AMD Ryzen 9 5950x good for hosting?

ChaoscripT

Hi,
Yes I know that performance with this CPU is great, some hosting provider owner who provide hosting with AMD EPYC CPU told me that the 5950x is desktop grade and don't fit to work much hours like server grade CPUs.

What do you think?

Regards

jar

It's fine. Desktop and server grade hardware are barely distinguishable these days under the right conditions. My best production servers are Ryzen boxes.

Shakib

How about i9 11900K?

Faster than any Ryzen and EPYC, awesome single core performance.

chocolateshirt

@Shakib said:
How about i9 11900K?

Faster than any Ryzen and EPYC, awesome single core performance.

Is it faster than Ryzen 5950x ?

bruh21

@chocolateshirt said:

@Shakib said:
How about i9 11900K?

Faster than any Ryzen and EPYC, awesome single core performance.

Is it faster than Ryzen 5950x ?

single core, possibly, but 11900k has less cores

user54321

@Shakib said:
How about i9 11900K?

Faster than any Ryzen and EPYC, awesome single core performance.

between 0-9% faster singlecore but multicore 40-50% slower, great performance

As long as you don't host a single gameserver that scales only with one core, a 5950x will be the better choice, also it does draw less power even though it has 50% more cores.

bruh21

@user54321 said:

@Shakib said:
How about i9 11900K?

Faster than any Ryzen and EPYC, awesome single core performance.

between 0-9% faster singlecore but multicore 40-50% slower, great performance

As long as you don't host a single gameserver that scales only with one core, a 5950x will be the better choice, also it does draw less power even though it has 50% more cores.

intel draws so much power because of their 14nm process. the fact that they can even compete somewhat on this node is impressive, but the power consumption suffers

SirFoxy

@bruh21 said:

@user54321 said:

@Shakib said:
How about i9 11900K?

Faster than any Ryzen and EPYC, awesome single core performance.

between 0-9% faster singlecore but multicore 40-50% slower, great performance

As long as you don't host a single gameserver that scales only with one core, a 5950x will be the better choice, also it does draw less power even though it has 50% more cores.

intel draws so much power because of their 14nm process. the fact that they can even compete somewhat on this node is impressive, but the power consumption suffers

better for gameservers as many don't support multicore

Shakib

PHP only cares about single core performance.

Do you really need all of your 16 cores and 32 HT?

Ryzen providers has more cores and RAM, gets to sell more VPS from the server. It would benefit the provider the most, but wouldn't go well with @Daniel15 dedicated core/thread scenario.

Inefficient, more watts usage shouldn't be something a client should worry about i guess. It's something a provider would consider as more power usage will cost more.

You'd want more single core performance when hosting websites or buying VPS. Because you're not buying 32 vCore, you're only buying a few.

Something i have talked about before > https://www.lowendtalk.com/discussion/172906/ryzen-9-3900x-vs-core-i9-11900k

I hope someone here would explain this better than me.

Daniel15

@Shakib said: PHP only cares about single core performance.

Per request, sure, but a server under heavy load will have multiple requests running in parallel, up to the maximum configured for pm.max_children in your FPM config. Even within a single request, you can use something like the parallel extension to have multiple threads within one request.

In any case, sites that need to scale up that far generally don't use PHP

user54321

@SirFoxy said:
better for gameservers as many don't support multicore

to be correct two gameservers which only use a single thread and nothing else on the system, if you have more the turbo will be lower and the AMD CPU will be faster even singlecore.

cybertech

they are not wrong from provider point of view. but customer is always right. just look at crowncloud and terrahost 5950X. tons of 3900X providers who know what users want

ChaoscripT

@Daniel15 said:

@Shakib said: PHP only cares about single core performance.

Per request, sure, but a server under heavy load will have multiple requests running in parallel, up to the maximum configured for pm.max_children in your FPM config. Even within a single request, you can use something like the parallel extension to have multiple threads within one request.

In any case, sites that need to scale up that far generally don't use PHP

Most of my websites are with Wordpress CMS, uses PHP/MySQL, if there is any chance to use it with multi cores, and it's gives more performance it will be good.

