Adding passmark benchmark results running on AMD Ryzen 5 7600X

hilliao · Jun 3, 2023 · 797fe1b · 797fe1b
1 parent 41481cf
commit 797fe1b
Showing 1 changed file with 102 additions and 59 deletions.
diff --git a/cpubench-passmark/README.md b/cpubench-passmark/README.md
@@ -23,10 +23,10 @@ You can adjust the following variables to simulate your deployed workloads.
 * MB_MEMORY
 
 ```commandline
-export NUM_CPUs=4
-export MB_MEMORY=512
-export CONTAINER_NAME=gcloud-${NUM_CPUs}cpus
-docker run --name $CONTAINER_NAME -d --cpus=$NUM_CPUs --memory=${MB_MEMORY}m gcr.io/google.com/cloudsdktool/cloud-sdk:latest sleep infinity
+export NUM_CPUs=4 && \
+export MB_MEMORY=512 && \
+export CONTAINER_NAME=gcloud-${NUM_CPUs}cpus && \
+docker run --name $CONTAINER_NAME -d --cpus=$NUM_CPUs --memory=${MB_MEMORY}m gcr.io/google.com/cloudsdktool/cloud-sdk:latest sleep infinity && \
 docker exec -it $CONTAINER_NAME bash
 ```
 
@@ -73,7 +73,7 @@ Executing the `docker run` command above, the passmark benchmark results running
 AMD Ryzen™ Threadripper™ PRO 3945WX workstation uses the command
 to start a docker container with vCPU limit set to 4, memory limit set to 512 MiB.
 
-### AMD threadripper Docker Passmark.yaml
+### AMD Ryzen Threadripper 3945wx Passmark results
 ```yaml
 Results:
   Results Complete: true
@@ -97,27 +97,30 @@ Results:
   m_CPU_enc_ECDSA: 4084348912.621902
 ```
 
-## Example result running on a container on Apple MacMini M2 with 4 CPUs
-Compared to the latest Apple silicon M2, the results are drastically different from running the benchmark on bare metal.
-Key performance metrics show large scale degradation for the following metric categories. The primary reason is that
-the docker container is running as emulated x86.
-`uname -a` returns `Linux 1e8c0cf3f9f8 5.15.49-linuxkit #1 SMP PREEMPT Tue Sep 13 07:51:32 UTC 2022 x86_64 GNU/Linux`
-
-### WINNER: AMD
-
-* CPU_INTEGER_MATH: 12134 is a quarter less than AMD's 15857
-* CPU_FLOATINGPOINT_MATH: 745 is merely 8% of what AMD's 9008
-* CPU_PRIME: 36 is the bright spot, making it 36% higher than AMD's 26.5
-* CPU_ENCRYPTION: 309 is shockingly 20 times worse compared to AMD's 5465
-* CPU_COMPRESSION: 26812 is 1/3 of AMD's 80428
-* CPU_SINGLETHREAD: 557 is 5 times worse than AMD's 2676
-* CPU_MATRIX_MULT_SSE: 35 is basically a joke compared to AMD's 4862. It's not even 1%. 
-* Encoding and Encryption benchmark scores combined are only half compared to AMD: (238986041+529282543+205998641)/(7635385602+5474155663+4084348912) => 0.05666357147629465
-
-Average metrics comparison: (12134/15857+745/9008+36/26.5+309/5465+26812/80428+557/2676+35/4862+238986041/7635385602+529282543/5474155663+205998641/4084348912)/10
-=> 0.299. Overall, the average of metrics shows Apple M2 is **70%** worse than AMD's 
+### AMD Ryzen 5 7600X Docker passmark results
+```yaml
+Results:
+  NumTestProcesses: 12
+  CPU_INTEGER_MATH: 21144.98333333333
+  CPU_FLOATINGPOINT_MATH: 14928.092017085815
+  CPU_PRIME: 41.362317335271563
+  CPU_SORTING: 11532.426624071779
+  CPU_ENCRYPTION: 7205.451012259553
+  CPU_COMPRESSION: 114850.06649693614
+  CPU_SINGLETHREAD: 4250.1011503066911
+  CPU_PHYSICS: 483.3072861346123
+  CPU_MATRIX_MULT_SSE: 7801.640998612419
+  CPU_mm: 695.40482911227195
+  CPU_sse: 3339.0007338346331
+  CPU_fma: 10158.386486041807
+  CPU_avx: 7854.5713382428139
+  CPU_avx512: 9854.6054363304229
+  m_CPU_enc_SHA: 10084425178.299089
+  m_CPU_enc_AES: 6816653751.0059433
+  m_CPU_enc_ECDSA: 5765310072.5881863
+```
 
