github/tfhe-rs
1
1
Fork 0
mirror of https://github.com/zama-ai/tfhe-rs.git synced 2026-01-07 22:04:10 -05:00
Code Issues Packages Projects Releases Wiki Activity

chore(gpu): update number of streams in erc20 throughput bench

Browse Source
This commit is contained in:
Agnes Leroy
2025-12-05 17:12:48 +01:00
committed by Agnès Leroy
parent 5273f61593
commit 100b4200c2
1 changed files with 4 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
tfhe-benchmark/benches/high_level_api/erc20.rs
View File

@@ -522,8 +522,9 @@ fn cuda_bench_transfer_throughput<FheType, F>(
let params = client_key.computation_parameters();
let params_name = params.name();
// 200 * num_gpus seems to be enough for maximum throughput on 8xH100 SXM5
let num_elems = 200 * num_gpus;
// 300 * num_gpus seems to be enough for maximum throughput on 8xH100 SXM5
// and is a multiple of the number of streams per GPU to avoid a bigger batch on one stream
let num_elems = 300 * num_gpus;
group.throughput(Throughput::Elements(num_elems));
let bench_id = format!(
@@ -540,7 +541,7 @@ fn cuda_bench_transfer_throughput<FheType, F>(
.map(|_| FheType::encrypt(rng.gen::<u64>(), client_key))
.collect::<Vec<_>>();
let num_streams_per_gpu = 8; // Hard coded stream value for FheUint64
let num_streams_per_gpu = 6; // Hard coded stream value for FheUint64
let chunk_size = (num_elems / num_gpus) as usize;
b.iter(|| {