Regards.

AXYZE

5950X is the best hosting CPU. Awesome performance for PHP, Node.js, Python(Django), nginx etc.

For PHP 11900K can be a little bit faster, but for nodejs there is bigger difference, but in favor of 5950X. Overall 5950X is best buy. One box that can do anything.

Shakib

@AXYZE said:
5950X is the best hosting CPU. Awesome performance for PHP, Node.js, Python(Django), nginx etc.

For PHP 11900K can be a little bit faster, but for nodejs there is bigger difference, but in favor of 5950X. Overall 5950X is best buy. One box that can do anything.

The problem is most people here ain't gonna buy the whole box. Most likely they will get a few vCores on a VPS.

Overall performance gonna decrease for all VPS once more VMs gets crammed into that one box. Which also affects I/O and bandwidth speeds.

Benchmarks don't lie (well, not usually).
I have seen quite a bit of single core performance difference between i9 11900K and Ryzen 3900X, 5950X when i compared with multiple benchmarks of other provider's Ryzen VPS.

I am not denying that Ryzen is faster on multi core performance.
But, more frequency and single core performance is what really matters in my machines and i care about. 🙃

coolice

@Shakib said:

@AXYZE said:
5950X is the best hosting CPU. Awesome performance for PHP, Node.js, Python(Django), nginx etc.

For PHP 11900K can be a little bit faster, but for nodejs there is bigger difference, but in favor of 5950X. Overall 5950X is best buy. One box that can do anything.

The problem is most people here ain't gonna buy the whole box. Most likely they will get a few vCores on a VPS.

Overall performance gonna decrease for all VPS once more VMs gets crammed into that one box. Which also affects I/O and bandwidth speeds.

Benchmarks don't lie (well, not usually).
I have seen quite a bit of single core performance difference between i9 11900K and Ryzen 3900X, 5950X when i compared with multiple benchmarks of other provider's Ryzen VPS.

I am not denying that Ryzen is faster on multi core performance.
But, more frequency and single core performance is what really matters in my machines and i care about. 🙃

Do not watch Ghz as AMD is executing more instructions per Ghz

Actually pure K is slower than 3950x
AMD Ryzen 9 5950X Single Thread Rating: 3496
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+9+5950X&id=3862

Intel Core i9-11900K @ 3.50GHz
Single Thread Rating: 3489
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i9-11900K+@+3.50GHz&id=3904

however there is Intel Core i9-11900KF @ 3.50GHz which have 3% better result
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i9-11900KF+@+3.50GHz&id=4240

BUT with 25 times lower samples number -

And In VPS scenario what you care is how much it allows to boost cores for example - when other 4 cores vps user is pushing 2 of its cores at 100% how much your 2 cores will boost - Intel gives vague answer to that ... it will allow more cores to boost if there is enough power and cooling and it not going over tdp

P.S Intel continue with its main trick for desktop cpus ECC Memory Supported No

For both Intel® Core™ i9-11900K and Intel® Core™ i9-11900KF

that is the main reason people make difference between Server and Desctop CPUs

All AMD Ryzens has ECC Support

Shakib

Those are just talks.

Most people here won't be getting 5950X on their VPS, they will still get 3900X or 3950X as those are available here on LET.

Again, i am talking about VPS performance, not the whole box. If you do VPS benchmark of both Ryzen and i9-11900K, you'll see how much behind the Ryzen is. Even the 5950X.

coolice

Well run this benchmark it is single core one PHP....

https://github.com/rusoft/php-simple-benchmark-script

Here is mine Ubuntu 20.04 server if I go 21.04 with 5.11 kernel it will be 5-10 percent better also it is KVM VM, result will be better for LXC