-### Mac Mini M2 Docker Passmark.yaml
+### Mac Mini M2 Docker Passmark results
 ```yaml
 Results:
   Results Complete: true
@@ -141,6 +144,20 @@ Results:
   m_CPU_enc_ECDSA: 205998641.68711933
 ```
 
+## Example result running on a container on Apple MacMini M2 with 4 CPUs
+Compared to the latest Apple silicon M2, the results are drastically different from running the benchmark on x86 bare metal.
+Key performance metrics show large scale degradation for the following metric categories. The primary reason is that
+the docker container is running as emulated x86 on M2's ARM architecture.
+`uname -a` returns `Linux 1e8c0cf3f9f8 5.15.49-linuxkit #1 SMP PREEMPT Tue Sep 13 07:51:32 UTC 2022 x86_64 GNU/Linux`
+
+### WINNER: <font color="green">AMD</font>
+
+Average metrics comparison: (12134/15857+745/9008+36/26.5+309/5465+26812/80428+557/2676+35/4862+238986041/7635385602+529282543/5474155663+205998641/4084348912)/10
+=> 0.299. Overall, the average of metrics shows Apple M2 is <font color="red">**70%**</font> worse than AMD Threadripper
+
+Average metrics comparison: (12134/21144+745/14928+36/41.3+309/7205+26812/114850+557/4250+35/7801+238986041/10084425178+529282543/6816653751+205998641/5765310072)/10
+=> 0.20 Overall, the average of metrics shows Apple M2 is <font color="red">**80%**</font> worse than AMD Ryzen 5 7600X
+
 ## Benchmark results in Kubernetes between AMD and Apple M2 with 4 CPUs
 The Kubernetes cluster used on AMD Ryzen Threadripper is [Anthos clusters on Bare metal](https://cloud.google.com/anthos/clusters/docs/bare-metal/latest)
 which is a Google Cloud product. The Kubernetes cluster used on Apple M2 is [MicroK8s](https://microk8s.io/).
@@ -149,15 +166,56 @@ I am making an average on the ratio between the mectrics from the 2 processors t
 The Anthos cluster has 8 CPUs allocated to it in KVM. The Microk8s cluster had 6 CPUs allocated to it in
 [multipass](https://multipass.run/).
 
-### WINNER: AMD
+### AMD Ryzen Threadripper 3945wx Anthos K8s passmark results
+running on pod gcloud.yaml on a KVM instance with 6-8 vCPUs
+```yaml
+Results:
+  NumTestProcesses: 8
+  CPU_INTEGER_MATH: 21294.402833333334
+  CPU_FLOATINGPOINT_MATH: 15550.762873294345
+  CPU_PRIME: 66.96481227570203
+  CPU_SORTING: 12563.285423741843
+  CPU_ENCRYPTION: 5427.6943984052141
+  CPU_COMPRESSION: 85259.102123164077
+  CPU_SINGLETHREAD: 2437.3673388481893
+  CPU_PHYSICS: 1023.2969898574404
+  CPU_MATRIX_MULT_SSE: 7801.7738157035983
+  CPU_mm: 949.51738450057883
+  CPU_sse: 3717.052978099593
+  CPU_fma: 11386.89534272076
+  CPU_avx: 8301.3731262904439
+  CPU_avx512: 0
+  m_CPU_enc_SHA: 7554238459.3581209
+  m_CPU_enc_AES: 5073027587.4781075
+  m_CPU_enc_ECDSA: 4446784197.6702099
+```
 
-* Score of the top 9 metrics from CPU_INTEGER_MATH to CPU_MATRIX_MULT_SSE: (18913/21294+31787/15550+61/67+16528/12563+912/5427+33929/85259+3219/2437+1109/1023+6372/7801)/9
-=> 0.9940215056724604 shows the Apple M2 is about **.5%** worse than AMD
-* Score of the top 9 metrics plus the 3 m_CPU_enc_* metrics: (18913/21294+31787/15550+61/67+16528/12563+912/5427+33929/85259+3219/2437+1109/1023+6372/7801+1545555081/7554238459+409409970/5073027587+915243407/4446784197)/12
-=> 0.7864427225740651 shows significant degradation in performance. Apple M2 is **22%** worse than AMD.
+### AMD Ryzen 5 7600X Anthos K8s passmark results
+running on pod gcloud.yaml on a KVM instance with 6-8 vCPUs
+```yaml
+Results:
+  NumTestProcesses: 8
+  CPU_INTEGER_MATH: 29434.186000000002
+  CPU_FLOATINGPOINT_MATH: 21346.093324722933
+  CPU_PRIME: 87.20001821401361
+  CPU_SORTING: 16445.807581311186
+  CPU_ENCRYPTION: 6862.0079143045823
+  CPU_COMPRESSION: 96026.907982954246
+  CPU_SINGLETHREAD: 3583.5580584712411
+  CPU_PHYSICS: 925.18616154936183
+  CPU_MATRIX_MULT_SSE: 9437.3984514737804
+  CPU_mm: 930.17880674484582
+  CPU_sse: 4053.2585292291978
+  CPU_fma: 12885.9802895905
+  CPU_avx: 10101.16249210229
+  CPU_avx512: 10709.192494973136
+  m_CPU_enc_SHA: 9499170583.2338409
+  m_CPU_enc_AES: 6320709617.1549015
+  m_CPU_enc_ECDSA: 5766130231.8607836
+```
 
