Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!


Processor Model vs Core Count vs Core Speed
New on LowEndTalk? Please Register and read our Community Rules.

All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.

Processor Model vs Core Count vs Core Speed

yokowasisyokowasis Member
edited September 2020 in General

I know it's a decade old question, and the answer is depends, but I still can't get my head around it. Especially on Lowendmarket where every drop of penny count.

My usage is for web / db server, nothing fancy. Just Nginx + mysql for Wordpress and NodeJS.

Let's look at the case

  1. Intel Xeon E5645 12 cores, 24 threads @ 2.401GHz VS 2 x Dual Intel Xeon L5630 16 cores, 32 threads @ 2.13 GHz.

  2. Intel Xeon E3-1265L v3 4 Cores, 8 Threads VS 2 x Intel Atom C2750 16 Cores, 16 Threads @ 2.4GHz

  3. Intel i3-2130 2 cores, 4 threads @ 3.400GHz VS Intel Atom C2750 8 Cores, 8 Threads @ 2.4GHz

All of them have similar Price, but different cores count / speed and very different processor model.

Which of each case it will perform better and can handle more concurrent user / visitor ?

Does it always best to prioritize core count over everything else ? for a web and database server. Even tough it's just an atom. I mean 16 Cores @ 2.4 Ghz, isn't that bad. Also can the intel atom really beat the i3 in this regards ?

Comments

  • edited September 2020

    I will look at geekbench and pcmark score to see the CPU performance

  • E3

    Thanked by 1pedagang
  • @chocolateshirt said:
    I will look at geekbench and pcmark score to see the CPU performance

    So, is there any way to know the which one better except to buy the said dedicated server and test it ?

    Also, how do I calculate 2 dedicated server vs 1 dedicated server in term of bench (Case Number 2, 1 E3 vs 2 Intel Atom) ? do I just add the score of multi core score of the geekbench ?

  • Core Speed > Cores

    Unless its something like 1 vs 8 and your sites are heavy on traffic + database calls.

  • @yokowasis said:

    @chocolateshirt said:
    I will look at geekbench and pcmark score to see the CPU performance

    So, is there any way to know the which one better except to buy the said dedicated server and test it ?

    Also, how do I calculate 2 dedicated server vs 1 dedicated server in term of bench (Case Number 2, 1 E3 vs 2 Intel Atom) ? do I just add the score of multi core score of the geekbench ?

    You can search on:

    https://browser.geekbench.com/

    and

    https://www.cpubenchmark.net/cpu_list.php

    To decide which one have higher score. Sometime there are value for dual processor, if there are no data, you could multiply the value of single processor by 150% - 180%

  • HostMayoHostMayo Member, Host Rep

    I also find the comparison between E5 and E3 confusing specially due to large number of cores offered by L5630. Its an old processor but with large number of physical cores and the E3 option is Atom not a Xeon which another reason not to go for that. Even we roll out Atom and say E3 with Xeon it should be faster in general opinion but it doesnt makes sense for large number of cocurrent users which would be better supported with E5 due to actual larger physical cores.

  • @corbpie said:
    Core Speed > Cores

    Unless its something like 1 vs 8 and your sites are heavy on traffic + database calls.

    My sites are heavy on traffic + database calls. I could say 1k database call / sec on busy hours.

  • IonSwitch_StanIonSwitch_Stan Member, Host Rep
    edited September 2020

    As @chocolateshirt alluded to -- I find that for a VERY corse comparison of performance that passmark (single, and all cores) is directionally accurate. There are a large number of factors that can impact performance, but assuming all the components are similar -- a machine with a passmark of 2000 is going to be roughly 100% faster than a machine with a passmark of 1000.

    Your application also matters -- MySQL is single threaded for each query, each query (unless paralleled by your application) will be limited by the single threaded passmark score. Lots of users will use lots of threads generally however.