-------------------------------------------------------------------------------------------
|                                  PHP BENCHMARK SCRIPT                                   |
-------------------------------------------------------------------------------------------
Start               : 2021-09-12 08:59:45
Server              : Linux/5.4.0-81-generic x86_64
Platform            : Linux
System              : Ubuntu 20.04.3 LTS
CPU                 :
              model : AMD Ryzen 9 5950X 16-Core Processor
              cores : 16
          available : 16
                MHz : 3393.624MHz
Memory              : 256 Mb available
Benchmark version   : 1.0.39
PHP version         : 7.4.23
  available modules :
           mbstring : yes
               json : yes
               pcre : yes
Max execution time  : 600 sec
Crypt hash algo     : MD5
-------------------------------------------------------------------------------------------
TEST NAME                      :      SECONDS |       OP/SEC |      OP/SEC/MHz |    MEMORY
-------------------------------------------------------------------------------------------
01_math                        :    2.237 sec | 447.08 kOp/s | 131.74  Ops/MHz |      2 Mb
02_string_concat               :    0.218 sec |  35.39 MOp/s |  10.43 kOps/MHz | 128.84 Mb
03_1_string_number_concat      :    0.021 sec | 240.01 MOp/s |  70.73 kOps/MHz |      4 Mb
03_2_string_number_format      :    1.255 sec |   3.98 MOp/s |   1.17 kOps/MHz |      4 Mb
04_string_simple_functions     :    2.183 sec | 595.46 kOp/s | 175.46  Ops/MHz |      4 Mb
05_string_multibyte            :    0.939 sec | 138.44 kOp/s |  40.79  Ops/MHz |      4 Mb
06_string_manipulation         :    2.504 sec | 519.15 kOp/s | 152.98  Ops/MHz |      4 Mb
07_regex                       :    1.716 sec | 757.52 kOp/s | 223.22  Ops/MHz |      4 Mb
08_1_hashing                   :    2.226 sec | 584.12 kOp/s | 172.12  Ops/MHz |      4 Mb
08_2_crypt                     :    6.457 sec |   1.55 kOp/s |   0.46  Ops/MHz |      4 Mb
09_json_encode                 :    2.776 sec | 468.38 kOp/s | 138.02  Ops/MHz |      4 Mb
10_json_decode                 :    3.585 sec | 362.59 kOp/s | 106.84  Ops/MHz |      4 Mb
11_serialize                   :    2.359 sec | 551.04 kOp/s | 162.37  Ops/MHz |      4 Mb
12_unserialize                 :    2.744 sec | 473.78 kOp/s | 139.61  Ops/MHz |      4 Mb
13_array_fill                  :    1.658 sec |  30.16 MOp/s |   8.89 kOps/MHz |     12 Mb
14_array_range                 :    0.310 sec | 323.01 kOp/s |  95.18  Ops/MHz |     12 Mb
14_array_unset                 :    1.263 sec |  39.59 MOp/s |  11.67 kOps/MHz |     12 Mb
15_loops                       :    0.855 sec | 233.93 MOp/s |  68.93 kOps/MHz |      4 Mb
16_loop_ifelse                 :    0.326 sec | 153.49 MOp/s |  45.23 kOps/MHz |      4 Mb
17_loop_ternary                :    0.543 sec |  92.06 MOp/s |  27.13 kOps/MHz |      4 Mb
18_1_loop_defined_access       :    0.246 sec |  81.21 MOp/s |  23.93 kOps/MHz |      4 Mb
18_2_loop_undefined_access     :    2.765 sec |   7.23 MOp/s |   2.13 kOps/MHz |      4 Mb
19_type_functions              :    0.427 sec |   7.03 MOp/s |   2.07 kOps/MHz |      4 Mb
20_type_conversion             :    0.425 sec |   7.06 MOp/s |   2.08 kOps/MHz |      4 Mb
21_0_loop_exception_none       :    0.017 sec | 236.64 MOp/s |  69.73 kOps/MHz |      4 Mb
21_1_loop_exception_try        :    0.040 sec | 100.46 MOp/s |  29.60 kOps/MHz |      4 Mb
21_2_loop_exception_catch      :    1.601 sec |   2.50 MOp/s | 736.32  Ops/MHz |      4 Mb
22_loop_null_op                :    0.739 sec |  67.64 MOp/s |  19.93 kOps/MHz |      4 Mb
23_loop_spaceship_op           :    0.799 sec |  62.56 MOp/s |  18.44 kOps/MHz |      4 Mb
26_1_class_public_properties   :    0.095 sec |  52.61 MOp/s |  15.50 kOps/MHz |      4 Mb
26_2_class_getter_setter       :    0.370 sec |  13.52 MOp/s |   3.98 kOps/MHz |      4 Mb
26_3_class_magic_methods       :    0.499 sec |  10.02 MOp/s |   2.95 kOps/MHz |      4 Mb
-------------------------------------------------------------------------------------------
Total time:                    :   44.195 sec |  13.63 MOp/s |   4.02 kOps/MHz |
Current PHP memory usage:      :        4 Mb
Peak PHP memory usage:         :   125.25 Mb

