Can't run Geekbench 6 on @TencentLighthouse 's promotional 1y 2vCPUs 2GB RAM VM

manish

I got a 1-year 2vCPUs Linux VM for 10.08 USD/year [promotional pricing] with the following specs:
2vCPUs2GB, SSD 40GB, Peak Bandwidth 20M, Monthly Traffic 512GB

As listed at: https://www.tencentcloud.com/act/pro/lighthouse

It freezes the system every time I run Geekbench 6 on it. I have no way to do anything but reboot from their console. It won't even let me ssh login from another tab etc.

System information according to the Geekbench 6:

System Information
  Operating System              Debian GNU/Linux 13 (trixie)
  Kernel                        Linux 6.12.94+deb13-amd64 x86_64
  Model                         Tencent Cloud CVM
  Motherboard                   N/A
  BIOS                          SeaBIOS seabios-1.9.1-qemu-project.org

CPU Information
  Name                          AMD EPYC 9754
  Topology                      1 Processor, 1 Core, 2 Threads
  Identifier                    AuthenticAMD Family 25 Model 160 Stepping 2
  Base Frequency                2.25 GHz
  L1 Instruction Cache          64.0 KB
  L1 Data Cache                 64.0 KB
  L2 Cache                      512 KB
  L3 Cache                      256 MB
  Instruction Sets              sse2 sse3 pclmul fma3 sse41 aesni avx avx2 avx512-f avx512-dq shani avx512-bw avx512-vl vaes avx512-vnni

Memory Information
  Size                          1.90 GB

So, the CPU looks good to handle this.

Their support is prompt and replied as follows:

Upon checking the logs, we found that the CPU was running at 100%​ and the system eventually ran out of memory (OOM). The system subsequently killed the geekbench_avx2​ business process.
Please note that severe memory shortages can cause excessive disk read operations, leading to I/O bottlenecks and system freezes.
We recommend:
1. Optimizing your application to reduce memory consumption, or
2. Upgrading to a higher-spec instance plan to accommodate your workload.  https://www.tencentcloud.com/document/product/1103/41562 

Of course a synthetic benchmark will peak the CPU to max for a while but it is the job of the hypervisor to throttle or control it? Isn't it? It should not just freeze, right? or am I wrong?

Can someone explain me why is this happening? I am running Debian 13.5 with no load like VPN, web servers etc. Just sshd server.

@TencentLighthouse Please look into this. Thank you!

Here is the results of @ddps 's 1core 2G RAM VM [1 Core less and same amount of RAM]:
https://browser.geekbench.com/v6/cpu/18569350

EthernetServers

What did their support team say when you contacted them on their website?

manish

@EthernetServers said: What did their support team say

I have included it in the post. Basically they want me to upgrade!

Upon checking the logs, we found that the CPU was running at 100%​ and the system eventually ran out of memory (OOM). The system subsequently killed the geekbench_avx2​ business process.
Please note that severe memory shortages can cause excessive disk read operations, leading to I/O bottlenecks and system freezes.

Could it be related to disk read operations? Or their hypervisor is not tuned to handle CPU burst loads?

Best Regards

dbadude

Dont run it in a ssh session, put in a background session. Like; nohup yourbenchscript &

manish

@dbadude said: Dont run it in a ssh session, put in a background session

I never had to do this. How would this help please?

totally_not_banned

@manish said:
Could it be related to disk read operations?

Maybe. If you run into a situation where the system doesn't have anything good to swap (assuming you have swap enabled of course) and starts endlessly juggling different chunks in and out of RAM (since it really needs all of them but can't fit everything into RAM at the same time) it's not going to be pretty and can basically feel like a hard lock even on bare metal. Add a virtualized disk shared with X other VMs and blasted cores...

Theoretically 2G should be plenty in my opinion but given how much senseless resource waste modern Linux installations regularly are i wouldn't be surprised if i'm wrong.

suut

Is it the latest version of Geekbench 6? If I recall correctly, there were issues with the previous version of Geekbench 6 failing to run.

dbadude

@manish said:

@dbadude said: Dont run it in a ssh session, put in a background session

I never had to do this. How would this help please?

if memory and cpu is getting tight some processes (like your ssh session) can be killed by the kernel or hypervisor. If you want your benchmark to finish it shouldnt be run in a ssh session but in the background, so when the ssh session is killed, the child (your benchmark) session isnt killed too.