    So looking at the processors you listed --

    E5645 x 2
    All Cores Passmark 8822
    Single Core Passmark 1082

    L5630
    All Cores Passmark 6450
    Single Core passmark 964

    **E3-1265L **
    All Cores Passmark 6019
    Single Core Passmark 2005

    C2750
    All Cores Passmark 2401
    Single Core Passmark 382

    i3-2130
    All Cores Passmark 1968
    Single Core Passmark 1590

    A site hosted on a C2750 is going to be the slowest of all the systems listed, by 3-10x depending on what you compare to. The dual E5645 is super old, and not the fastest per core, but going to possibly have 25% more capacity than the E3-1265L. The E3-1265L will provide the fast single user experience (ie, page load time) under less than 100% load, and after 100% load the E5645 might be slightly faster.

    The E3-1265L will probably also use ~100 watts to serve the same traffic the E5645 will burn 200 watts to serve...

    Again, this is a very rough estimate of performance, only testing your applications and such will really say how the site might perform, and factors such as memory capacity, disk speed, etc might impact performance considerably. YYMV.

    Compare this all to what the cool kids are using for their VPS hosts --

    Ryzen 9 3900x
    All cores passmark: 32800
    Single Core Passmark: 2730

    You could basically run the same "cpu capacity" on a 4 vCPU VPS on a Ryzen 3900x as a dual processor E5645 dedicated server. If your running SSD (E5645) vs NVME VPS from someone like @Francisco or @seriesn ...its going to mop the floor on a 10 year old machine.

  • @IonSwitch_Stan said:
    As @chocolateshirt alluded to -- I find that for a VERY corse comparison of performance that passmark (single, and all cores) is directionally accurate. There are a large number of factors that can impact performance, but assuming all the components are similar -- a machine with a passmark of 2000 is going to be roughly 100% faster than a machine with a passmark of 1000.
    SNIP

    So, instead of all of my cpu listed above, I should just go with Hetzner AX41 ?

    AMD Ryzen 5 3600 has passmark 17851, it's wayyyyy higher than all of my listed CPU, with price almost the same. Although it merely has only 6 core ? Does this mean we already come to a point where clockspeed and cores count irrelevant ?

    Thanked by 1akhfa
  • pedagangpedagang Member
    edited September 2020

    rather than atom, better rent vps
    old amd still not bad

    just maybe

    (feeling doang Bos)

  • Yes, if overall CPU performance is the only relevant aspect, the AX41-NVMe is the best dedicated server, although you should also consider the NetCup RS servers (Epyc VPS with dedicated resources). Those two are the best options, by far.

    Thanked by 1pbx
  • pbxpbx Member
    edited September 2020

    Core count and performance per core are both important, but for web hosting you want good core performance, as it'll make your php (or nodejs) execution time lower.

    A recent CPU with good single core performance and several core will always be the best, but it can get expensive. Don't forget that IOPS are important as well as you seem to run DB intensive tasks: you probably want an SSD.

    Keep in mind that there will always be a bottleneck. It can be CPU power, but if can be disk I/O.

    Netcup's root servers can be a great option if a shared environment can work for your needs, as @AC_Fan mentioned. Storage is fast and RAID protected, and if you don't need a full Epyc dedi it can be enough for your workload at a fraction of the cost of a dedi, or at the same price that lower end stuff with better performance and no worries about hardware failures as the node is monitored. See here for a bench.

    Maybe you can work on lowering your hardware requirements (as we know that on LET every penny counts!): caching can help a lot to drastically reduce the need for CPU power. If you have enough RAM you also can use that to limit disk I/O you'll need: what matters isn't only the CPU as RAM as well as storage capacity and (even more importantly) the way you setup your server have to be taken into account... But to finish on your question, as you mention WP & NodeJS, I'd say that fast cores > many slow cores!

  • jsgjsg Member, Resident Benchmarker
    edited September 2020

    @yokowasis said:
    My usage is for web / db server, nothing fancy. Just Nginx + mysql for Wordpress and NodeJS.

    @yokowasis said:
    My sites are heavy on traffic + database calls. I could say 1k database call / sec on busy hours.

    Pardon me but the only reasonable answer directly relating to OP's question came from @IonSwitch_Stan. A small thumbs up also to those who recommended a good VDS like netcups, but OP asked about dedis.

