Why is my AMD EPYC VPS 3-5x slower than my Intel Xeon VPS?

zGato

@melp57 care to comment?

for reference:

Noct

@zGato said:
@melp57 care to comment?

for reference:

I found the prospective customer service response over there was nothing like I find many other places. A lazy and uninformed provider as well as a shill butting in to each enquiry.

ascicode

@zGato said:
@melp57 care to comment?

for reference:

wrong thread?

ralf

@dev_vps said:

@lowendfolk said:

It's idling time!

if your use case is “idling” then c1v VPS would be perfect.

"idling" at least requires the machine to be up

TonyB

The problem with the frequency information is it's just what the CPU reports, not necessarily what the cores are running at. You'd hope the underlying hardware runs in performance mode, but that may not be the case. In performance mode, that also would mean it's likely assuming cooling is not an issue; the CPUs are running an all-core turbo. That also means benchmarks of the same CPUs in an idle environment might not be the same. Finally, not utilizing host passthrough can have an impact.

Here are a few examples. I tried to use underlying hardware that is as idle as possible but I don't have 100% idle options to test with. I cannot control the I/O portion though so there is variation there but I'll include it for the first one.

AMD EPYC 9354 (Base Clock: 3.25Ghz | All Core Turbo: 3.75Ghz)

Basic System Information:
---------------------------------
Uptime     : 0 days, 0 hours, 4 minutes
Processor  : AMD EPYC 9354 32-Core Processor
CPU cores  : 4 @ 3249.990 MHz

fio Disk Speed Tests (Mixed R/W 50/50) (Partition /dev/sda1):
---------------------------------
Block Size | 4k         (IOPS) | 64k        (IOPS)
  ------   | ---            ----  | ----        ----
Read    | 404.63 MB/s (101.1k) | 2.33 GB/s  (36.4k)
Write   | 405.70 MB/s (101.4k) | 2.34 GB/s  (36.6k)
Total   | 810.33 MB/s (202.5k) | 4.68 GB/s  (73.1k)
        |                   |                    
Block Size | 512k       (IOPS) | 1m         (IOPS)
  ------   | ---            ----  | ----        ----
Read    | 2.16 GB/s     (4.2k) | 1.85 GB/s  (1.8k)
Write   | 2.27 GB/s     (4.4k) | 1.98 GB/s  (1.9k)
Total   | 4.44 GB/s     (8.6k) | 3.83 GB/s  (3.7k)

Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value                      
                |                            
Single Core     | 2164                       
Multi Core      | 6880                       
Full Test       | https://browser.geekbench.com/v6/cpu/9949910

Same hardware and test without host passthrough using the same model you had ( AMD EPYC Processor (with IBPB))

Basic System Information:
---------------------------------
Uptime  : 0 days, 0 hours, 0 minutes
Processor  : AMD EPYC Processor (with IBPB)
CPU cores  : 4 @ 3249.990 MHz

fio Disk Speed Tests (Mixed R/W 50/50) (Partition /dev/sdb1):
---------------------------------
Block Size | 4k         (IOPS) | 64k        (IOPS)
  ------   | ---            ----  | ----        ----
Read    | 514.33 MB/s (128.5k) | 2.23 GB/s  (34.9k)
Write   | 515.69 MB/s (128.9k) | 2.25 GB/s  (35.1k)
Total   | 1.03 GB/s   (257.5k) | 4.49 GB/s  (70.1k)
        |                   |                    
Block Size | 512k       (IOPS) | 1m         (IOPS)
  ------   | ---            ----  | ----        ----
Read    | 2.11 GB/s     (4.1k) | 1.89 GB/s  (1.8k)
Write   | 2.22 GB/s     (4.3k) | 2.02 GB/s  (1.9k)
Total   | 4.33 GB/s     (8.4k) | 3.91 GB/s  (3.8k)

Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value                      
                |                            
Single Core     | 2043                       
Multi Core      | 6645                       
Full Test       | https://browser.geekbench.com/v6/cpu/9952247

I'm seeing a 3-5% decrease not using the host passthrough in this specific case. It's possible older CPU models might not see the same impact.

