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.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Real world comparison for more cores vs more clock speed
Let's say we have an server with dual hex core 1.9 GHz CPUs (24 threads) that runs 50 virtual machines with an average load of 2-4.
Would this server run better or worse or equal if it has single hex core 3.2 GHz CPU (12 threads) that has the same CPU benchmark score (cpubenchmark.net) as dual hex core CPUs?
Real world CPUs:
Dual Hex Core: http://cpubenchmark.net/cpu.php?cpu=Intel+Xeon+E5-2420+@+1.90GHz&id=1213&cpuCount=2
Single Hex Core: http://www.cpubenchmark.net/cpu.php?cpu=Intel+Xeon+E5-1650+@+3.20GHz&id=1211
Adding another argument:
If the load of the first server was 13, would it be still fine on the second 12 threaded CPU?
Comments
This is my personal opinion and from previous experience, for real world usage, 2 2GHz core is better than 1 4GHz core.
@seriesn: one 4GHz core is much better than 2 2GHz cores. End of story.
Sure, wait till the load spikes up
Wrong! @Bogdacutuu @seriesn!
a 3.2 ghz P4 is not faster than at xeon e5 2420 core. you can't go just by ghz any more.
That reminds me of host who was selling VPS last year from a p4. SonWebHost was his name.
it really all depends on the processor design!
@24khost I meant identical cores (the only difference between them being clock speed).
Larger room vs smaller room with higher ceiling?
It's difficult to make an irrefutable statement in this matter, but I would agree that more cores will handle a load better. Plenty of applications that are single threaded and benefit from a higher core speed, but hardly worth building a system to accomodate developers who won't do what they need to do, unless that is the purpose of the system.
so right now he has dual hex cores vs a single hex core. That is half the amount of threads. at that point the ghz doesn't really matter that much. I would take the more threads setup.
@jarland Having less cores with high clock speed is better in my opinion, since it can handle both single and multithreaded apps without issues.
@jarland thank you for explaining it better.
Single threaded apps benefit from CPU Speed
Multi threaded apps benefit from more cores if the app can take advantage of all of them.. Then speed becomes a factor. IE app can only do 8 threads, then you're capped until you up the GHz.
@Bogdacutuu
lets put 50 vps containers on those 2 systems. Your going to have higher loads due to containers having to wait for an available thread. 2x as long. I would take the more threads.
Think this platform as a VMs of almost any kind of usage. Like they are running single threaded game servers and multi threaded database servers etc. This eliminates the behavior for single threaded and multi threaded applications take advantage of different kind of CPUs.
Would the 2 system do the same? Or one would be better?
Dual cpu would be better, even at lower clock speeds. more threads equals less cpu wait
@24khost, check the load. It's 3-4. So in 12 threaded CPU, there will be still free threads.
I believe if the load was over 12, 24 threaded slow core CPU would be better.
@serverian Server load also depends on clock speed
@Bogdacutuu, yeah sorry. I meant having a load amount for the second server that equals to 13 in first server. Not exactly 13
You do as you wish but I will take the more threads everyday.
Typically I follow this formula:
TotalThreads/4 + TotalCores/2 * Ghz/core
So a few examples:
Dual L5420 = no threading + 8/2 * 2.5 = 10 Performance rating
E3-1240v2 = 8/4 + 4/2 * 3.4 = 13.6 Performance rating
E5-2640 = 12/4 + 6/2 * 2.5 = 15 Performance rating
The higher the number the better typically. This is just math I've done over time, it may not hold true for all processors or all systems.
Oh yeah!
Man we really need a Wiki page to keep track of these glories from the past.
DepotVPS, SonWebHost...
No
IMHO, to get a Load of 13 in the first one, probably you need to push more work than in the 12 threaded one....
But maybe someone can try to run some stuff or something
For a lot of users I would say go with the added cores since it will spread the load out across them. However some apps can only use one core so in that case a higher clock speed might be better.
There may be some exceptions but I'd say it would run worse.
This.
For virtualization the number of threads is extremely important, especially on OVZ and Xen-PV.
The less switching is done, the better the performance and stability.
It also happens to have abusers that lock a few cores, if that happens, the server will have many others for the rest of people.
So, definitely more cores is better.
You fail to consider instructions per clock? According to benchmarks, the e3-1240v2 is 2x as good as l5420
So a few examples:
Dual L5420 = no threading + 8/2 * 2.5 = 10 Performance rating
E3-1240v2 = 8/4 + 4/2 * 3.4 = 13.6 Performance rating
E5-2640 = 12/4 + 6/2 * 2.5 = 15 Performance rating
Crude, but I like it as a rough guesstimate.
Personally, I have a benchmark program which crunches numbers as fast as it can in a minute, runs two tests- single thread, and all threads simultaneously. The idea is to report the raw CPU power per-core, and the power of all cores running simultaneously. Interesting that it shows my Dual L5420 running 8 simultaneous CPU math does outperform a E3-1270 running the same 8 simultaneous CPU math, but on the one-thread test, a E3 core does slightly outperform a L5420 core. That is correct- a E3-1270 core at 3.4GHz was only slightly faster for me than a single L5420 core at 2.5GHz, maybe because of the beefy cache on the L5420, but demonstrates how you cannot just compare GHz vs GHz to figure out real performance. This also seems to indicate that if I was doing a lot of simultaneous processing, more real cores should be better than more hyperthreaded cores. Never used Unixbench so I don't know if that reports differently.
So, there are a lot of factors in real-world comparisons.
When in doubt, and if prices are equal, I always go with the faster per-core choice. Most of the time, I would rather drive on a 4-lane 120mph highway than a 8-lane 60mph highway, because the 8 lanes would only be a better choice if traffic is always heavy, and if traffic on that road really is always heavy, real lanes would be slightly better than virtual lanes.
There is no simple answer to the question. The answer is it depends. The application drives the requirements. For a VPS type of use I'd say that more cores will usually be the best solution. However, going really big like a 4 chip E5-4650 box will cost so much that it's not going to have the best preformance per $. I recently went through this process and settled on 2 chip E5-2670's boxes. It wasn't for VPS usage though.
More cores != more virtual machines.
More memory != more virtual machines.
There is only so much cpu cache. Hence as you ramp up the running 'processes', memory access becomes far more important as your cache misses increase. You cant work on something you dont have.
More memory != more virtual machines.
True, but it does help.
As with the cache, more cores usually mean more cache, but, obviously, in big monster server with 10 k threads, it will probably miss almost continuously.
A smaller server with less machines gives more performance per machine and is more stable than a big one, but there are other things to consider here too, such as IOPS, even with large arrays, the IOPS dont scale at the same pace as the number of VMs, there is also the network port limitation...
Small VMs, small server, big VMs, big server.
I prefer more core over clock speed.