    IonSwitch_Stan is right, passmark is a useful and reasonable first impression number, but one should dig deeper for such a decision. I did. Here are some general remarks and my remarks re each candidate:

    • All those all cores and single core benchmark numbers are nice and dandy but what I'm looking for (and recommend to others as well) is the ratio MC vs SC plus additionally the ratio number of cores vs MC/SC. Reason: If we get a multicore processor we of course want to get matching performance that is, ideally a CPU with x cores would also have x times the single core performance. In reality though that ratio usually is mediocre to poor. Example: the i3-2130 has 2 cores and 4 HTs, yet its multicore performance is just 1.23 times the single core performance; that's ridiculous.
    • Cache - Do not underestimate the importance of a processors cache. In that regard the C2750 is a very poor choice. Always make sure that your choice of processor has at least 2 MB of cache per core.
    • TDP ~ electrical power consumption, which is one of the major cost factors in hosting. E5645 (80W) vs L5630 (40W) will have a significant impact on your invoice. Keep in mind that 1 or 2 HU colo offers typically come with only 100W of el. power included. Having to buy an additional 200 W can double your colo cost. What I recommend to look for is "Watt per 1k performance". Poor example: the i3-2130 gobbles 33 W per 1k performance. The C2750 on the other hand needs just 8.3 W per 1k performance.
    • Desktop vs server CPUs - try to avoid desktop processors, especially those with graphics on chip. They tend to waste energy and to be optimized for workloads that one usually isn't interested in on a server.
    • VT-x and VT-d support. Having a dedi whose processors doesn't support virtualization more often than not turns out to be a bummer.
    • Other "goodies" - AVX is not highly desirable for the vast majority of server loads, SSE 4.2 is good enough and unlike AVX doesn't come with a premium price. AES and No eXecute are highly desirable.
    • Memory - obviously the more the better but don't go crazy over DDR3 vs DDR4 and the Mem clock speeds. THE important thing that makes a system fast or slow is processor cache, followed by amount of memory. Memory speed is just an additional plus but trust me, 98% of users running a DB and a web server can happily sail with DDR3 memory (and ideally lots of it).

    Now to the candidates:

    • E5645 - All in all a decent choice but it needs lots of power and doesn't have a good el. power vs. performance ratio and a rather mediocre MC vs. SC ratio.
    • L5630 - My favourite. Low power vs performance, largest cache of all candidates (3MB/core). The only minus is that it also has the slowest memory (but funnily the highest memory to CPU bus speed). Considering all other factors - esp. the large cache and the outstanding MC vs SC performance ratio - this in my view is clearly the winner.
    • E3-1265L - in my view the second best choice. Very decent SC performance, low power consumption, 2 MB cache per core, all in all a nice choice. Its minus is the on board GPU but on the plus side it supports AVX (in case you feel to need that).
    • C 2750 - Obviously decent wrt el. power consumption but its performance is just ridiculous (SC well below 400) plus, to make it worse it has by far the smallest cache (and no NX). My view: A clear no go.
    • i3-2130 - In my view tied in the race which is the worst candidate. Horrible el. power per performance ratio, poor MC vs SC ratio, GPU on board, just 2 core, no VT-d, no AES ... just miserable.

    So, my suggestion is to pick either the L5630 or the E3-1265L, similar MC performance, the E3 has much higher SC performance (in case you want/need that), the L5630 being better balanced in my view, but both are good choices. And both can easily handle 1k req/s.

    Some said/mentioned
    ... Ryzen ...

    Are you joking? That processor although very nice and I love it too, is far, far beyond OPs needs.

    Finally a short remark re netcup (or other) VDS: Nuh, at the end of the day a dedi is better. One main reason: no shared disks (which can draw down even the best system unless provider installs multiple RAID 10 NVMes (as in 6 per server) which costs a ton of money.
    Well noted, I'm a happy netcup VDS customer and I like my VDS there a lot but then it runs a well known, not too dynamic load and a relatively light site.

  • How much RAM?

