No more 3.4 GHZ CPU cores?

2

Comments

  • edited April 2015
    I first discovered 2.1Ghz (Vultr 1) a long time ago with my first NJ storage instance. I assumed 2.1 were used because storage doesn't need much CPU.

    My latest NJ storage instance is 3.4Ghz (Vultr 2). I've yet to experience 2.4 (Vultr 3?) or its capabilities...

    19::32 (11) "/tmp" jamie@catflap% cat plop
    CPU: Vultr Virtual CPU 1 (2133.46-MHz K8-class CPU)
    Vendor ID: "GenuineIntel"; CPUID level 4

    Intel-specific functions:
    Version 000106a3:
    Type 0 - Original OEM
    Family 6 - Pentium Pro
    Model 26 -
    Stepping 3
    Reserved 0

    Extended brand string: "Vultr Virtual CPU 1"

    dmesg: <FPU,DE,PSE,TSC,MSR,PAE,MCE,CX8,APIC,SEP,MTRR,PGE,MCA,CMOV,PAT,PSE36,CLFLUSH,MMX,FXSR,SSE,SSE2,HTT,SSE3,SSSE3,CX16,SSE4.1,SSE4.2,x2APIC,POPCNT,HV,SYSCALL,NX,LM,LAHF>
    dmesg: TSC: P-state invariant
    cpuid: FPU DE PSE TSC MSR PAE MCE CX8 APIC SEP MTRR PGE MCA CMOV FGPAT PSE-36 CLFSH MMX FXSR SSE SSE2 HT SSE3 SSSE3 CX16 SSE4.1 SSE4.2 x2APIC POPCNT SYSCALL XD-bit EM64T LAHF
    clang: -mtune=corei7 -march=corei7 -mmmx -msse -msse2 -msse3 -mssse3 -mcx16 -msse4.1 -msse4.2 -mpopcnt

    --------
    CPU: Vultr Virtual CPU 2 (3392.30-MHz K8-class CPU)
    Vendor ID: "GenuineIntel"; CPUID level 13

    Intel-specific functions:
    Version 000306c1:
    Type 0 - Original OEM
    Family 6 - Pentium Pro
    Model 60 -
    Stepping 1
    Reserved 0

    Extended brand string: "Vultr Virtual CPU 2"

    dmesg: <FPU,DE,PSE,TSC,MSR,PAE,MCE,CX8,APIC,SEP,MTRR,PGE,MCA,CMOV,PAT,PSE36,CLFLUSH,MMX,FXSR,SSE,SSE2,HTT,SSE3,PCLMULQDQ,SSSE3,FMA,CX16,PCID,SSE4.1,SSE4.2,x2APIC,MOVBE,POPCNT,TSCDLT,AESNI,XSAVE,OSXSAVE,AVX,HV,SYSCALL,NX,RDTSCP,LM,LAHF,BMI1,HLE,AVX2,BMI2,ENHMOVSB,INVPCID,RTM>
    dmesg: TSC: P-state invariant
    cpuid: FPU DE PSE TSC MSR PAE MCE CX8 APIC SEP MTRR PGE MCA CMOV FGPAT PSE-36 CLFSH MMX FXSR SSE SSE2 HT SSE3 PCLMULDQ SSSE3 FMA CX16 PCID SSE4.1 SSE4.2 x2APIC MOVBE POPCNT TSC-Deadline AESNI XSAVE OSXSAVE AVX "Feature 31" SYSCALL XD-bit RDTSCP EM64T LAHF FSGSBASE BMI1 HLE AVX2 SMEP BMI2 "Enhanced REP MOVSB/STOSB" INVPCID RTM
    clang: -mtune=core-avx2 -march=core-avx2 -mmmx -msse -msse2 -msse3 -mpclmul -mssse3 -mfma -mcx16 -msse4.1 -msse4.2 -mpopcnt -maes -mavx -mbmi -mavx2 -mbmi2 -mrtm
  • Ah... the new CPU's are basically Vultr-2, but at 2.4Ghz - that makes things easier at least!
  • Thanks for the warning Ben. Time to shop.
  • edited May 2015
    Ok assuming I did this correctly.

