Tensix Processor add-in boards are built using the
| Memory | 8GB LPDDR4 | 8GB LPDDR4 |
| Memory Speed | 3.2 GT/sec | 3.7 GT/sec |
| Memory Bandwidth | 102 GB/sec | 118 GB/sec |
| TeraFLOPs (FP8) | 221 | 332 |
| TeraFLOPs (FP16) | 55 | 83 |
| TeraFLOPs (BFP8) | 55 | 83 |
| TeraFLOPS (FP8) | 221 | 332 |
| TeraFLOPS (FP16) | 55 | 83 |
| TeraFLOPS (BLOCKFP8) | 55 | 83 |
| TBP (Total Board Power) | 75W | 200W |
| External Power | 1x 6-pin PCIe *(required for powering fan)* | 1x 6+2-pin PCIe<br />1x 6-pin PCIe |
| System Interface | PCI Express 4.0 x16 | PCI Express 4.0 x16 |
Expand All @@ -39,11 +39,11 @@ The Grayskull™ e75 and e150 Tensix Processor add-in boards are built using the

The e75/e150 support the following data precision formats:

| Format | Bit Depth |
| -------------------- | ---------------- |
| Floating point | FP8, FP16, BF16 |
| Block floating point | BFP2, BFP4, BFP8 |
| Vector | VTF19 |
| Format | Bit Depth |
| -------------------- | ---------------------------- |
| Floating point | FP8, FP16, BF16 |
| Block floating point | BLOCKFP2, BLOCKFP4, BLOCKFP8 |
| Vector | VTF19 |

## Minimum System Requirements

Expand Down
8 changes: 4 additions & 4 deletions core/aibs/wormhole/specifications.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -22,9 +22,9 @@ The Wormhole™ n150s and n300s Tensix Processor add-in boards are built using t
| Memory | 12GB GDDR6 | 24GB GDDR6 |
| Memory Speed | 12 GT/sec | 12 GT/sec |
| Memory Bandwidth | 288 GB/sec | 576 GB/sec |
| TeraFLOPs (FP8) | 262 | 466 |
| TeraFLOPs (FP16) | 74 | 131 |
| TeraFLOPs (BFP8) | 148 | 262 |
| TeraFLOPS (FP8) | 262 | 466 |
| TeraFLOPS (FP16) | 74 | 131 |
| TeraFLOPS (BLOCKFP8) | 148 | 262 |
| TBP (Total Board Power) | 160W | 300W |
| External Power | 1x 4+4-pin EPS12V | 1x 4+4-pin EPS12V |
| Connectivity | 2x Warp 100 Bridge<br />2x QSFP-DD 400GbE (Active)* | 2x Warp 100 Bridge<br />2x QSFP-DD 400GbE (Active)* |
Expand Down Expand Up @@ -57,7 +57,7 @@ The n150/n300 support the following data precision formats:
| Format | Bit Depth |
| -------------------- | --------------------------------------- |
| Floating point | FP8, FP16, BF16<br />FP32 (Output Only) |
| Block floating point | BFP2, BFP4, BFP8 |
| Block floating point | BLOCKFP2, BLOCKFP4, BLOCKFP8 |
| Integer | INT8<br />INT32 (Output Only) |
| Unsigned Integer | UINT8 |
| TensorFloat | TF32 |
Expand Down

0 comments on commit 5ddb5c5

Please sign in to comment.