Shakib

@coolice , post YABS & monster bench as well.

coolice

YABS use geekbench 5 and that is a desktop benchmark

Geekbench 5 measures your processor's single-core and multi-core power, for everything from checking your email to taking a picture to playing music, or all of it at once. Test your system's potential for gaming, image processing, or video editing with the Compute Benchmark. Test your GPU's power with support for the OpenCL, CUDA, and Metal APIs. New to Geekbench 5 is support for Vulkan, the next-generation cross-platform graphics and compute API.

I literally do not care for for any of the mentioned parameters in a VM

I care about PHP and MySQL 99.5% and 0.5 Node.js

So you post i9-11900k PHP Benchmark

P.S Monster Bench use Geekbench too so same non server benchmarks

tototo

HostCram LXC-1G plan
https://www.lowendtalk.com/discussion/173817/welovelet-i9-11900k-samsung-nvme-vps-for-1-first-month-renews-at-3-50-month-hostcram-llc/p1

root@lxc:~/php-simple-benchmark-script-master# apt install php
root@lxc:~/php-simple-benchmark-script-master# php ./bench.php
taskset: failed to set pid 4670's affinity: Invalid argument
-------------------------------------------------------------------------------------------
<<< WARNING >>>
CPU is in powersaving mode? Set CPU governor to 'performance'!
 Fire up CPU and recalculate MHz!
-------------------------------------------------------------------------------------------
Extenstion 'mbstring' not loaded or not compiled! Multi-byte string tests will produce empty result!
-------------------------------------------------------------------------------------------
|                                  PHP BENCHMARK SCRIPT                                   |
-------------------------------------------------------------------------------------------
Start               : 2021-09-13 05:51:20
Server              : Linux/5.4.128-1-pve x86_64
Platform            : Linux
System              : Debian GNU/Linux 9 (stretch)
CPU                 :
              model : 11th Gen Intel(R) Core(TM) i9-11900K @ 3.50GHz
              cores : 8
          available : 5
                MHz : 5098.905MHz
Memory              : 256 Mb available
Benchmark version   : 1.0.39
PHP version         : 7.0.33-0+deb9u11
  available modules :
           mbstring : no
               json : yes
               pcre : yes
Max execution time  : 600 sec
Crypt hash algo     : MD5
-------------------------------------------------------------------------------------------
TEST NAME                      :      SECONDS |       OP/SEC |      OP/SEC/MHz |    MEMORY
-------------------------------------------------------------------------------------------
01_math                        :    1.280 sec | 781.28 kOp/s | 153.22  Ops/MHz |      4 Mb
02_string_concat               :    0.210 sec |  36.66 MOp/s |   7.19 kOps/MHz | 128.84 Mb
03_1_string_number_concat      :    1.147 sec |   4.36 MOp/s | 854.87  Ops/MHz |      4 Mb
03_2_string_number_format      :    1.028 sec |   4.86 MOp/s | 953.89  Ops/MHz |      4 Mb
04_string_simple_functions     :    1.274 sec |   1.02 MOp/s | 200.13  Ops/MHz |      4 Mb
05_string_multibyte            :    -.--- sec |     -.--Op/s |     -.--Ops/MHz |         0
06_string_manipulation         :    2.067 sec | 629.06 kOp/s | 123.37  Ops/MHz |      4 Mb
