GeekRadar / The comparison desk
COMPARE.
Tested systems
AMD

8 × AMD Instinct MI355X 288GB HBM3e

Tested system

Processor
AMD EPYC 9575F
Graphics
AMD Instinct MI355X 288GB HBM3e
Accelerator count
8
Nodes
1

Recorded results

Only the same benchmark, version, workload, scenario and unit share a row. Missing results are not zero. System results include the full submitted hardware and software configuration.

MLPerf Inference / dlrm-v3Offline · v6.1 · Samples/s

8 × AMD Instinct MI355X 288GB HBM3e

13,027.70* Samples/s

Accuracy target / result
DLRM_NE: 0.8681176645656181 DLRM_ACC: 0.6963383089448092 DLRM_AUC: 0.7862169525257562
Weight precision
fp16,bf16,fp8
Division
closed
System type
datacenter
Availability
available
Cores per processor
64
Processors per node
8
Nodes
1
Total accelerators
8
Software stack
PyTorch 2.10.0+git8514f05, ROCm 7.2.2
Operating system
Ubuntu 22.04.5 LTS
MLPerf Inference / dlrm-v3Server · v6.1 · Queries/s

8 × AMD Instinct MI355X 288GB HBM3e

12,198.80* Queries/s

Accuracy target / result
DLRM_NE: 0.8681164332983042 DLRM_ACC: 0.696331698458935 DLRM_AUC: 0.786212286822123
Weight precision
fp16,bf16,fp8
Division
closed
System type
datacenter
Availability
available
Cores per processor
64
Processors per node
8
Nodes
1
Total accelerators
8
Software stack
PyTorch 2.10.0+git8514f05, ROCm 7.2.2
Operating system
Ubuntu 22.04.5 LTS
MLPerf Inference / gpt-oss-120bOffline · v6.1 · Tokens/s

8 × AMD Instinct MI355X 288GB HBM3e

121,818* Tokens/s

Accuracy target / result
exact_match: 83.706
Weight precision
fp4
Division
closed
System type
datacenter
Availability
available
Cores per processor
64
Processors per node
8
Nodes
1
Total accelerators
8
Software stack
PyTorch 2.10.0+git8514f05, ROCm 7.2.2
Operating system
Ubuntu 22.04.5 LTS
MLPerf Inference / gpt-oss-120bServer · v6.1 · Tokens/s

8 × AMD Instinct MI355X 288GB HBM3e

113,241* Tokens/s

Accuracy target / result
exact_match: 83.451
Weight precision
fp4
Division
closed
System type
datacenter
Availability
available
Cores per processor
64
Processors per node
8
Nodes
1
Total accelerators
8
Software stack
PyTorch 2.10.0+git8514f05, ROCm 7.2.2
Operating system
Ubuntu 22.04.5 LTS
MLPerf Inference / llama2-70b-99.9Offline · v6.1 · Tokens/s

8 × AMD Instinct MI355X 288GB HBM3e

106,517* Tokens/s

Accuracy target / result
ROUGE1: 44.5481 ROUGE2: 22.1846 ROUGEL: 28.9482 TOKENS_PER_SAMPLE: 283.2
Weight precision
fp4
Division
closed
System type
datacenter
Availability
available
Cores per processor
64
Processors per node
8
Nodes
1
Total accelerators
8
Software stack
PyTorch 2.10.0+git8514f05, ROCm 7.2.2
Operating system
Ubuntu 22.04.5 LTS
MLPerf Inference / llama2-70b-99.9Server · v6.1 · Tokens/s

8 × AMD Instinct MI355X 288GB HBM3e

103,275* Tokens/s

Accuracy target / result
ROUGE1: 44.5266 ROUGE2: 22.1733 ROUGEL: 28.9289 TOKENS_PER_SAMPLE: 284.2
Weight precision
fp4
Division
closed
System type
datacenter
Availability
available
Cores per processor
64
Processors per node
8
Nodes
1
Total accelerators
8
Software stack
PyTorch 2.10.0+git8514f05, ROCm 7.2.2
Operating system
Ubuntu 22.04.5 LTS
MLPerf Inference / wan-2.2-t2v-a14bOffline · v6.1 · Samples/s

8 × AMD Instinct MI355X 288GB HBM3e

0.07* Samples/s

Accuracy target / result
vbench_score: 69.8899
Weight precision
bf16, fp8, fp4
Division
closed
System type
datacenter
Availability
available
Cores per processor
64
Processors per node
8
Nodes
1
Total accelerators
8
Software stack
PyTorch 2.10.0+git8514f05, ROCm 7.2.2
Operating system
Ubuntu 22.04.5 LTS
MLPerf Inference / wan-2.2-t2v-a14bSingleStream · v6.1 · Latency (s)

8 × AMD Instinct MI355X 288GB HBM3e

16.12* Latency (s)

Accuracy target / result
vbench_score: 69.9691
Weight precision
bf16, fp8, fp4
Division
closed
System type
datacenter
Availability
available
Cores per processor
64
Processors per node
8
Nodes
1
Total accelerators
8
Software stack
PyTorch 2.10.0+git8514f05, ROCm 7.2.2
Operating system
Ubuntu 22.04.5 LTS

Processing & platform

Processor
AMD EPYC 9575F
Graphics
AMD Instinct MI355X 288GB HBM3e

Connectivity & software

Operating system
Ubuntu 22.04.5 LTS

System configuration

Reported host core count
64
Processors per node
8
Accelerator count
8
Nodes
1
Software stack
PyTorch 2.10.0+git8514f05, ROCm 7.2.2
Submission division
closed

* Provisional data