    The CPU is one thing, the amount of RAM another. A Website / DB is not CPU intensive, considering this is rarely involving complex calculations/operations.

    Thanked by 1pbx
  • dodheimsgarddodheimsgard Member
    edited September 2020

    Nevermind :wink:

  • eva2000eva2000 Veteran
    edited September 2020

    @yokowasis said: All of them have similar Price, but different cores count / speed and very different processor model.

    Which of each case it will perform better and can handle more concurrent user / visitor ?

    Does it always best to prioritize core count over everything else ? for a web and database server. Even tough it's just an atom. I mean 16 Cores @ 2.4 Ghz, isn't that bad. Also can the intel atom really beat the i3 in this regards ?

    It's not just about cpu core count and clock speed. It's also about the relative cpu architecture and how old it is. Xeon E56xx is ancient compared to Xeon E3-12xx or E5-16xx etc. As to is Xeon E3-12xx v1 vs Xeon E3-12xx v6. A 9yr old Xeon E3-1230v1 4C/8T was about as fast a 10+ yr old dual Xeon E5620 Westmere cpus with 2x (4C/8T) = 8C/16T https://ark.intel.com/content/www/us/en/ark/products/47925/intel-xeon-processor-e5620-12m-cache-2-40-ghz-5-86-gt-s-intel-qpi.html

    Example Xeon E3-1230 v1 vs v3 vs v5 vs v6 https://ark.intel.com/content/www/us/en/ark/compare.html?productIds=97474,88182,75054,52271

    Even then it depends on your workloads. For instance, the jump between Xeon E3-1230 v1/v2 and v5+ above for cryptographic performance and HTTPS can be as high as 40-80% while other usage cases might only between 10-30% better. Noy many test and benchmark sites for HTTP/2 HTTPS and test with non-HTTPS or HTTP/1.1 HTTPS so don't see the improvements.

    Out of all the cpus listed, E3-1265Lv3 is relatively the fastest of the bunch. But depends on the price charged really. For the price, I see web hosts charge for them, you could get better and faster cpus for similar or slightly higher price.

    To know the relative performance of each generation of cpus, do some digging on 3 web sites for comparison benchmarks for an idea.

    @yokowasis said: Which of each case it will perform better and can handle more concurrent user / visitor ?

    Just having a better server doesn't equate to handling more concurrent users. That depends on how you configure your web stack from web server, PHP to MySQL database up to Kernel/TCP level configuration. But if you really want to scale user concurrency wise, then more cpu cores & memory, the better - provided you optimally configure your server to take advantage of such resources.

    And don't forget web performance is only one task a server does. Factor in data/database backup and restore/recovery speeds then your usage requirements and server selection might change. i.e. if you want to be able to backup and restore 500GB of data in 10-20 mins versus 60+ minutes would require different server resource requirements.

    Thanked by 2pbx TimboJones
  • @yokowasis said: My sites are heavy on traffic + database calls. I could say 1k database call / sec on busy hours.

    Depending on the type of MySQL queries, that is fairly light for today's modern cpus and stacks. Look into caching to offload and reduce MySQL database calls. 9+ years ago had a vBulletin forum client pushing 4,000 MySQL queries/second 24x7. Offloading some of the reads to Memcached cluster allowed ~ 3,200 MySQL queries/second to be offloaded to Memcached cluster - leaving just 800 MySQL queries/second hitting MySQL! And if you can cache on front end side, that would help too i.e. Cloudflare CDN cache, Nginx proxy caching, Nginx/PHP-FPM fastcgi_cache, Litespeed cache etc.

    Thanked by 2pbx coreflux
  • pbxpbx Member
    edited September 2020

    @jsg said: at the end of the day a dedi is better. One main reason: no shared disks

    OP didn't mention anything about disk. A good VDS can be better in this regard (iops, raw speed, etc.) than a dedi with a single HDD or a crappy SSD. Of course, if you we compare the VPS to a good dedi with an NVMe or DC grade SSD setup it's different.

    Here we don't know what kind of storage (nor do we know what amount of RAM!) @yokowasis would get with each CPU... And this stuff matters a lot...