-### Mac Mini M2 microk8s passmark.yaml
-running on pod ubuntu.yaml as gcloud.yaml caused crash loop.
+### Mac Mini M2 microk8s passmark results
+running on pod ubuntu.yaml because gcloud.yaml caused crash loop.
 ```yaml
   CPU_INTEGER_MATH: 18913.772833333333
   CPU_FLOATINGPOINT_MATH: 31787.635753181163
@@ -178,34 +236,19 @@ running on pod ubuntu.yaml as gcloud.yaml caused crash loop.
   m_CPU_enc_ECDSA: 915243407.75800788
 ```
 
-### AMD Threadripper Anthos K8s passmark.yaml
-running on pod gcloud.yaml
-```yaml
-Results:
-  Results Complete: true
-  NumTestProcesses: 8
-  CPU_INTEGER_MATH: 21294.402833333334
-  CPU_FLOATINGPOINT_MATH: 15550.762873294345
-  CPU_PRIME: 66.96481227570203
-  CPU_SORTING: 12563.285423741843
-  CPU_ENCRYPTION: 5427.6943984052141
-  CPU_COMPRESSION: 85259.102123164077
-  CPU_SINGLETHREAD: 2437.3673388481893
-  CPU_PHYSICS: 1023.2969898574404
-  CPU_MATRIX_MULT_SSE: 7801.7738157035983
-  CPU_mm: 949.51738450057883
-  CPU_sse: 3717.052978099593
-  CPU_fma: 11386.89534272076
-  CPU_avx: 8301.3731262904439
-  CPU_avx512: 0
-  m_CPU_enc_SHA: 7554238459.3581209
-  m_CPU_enc_AES: 5073027587.4781075
-  m_CPU_enc_ECDSA: 4446784197.6702099
-```
+### WINNER: <font color="green">AMD</font>
+
+* Score of the top 9 metrics from CPU_INTEGER_MATH to CPU_MATRIX_MULT_SSE: (18913/21294+31787/15550+61/67+16528/12563+912/5427+33929/85259+3219/2437+1109/1023+6372/7801)/9
+=> 0.9940215056724604 shows the Apple M2 is about **.5%** worse than AMD
+* Score of the top 9 metrics plus the 3 m_CPU_enc_* metrics: (18913/21294+31787/15550+61/67+16528/12563+912/5427+33929/85259+3219/2437+1109/1023+6372/7801+1545555081/7554238459+409409970/5073027587+915243407/4446784197)/12
+=> 0.7864427225740651 shows significant degradation in performance. Apple M2 is <font color="red">**22%**</font> worse than AMD.
+* Score of the top 9 metrics plus the 3 m_CPU_enc_* metrics: (18913/29434+31787/21346+61/87+16528/16445+912/6862+33929/96026+3219/3583+1109/925+6372/9437+1545555081/9499170583+409409970/6320709617+915243407/5766130231)/12
+=> 0.6235723857629476 shows significant degradation in performance. Apple M2 is <font color="red">**38%**</font> worse than AMD.
 
-It's uncertain if the higher end Apple M2 processors would make a difference as the tests were executed in pods with
-CPU and memory resource limits. One thing for certain is that Apple M2 processors are definitely not designed for
-enterprise workloads. A 3 years newer processor should not be 22% worse than AMD's CPU from 2020.
+Apple M2 processors are definitely not designed for enterprise workloads.
+Comparing it with a 3 years older AMD Ryzen Threadripper processor, M2 at 4 cores is still underperformed.
+Comparing it with AMD Ryzen 5 7600X released 2 months later than the Apple M2 chip, the gap is even bigger at 
+<font color="red">**-38%**</font>
 
 ## Stress certain pods to simulate a busy Kubernetes cluster
 To simulate CPU heavy workloads in a container or a pod, install and execute the following packages