TensorFusion Remote/Local vGPU Benchmark Helm Chart

This Helm chart deploys the TensorFusion Remote/Local vGPU Benchmark application, which includes a deployment for running the benchmark tests and a cronjob for automated testing.

Benchmark Results

TorchBenchmark Results (2025-07-06)

To run the TorchBenchmark tests:

cd benchmark
python3 test.py -k "test_${model_name}_eval_cuda" -t ${eval_times}

Model	Native	NGPU Mode	Loss(NGPU)	Local	Loss(Local)	Same AZ	Loss(Same AZ)	Cross AZ	Loss(Cross AZ)
basic_gnn_edgecnn	41.15 s	40.95 s	-0.49%	43.48 s	5.66%	46.07 s	11.96%	54.97 s	33.58%
BERT_pytorch	249.02 s	248.84 s	-0.07%	251.26 s	0.90%	253.71 s	1.88%	261.62 s	5.06%
basic_gnn_gcn	15.05 s	15.24 s	1.26%	19.63 s	30.43%	29.70 s	97.34%	64.39 s	327.84%
basic_gnn_gin	9.47 s	9.53 s	0.63%	9.78 s	3.27%	12.66 s	33.69%	21.83 s	130.52%
hf_Albert	24.73 s	24.00 s	-2.95%	29.19 s	18.03%	39.19 s	58.47%	73.00 s	195.19%
hf_Bart	39.88 s	38.68 s	-3.01%	54.96 s	37.81%	94.17 s	136.13%	211.68 s	430.79%
hf_Bert	24.15 s	24.35 s	0.83%	29.55 s	22.36%	42.00 s	73.91%	75.86 s	214.12%
llama	39.91 s	41.20 s	3.23%	42.90 s	7.49%	45.80 s	14.76%	52.55 s	31.67%
hf_distil_whisper	170.61 s	170.87 s	0.15%	172.16 s	0.91%	178.75 s	4.77%	189.45 s	11.04%
hf_clip	191.60 s	191.70 s	0.05%	194.52 s	1.52%	197.51 s	3.08%	208.90 s	9.03%
hf_Whisper	58.98 s	59.18 s	0.34%	63.50 s	7.66%	66.66 s	13.02%	72.63 s	23.14%
Average Loss	-	-	0.00%	-	12.37%	-	40.82%	-	128.36%

MLPerf Results (2025-07-04)

To run the MLPerf benchmark:

mlcr run-mlperf,inference,_full,_r5.0-dev \
--model=bert-99 \
--implementation=reference \
--framework=pytorch \
--category=edge \
--scenario=SingleStream \
--execution_mode=valid \
--device=cuda \
--quiet --rerun

Mode	Time	Loss
Native	27.008 s	-
Local	29.930 s	10.82%
Same AZ	33.341 s	23.45%
Cross AZ	41.597 s	54.02%

Simulating AZ Latencies

To simulate different AZ (Availability Zone) network conditions, you can use the Linux Traffic Control (tc) tool to inject artificial network latency:

1. Inject network latency:

# For Same AZ simulation (0.3ms latency)
tc qdisc add dev lo root netem delay 0.3ms
# For Cross AZ simulation (1ms latency)
tc qdisc add dev lo root netem delay 1ms

2. Verify the latency:

ping target_host

3. Remove the artificial latency when done:

tc qdisc del dev lo root

Prerequisites

• Kubernetes 1.19+
• Helm 3.2.0+
• PV provisioner support in the underlying infrastructure
• A GPU node with NVIDIA drivers installed

Installing the Chart

To install the chart with the release name my-release:

helm install my-release ./helm/torchbench

The command deploys the benchmark application on the Kubernetes cluster with default configuration.

Configuration

The following table lists the configurable parameters of the chart and their default values.

Parameter	Description	Default
replicaCount	Number of replicas	1
image.repository	Image repository	crpi-wpzfqfci37r0ad3n.cn-hangzhou.personal.cr.aliyuncs.com/tensorfusionrobin/tensorfusionrobin
image.tag	Image tag	latest
image.pullPolicy	Image pull policy	Always
serviceAccount.create	Create service account	true
serviceAccount.name	Service account name	cronjob-sa
podAnnotations	Pod annotations	See values.yaml
podLabels	Pod labels	See values.yaml
resources	Pod resource requests and limits	See values.yaml
nodeSelector	Node selector	kubernetes.io/hostname: gpu-2
cronjob.schedule	Cronjob schedule	0 0 * * *
cronjob.concurrencyPolicy	Cronjob concurrency policy	Allow
cronjob.successfulJobsHistoryLimit	Number of successful jobs to keep	3
cronjob.failedJobsHistoryLimit	Number of failed jobs to keep	1

Usage

Running the Benchmark

The benchmark will run automatically according to the cronjob schedule. You can also manually trigger a benchmark run by:

1. Finding the cronjob:

kubectl get cronjob

2. Creating a job from the cronjob:

kubectl create job --from=cronjob/my-release-torchbench-test-runner manual-run

Viewing Results

To view the benchmark results:

kubectl logs -l app=my-release-torchbench

Customizing the Configuration

To customize the configuration, create a custom values file:

helm install my-release ./helm/torchbench -f custom-values.yaml

Uninstalling the Chart

To uninstall/delete the deployment:

helm uninstall my-release

Troubleshooting

If you encounter any issues:

1. Check the pod status:

kubectl get pods -l app=my-release-torchbench

2. Check the pod logs:

kubectl logs -l app=my-release-torchbench

3. Check the cronjob status:

kubectl get cronjob
kubectl get jobs

4. Check the service account:

kubectl get serviceaccount