Update benchmarking docs and add ResNet results - #276
Open
chloecrozier wants to merge 5 commits into
Open
Conversation
Contributor
|
| Filename | Overview |
|---|---|
| README.md | Updates headline benchmark and ResNet figures; the previous ambiguous socket core-count claim is no longer present. |
| docs/benchmarks/performance-dgx-spark.md | Reworks the benchmark methodology, results, scaling definitions, and reproduction commands; the formerly missing cross-host socket configurations are now referenced by their checked-in names. |
| docs/benchmarks/socket_benchmarking.md | Replaces lossy socket examples with cross-host loss-free TCP and paced UDP results. |
| docs/tutorials/daqiri-resnet-inference.md | Expands the inference tutorial to cover the five measured ResNet variants and their results. |
| examples/socket_bench.cpp | Adds target-rate pacing and benchmark-worker affinity support without leaving an eligible blocking issue from the prior threads. |
| examples/run_spark_bench.sh | Extends benchmark orchestration for the revised socket measurements. |
| examples/daqiri_bench_socket_udp_server_spark_xhost.yaml | Adds the server role configuration needed by the corrected cross-host UDP reproduction command. |
| examples/daqiri_bench_socket_udp_client_spark_xhost.yaml | Adds the matching paced cross-host UDP client configuration. |
| examples/daqiri_bench_socket_tcp_server_spark_xhost.yaml | Adds the server role configuration for cross-host TCP measurements. |
| examples/daqiri_bench_socket_tcp_client_spark_xhost.yaml | Adds the matching cross-host TCP client configuration. |
Reviews (8): Last reviewed commit: "#251 - Update cross-host TCP scaling res..." | Re-trigger Greptile
chloecrozier
force-pushed
the
ccrozier-251-benchmark-docs
branch
2 times, most recently
from
August 26, 2026 15:17
7ab1004 to
24cb5a7
Compare
Benchmark numbers disagreed across the README, the benchmarking pages, and the tutorials. Consolidate every published figure onto the cross-host 200 GbE pair, and rerun the socket cells so every published row is loss-free. - refresh raw Ethernet, RoCE, and socket figures; README mirrors the perf page - add five-model ResNet inference results - add socket core scaling, replacing a loopback table that showed 48% loss - record the PCIe Gen5 x4 host attach (126 Gb/s) as the ceiling behind every row Signed-off-by: Chloe Crozier <chloecrozier@gmail.com>
chloecrozier
force-pushed
the
ccrozier-251-benchmark-docs
branch
from
August 26, 2026 15:47
24cb5a7 to
0c79c2d
Compare
Signed-off-by: Denis Leshchev <dleshchev@nvidia.com>
Signed-off-by: Denis Leshchev <dleshchev@nvidia.com>
dleshchev
force-pushed
the
ccrozier-251-benchmark-docs
branch
from
September 1, 2026 23:03
18f02ed to
ffe79d6
Compare
Signed-off-by: Denis Leshchev <dleshchev@nvidia.com>
Signed-off-by: Denis Leshchev <dleshchev@nvidia.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Numbers disagreed across the README, benchmarking pages, and tutorials.
Consolidates them onto one measured cross-host testbed.
Updated:
Added (requested in review):
RTX PRO 6000 400 GbE benchmarks land in #162.
Test: mkdocs build --strict, check_doc_refs.py, check_html_links.py