Other CPUs though for reference but I'll just leave the CPU benchmark portion for them:

AMD EPYC 9654P (Base Clock: 2.4Ghz | All Core Turbo: 3.55Ghz)

Basic System Information:
---------------------------------
Uptime  : 0 days, 0 hours, 3 minutes
Processor  : AMD EPYC 9654P 96-Core Processor
CPU cores  : 4 @ 2396.400 MHz

Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value                      
                |                            
Single Core     | 1980                       
Multi Core      | 6365                       
Full Test       | https://browser.geekbench.com/v6/cpu/9949921

AMD EPYC 7543(Base Clock: 2.8Ghz | All Core Turbo: 3.7Ghz)

Basic System Information:
---------------------------------
Uptime     : 0 days, 0 hours, 1 minutes
Processor  : AMD EPYC 7543 32-Core Processor
CPU cores  : 4 @ 2794.748 MHz

Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value                         
                |                               
Single Core     | 1613                          
Multi Core      | 5136                          
Full Test       | https://browser.geekbench.com/v6/cpu/9950007

AMD EPYC 7643 (Base Clock: 2.3Ghz | All Core Turbo: 3.3Ghz)

Basic System Information:
---------------------------------
Uptime  : 0 days, 0 hours, 2 minutes
Processor  : AMD EPYC 7643 48-Core Processor
CPU cores  : 4 @ 2299.998 MHz
Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value                      
                |                            
Single Core     | 1454                       
Multi Core      | 4573                       
Full Test       | https://browser.geekbench.com/v6/cpu/9949930

AMD EPYC 7402 (Base Clock: 2.8Ghz | All Core Turbo: 3.3Ghz)

Basic System Information:
---------------------------------
Uptime     : 0 days, 0 hours, 2 minutes
Processor  : AMD EPYC 7402 24-Core Processor
CPU cores  : 4 @ 2794.748 MHz

Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value                         
                |                               
Single Core     | 1176                          
Multi Core      | 3783                          
Full Test       | https://browser.geekbench.com/v6/cpu/9949944

I suspect it's an EPYC 7002 series CPU running in an energy-efficient configuration. The performance is correct when that is taken into account. They probably don't want the significant increase in power for a 15% increase in CPU core frequency.

XNQ

what's your steal on the server? io wait? I had similar issues and ended up letting the VPS expire as to me it was unusable and didn't wanna deal with it anymore.

arnoldz

@XNQ said:
what's your steal on the server? io wait? I had similar issues and ended up letting the VPS expire as to me it was unusable and didn't wanna deal with it anymore.

I paid for the server for a month, and on January 18th, I won’t be renewing it. For now, I’ll be sticking with the 2 CPU cores, 2GB server from Crunchbits.

greenhost_cloud

It might be worth looking into resource allocation, caching mechanisms, or even how your applications interact with the underlying hardware.

arnoldz

@greenhost_cloud said:
It might be worth looking into resource allocation, caching mechanisms, or even how your applications interact with the underlying hardware.

I tested all three servers by installing the same apps on each—Fastpanel, wget, Drupal, and a MySQL database. So, the servers weren't heavily loaded, they were running light.

itachikonoha

@arnoldz said:

@greenhost_cloud said:
It might be worth looking into resource allocation, caching mechanisms, or even how your applications interact with the underlying hardware.

I tested all three servers by installing the same apps on each—Fastpanel, wget, Drupal, and a MySQL database. So, the servers weren't heavily loaded, they were running light.

Continuing from the last thread since this one is more appropriate.

Have you tried without fastpanel?

Just plain old nginx/caddy+php+mysql+Drupal?

arnoldz

@itachikonoha said:

@arnoldz said:

@greenhost_cloud said:
It might be worth looking into resource allocation, caching mechanisms, or even how your applications interact with the underlying hardware.

I tested all three servers by installing the same apps on each—Fastpanel, wget, Drupal, and a MySQL database. So, the servers weren't heavily loaded, they were running light.

Continuing from the last thread since this one is more appropriate.

Have you tried without fastpanel?

Just plain old nginx/caddy+php+mysql+Drupal?

No, I haven't tried without Fastpanel. But on my VDS from @ShockHosting, Fastpanel is also installed, and the process of clearing an empty Drupal 11 cache takes only about 2 seconds—a very good result.