    As @eva2000 said: caching rocks! Having enough RAM can help reduce the load on the CPU and storage layer, and that's why IMO a good VPS/VDS can be fine even for high traffic sites and the like, if the server is setup and optimized correctly.

  • imagine, if poll, i choose (1) Intel Xeon E5645
    for plenty web site

    if 1 or 2 web site, difinetly enough VPS or powerful dedi

    just imagine from common people

  • @pbx said:

    @jsg said: at the end of the day a dedi is better. One main reason: no shared disks

    OP didn't mention anything about disk. A good VDS can be better in this regard (iops, raw speed, etc.) than a dedi with a single HDD or a crappy SSD. Of course, if you we compare the VPS to a good dedi with an NVMe or DC grade SSD setup it's different.

    Here we don't know what kind of storage (nor do we know what amount of RAM!) @yokowasis would get with each CPU... And this stuff matters a lot...

    As @eva2000 said: caching rocks! Having enough RAM can help reduce the load on the CPU and storage layer, and that's why IMO a good VPS/VDS can be fine even for high traffic sites and the like, if the server is setup and optimized correctly.

    Assuming other component is all the same. But it's interesting point of view. Because the nature of mysql, I need to squeeze every last drop of iops. And having it not shared with other tenant I think is definitely help.

    Thank you for all of your insight guys. I will just grab a ax hetzner dedicated. I want to grab netcup RS server but then again, I am afraid shared disk speed, isn't adequate

    Thanked by 1pbx
  • jsgjsg Member, Resident Benchmarker

    @pbx said:
    OP didn't mention anything about disk.

    Wrong, he did. -> MySql.

  • pbxpbx Member
    edited September 2020

    @jsg said: MySql

    Dont be silly. He didn't mention the storage he would get, not the fact that he'd need to be able to have decent iops (which is possible with several good VPS/VDS providers nowadays): they don't use WD blacks in RAID10 anymore...

  • jsgjsg Member, Resident Benchmarker

    @pbx said:

    @jsg said: MySql

    Dont be silly. He didn't mention the storage he would get, not the fact that he'd need to be able to have decent iops (which is possible with several good VPS/VDS providers nowadays): they don't use WD blacks in RAID10 anymore...

    • DB always (also) means disk. And in fact OP confirmed that it's important to him.
    • I'm talking with providers and I find confirmed that disks still often is a problem on nodes. If you have 20 VMs on a node all having some DB working it is a problem and those funny IOPS numbers don't mean a whole lot nor does the NVMe speed which usually is rather theoretical and achieved only under optimal conditions.
    • Do you know whether a provider actually has 8 multi GB/s U2 drives and not just some much cheaper SSDs in an array, maybe along with a NVMe cache?
  • @yokowasis said:

    ax hetzner dedicated.

    smart choice Sir B) congrat

  • @jsg said: DB always (also) means disk. And in fact OP confirmed that it's important to him.

    Seems like you'll never change. Nobody said it wasn't important to him, and I just tried to say that he didn't specify what other hardware he'd get with each CPU.

    Anyway problem solved: @yokowasis made up his mind already :smile:

  • jsgjsg Member, Resident Benchmarker

    @pbx said:
    Seems like you'll never change. Nobody said it wasn't important to him, and I just tried to say that he didn't specify what other hardware he'd get with each CPU.

    Anyway problem solved: @yokowasis made up his mind already :smile:

    Funny. I'm right and confirmed by OP - yet you paint me as somehow stubborn ...

  • yokowasisyokowasis Member
    edited September 2020

    @jsg said:

    @pbx said:
    Seems like you'll never change. Nobody said it wasn't important to him, and I just tried to say that he didn't specify what other hardware he'd get with each CPU.

    Anyway problem solved: @yokowasis made up his mind already :smile:

    Funny. I'm right and confirmed by OP - yet you paint me as somehow stubborn ...

    Now now, no need to fight. I get what I came for.

    TL;DR
    Core count and core speed is irrelevant. Bench is.

    Thanks for both of you.

    Thanked by 1pbx
Sign In or Register to comment.