07_regex                       :    1.564 sec | 831.14 kOp/s | 163.00  Ops/MHz |      4 Mb
08_1_hashing                   :    1.888 sec | 688.41 kOp/s | 135.01  Ops/MHz |      4 Mb
08_2_crypt                     :    5.656 sec |   1.77 kOp/s |   0.35  Ops/MHz |      4 Mb
09_json_encode                 :    3.255 sec | 399.43 kOp/s |  78.34  Ops/MHz |      4 Mb
10_json_decode                 :    2.428 sec | 535.39 kOp/s | 105.00  Ops/MHz |      4 Mb
11_serialize                   :    1.428 sec | 910.38 kOp/s | 178.54  Ops/MHz |      4 Mb
12_unserialize                 :    2.059 sec | 631.28 kOp/s | 123.81  Ops/MHz |      4 Mb
13_array_fill                  :    1.452 sec |  34.42 MOp/s |   6.75 kOps/MHz |     12 Mb
14_array_range                 :    0.172 sec | 580.47 kOp/s | 113.84  Ops/MHz |     12 Mb
14_array_unset                 :    1.206 sec |  41.45 MOp/s |   8.13 kOps/MHz |     12 Mb
15_loops                       :    0.723 sec | 276.67 MOp/s |  54.26 kOps/MHz |      4 Mb
16_loop_ifelse                 :    0.450 sec | 110.99 MOp/s |  21.77 kOps/MHz |      4 Mb
17_loop_ternary                :    0.875 sec |  57.17 MOp/s |  11.21 kOps/MHz |      4 Mb
18_1_loop_defined_access       :    0.326 sec |  61.42 MOp/s |  12.05 kOps/MHz |      4 Mb
18_2_loop_undefined_access     :    2.143 sec |   9.33 MOp/s |   1.83 kOps/MHz |      4 Mb
19_type_functions              :    0.777 sec |   3.86 MOp/s | 757.26  Ops/MHz |      4 Mb
20_type_conversion             :    0.520 sec |   5.76 MOp/s |   1.13 kOps/MHz |      4 Mb
21_0_loop_exception_none       :    0.029 sec | 139.92 MOp/s |  27.44 kOps/MHz |      4 Mb
21_1_loop_exception_try        :    0.034 sec | 117.58 MOp/s |  23.06 kOps/MHz |      4 Mb
21_2_loop_exception_catch      :    1.527 sec |   2.62 MOp/s | 513.74  Ops/MHz |      4 Mb
22_loop_null_op                :    0.804 sec |  62.19 MOp/s |  12.20 kOps/MHz |      4 Mb
23_loop_spaceship_op           :    0.690 sec |  72.42 MOp/s |  14.20 kOps/MHz |      4 Mb
26_1_class_public_properties   :    0.090 sec |  55.81 MOp/s |  10.94 kOps/MHz |      4 Mb
26_2_class_getter_setter       :    0.282 sec |  17.72 MOp/s |   3.48 kOps/MHz |      4 Mb
26_3_class_magic_methods       :    0.598 sec |   8.36 MOp/s |   1.64 kOps/MHz |      4 Mb
-------------------------------------------------------------------------------------------
Total time:                    :   37.983 sec |  15.85 MOp/s |   3.11 kOps/MHz |
Current PHP memory usage:      :        4 Mb
Peak PHP memory usage:         :   125.44 Mb

Shakib

@coolice

What do you think?

MythHost

I can provide information on performance with the 5950x in terms of a full dedicated server. Attached below is the benchmark:

Yet-Another-Bench-Script:

# ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## #
#              Yet-Another-Bench-Script              #
#                     v2021-06-05                    #
# https://github.com/masonr/yet-another-bench-script #
# ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## #

Mon 13 Sep 2021 04:27:46 AM UTC

Basic System Information:
---------------------------------
Processor  : AMD Ryzen 9 5950X 16-Core Processor
CPU cores  : 32 @ 2191.432 MHz
AES-NI     : ✔ Enabled
VM-x/AMD-V : ✔ Enabled
RAM        : 125.8 GiB
Swap       : 40.0 GiB
Disk       : 49.9 GiB

