The following table shows a summary of throughput and latency for all variations.
GeMM Configuration |
perf (in MSPS) |
Latency(us) |
Matrices/s |
TOPs |
No.of Active Cores |
Vector Load |
No. of Active mem Banks |
Mem R/W Rate |
Active AIE Tiles |
Dynamic Power (mW) |
TOPs per Watt |
|---|---|---|---|---|---|---|---|---|---|---|---|
32x32x32 |
10541.176 |
0.097 |
10.29 x 10^6 |
1.34927 |
NA |
NA |
NA |
NA |
NA |
7469 |
0.180649 |
64x64x64 |
11027.692 |
0.371 |
2.69 x 10^6 |
1.41154 |
NA |
NA |
NA |
NA |
NA |
8659 |
0.163015 |
128x128x128 |
5589.083 |
2.931 |
3.41 x 10^5 |
1.43081 |
NA |
NA |
NA |
NA |
NA |
8665 |
0.165125 |
256x256x256 |
2799.316 |
23.411 |
4.27 x 10^4 |
1.43325 |
NA |
NA |
NA |
NA |
NA |
8682 |
0.165083 |
512x512x512 |
1399.957 |
187.25 |
5.34 x 10^3 |
1.43356 |
NA |
NA |
NA |
NA |
NA |
8538 |
0.167903 |
1024x1024x1024 |
699.997 |
1497.9 |
6.67 x 10^2 |
1.43359 |
NA |
NA |
NA |
NA |
NA |
8682 |
0.165123 |