fio Disk Speed Tests (Mixed R/W 50/50):
---------------------------------
Block Size | 4k            (IOPS) | 64k           (IOPS)
  ------   | ---            ----  | ----           ----
Read       | 632.36 MB/s (158.0k) | 1.00 GB/s    (15.6k)
Write      | 634.03 MB/s (158.5k) | 1.00 GB/s    (15.7k)
Total      | 1.26 GB/s   (316.5k) | 2.01 GB/s    (31.4k)
           |                      |
Block Size | 512k          (IOPS) | 1m            (IOPS)
  ------   | ---            ----  | ----           ----
Read       | 1.14 GB/s     (2.2k) | 1.10 GB/s     (1.0k)
Write      | 1.20 GB/s     (2.3k) | 1.18 GB/s     (1.1k)
Total      | 2.35 GB/s     (4.5k) | 2.29 GB/s     (2.2k)

iperf3 Network Speed Tests (IPv4):
---------------------------------
Provider        | Location (Link)           | Send Speed      | Recv Speed
                |                           |                 |
Clouvider       | London, UK (10G)          | 867 Mbits/sec   | 202 Mbits/sec
Online.net      | Paris, FR (10G)           | 863 Mbits/sec   | 379 Mbits/sec
WorldStream     | The Netherlands (10G)     | 852 Mbits/sec   | 233 Mbits/sec
Biznet          | Jakarta, Indonesia (1G)   | busy            | busy
Clouvider       | NYC, NY, US (10G)         | 922 Mbits/sec   | 904 Mbits/sec
Velocity Online | Tallahassee, FL, US (10G) | 918 Mbits/sec   | 729 Mbits/sec
Clouvider       | Los Angeles, CA, US (10G) | 924 Mbits/sec   | 912 Mbits/sec
Iveloz Telecom  | Sao Paulo, BR (2G)        | 842 Mbits/sec   | 248 Mbits/sec

Geekbench 5 Benchmark Test:
---------------------------------
Test            | Value
                |
Single Core     | 1750
Multi Core      | 15401
Full Test       | https://browser.geekbench.com/v5/cpu/9803424

Shakib

There you go...

MythHost

@Shakib said: There you go...

Quick question about your test. Is your 11900K overclocked? Your test reports your CPU Running at 5.1 GHz which is over the boost clock of the stock 11900K. For an apples to apples comparison, you should run it stock and let the system boost it up rather than applying a manual overclock. Don't get me wrong though, 5.1 GHz on an 11900K is very impressive and leads to a sexy single core performance score, but it's not something most people will see unless they own their own hardware and/or their host supports overclocking.

Shakib

@MythHost said:

@Shakib said: There you go...

Quick question about your test. Is your 11900K overclocked? Your test reports your CPU Running at 5.1 GHz which is over the boost clock of the stock 11900K. For an apples to apples comparison, you should run it stock and let the system boost it up rather than applying a manual overclock. Don't get me wrong though, 5.1 GHz on an 11900K is very impressive and leads to a sexy single core performance score, but it's not something most people will see unless they own their own hardware and/or their host supports overclocking.

Nope. All of them boosting up automatically, can boost clock upto 5.3 GHz as it seems.

Benchmark by @henix on LXC-1G (on a different node)

MythHost

@Shakib said: Nope. All of them boosting up automatically, can boost clock upto 5.3 GHz as it seems.

You're 100% right, my apologies. I was looking at the wrong row of a spread sheet.

coolice

@Shakib said:
@coolice

What do you think?

My Ubuntu underperforms with that PHP some dev lefts his hands

Here another one from CL8 VM on total you are fraction of second faster which mostly 80% of the difference came from crypt and you use pve kernel it means LXC mine is KVM VM

coolice

I forgot to mention it is loaded node it goes up to only MHz : 4431.509MHz Turbo
If it was empty to use 4.9 on single core I know AMD will be the winner

Shakib

@coolice said:
I forgot to mention it is loaded node it goes up to only MHz : 4431.509MHz Turbo
If it was empty to use 4.9 on single core I know AMD will be the winner

Sweet dreams.

On KVM VPS

coolice

It is not without boost you just do not see it if you run

watch -n.1 "lscpu | grep MHz

on the node you will see that it is or n1 if you want it slower

I got to reproduce that with 7.4.3 version from where but will be later today