    I ran unixbenchmark on my running mail server. Doveccot postfix mysql claimd amavis etc. Server was 768m. Score came out to 2179.8.

    Spun up new server same size same location score was 1422.8.

    Higher is better, correct? If so looking at the new benchmarks for mid-range server I am getting about the same performance as a 2 gig server with my 768m

    Am I understanding this correctly?


  • I just had a look at my Australian servers, both were 3.4GHz, I've been using them for about 7 months now. At least they haven't been changed (Although I'm afraid to reboot them). For me a change in CPU clock speed might mean having to change providers. I use Vultr VPSs for GarrysMod servers. The server software is only single threaded and Vultr has been the best place to run them for a while.

    Interestingly though, has anyone else noticed a disk speed drop as well? I'm only getting 250MB/s now. I think Vultr needs to update their benchmark page.
  • I still have, is FR, 3600.008mhz, (3,6Ghz), assuming its virtual core, dunno what to think :x
  • I just rebooted for kernel update. Core stayed at 3.6. Looking at my munin graphs it doesn't appear that any thing has changed with disk.
  • I just spun up 2x Vultr 2GB instances LA 3.60ghz and NY 2.40Ghz. Did a 4-way benchmark comparison to see how much of a difference there is between the 2 at https://community.centminmod.com/threads/digitalocean-vs-vultr-vs-linode-benchmark-compare.2889/.

    Enjoy :)
  • @eva200 - Thank you for that, it's really informative and was well worth the read.
  • edited May 2015
    @guyf2010 @eva2000 It seems current instances aren't being changed, at least for now, all new ones are. (Unless there is space on an old node, due to someone destroying a previously created instance)
  • @RagnaHosting your welcome

    Yeah it's going to be a lucky dip as to which you get 3.4/3.6Ghz or <2.4Ghz and what the underlying disk i/o performance/config you end up with.
  • @eva2000 Blimey, that's a brilliant thorough test you've done! Mucho kudos!

    I'm inferring from your comments that the "CPU" as seen by the guest is the same as the physical one on the host.

    Is it possible (and I've no idea - I've never fiddled with KVM hosts) for the CPU to be virtualised/throttled down to 2.4Ghz, or does this mean Vultr are actually now using slower physical CPUs than before?

    Cheers!
  • @jamie What benefit would they get from intentionally throttling down CPU speed? I am making assumptions here but, using E5s with more RAM as opposed to using E3 which is limited to 32GB of RAM is the main reason I see for the recent changes. Again, this isn't Vultr's fault, business is business. I really do hope they implement the option to choose those "legacy" E3 CPUs soon though.
  • edited May 2015
    @RagnaHosting Throttling guest CPU means more guests per machine!

    However, I think you've hit on the answer (also, with more cores per chip, more cores to dole out!)

    Are we concluding that KVM can't virtualise the CPU in any way (Other than allocating per core, and Feature-announcement masking)?
  • @jamie thanks - yes I am assuming guest cpu is same as VPS host node's of course we'll never really know
  • So will we ever get 3.4Ghz cpus back? I run game servers and 2.4GHz doesn't really do it for me. Most game servers I run cant take advantage of multiple cores. Maybe add a different server type like Gaming series? Or maybe do something like ramnode is doing?
  • 3.4/3.6 Ghz cores as an (more expensive) option is definitely something I would like to have as well. Thanks!
  • @FMSMITH91 I agree with that completely. I also run game servers, but the lack of DDoS protection and the fact that they've suddenly dropped support for faster cores has forced me to move elsewhere.

    Vultr, I think you messed up by suddenly dropping support for the faster processors. I feel like you should've warned everyone first, or kept the option available for the faster servers. The way I see it, those faster processors aren't going away so you could've allowed people to purchase those for an extra charge while having 2.5GHz as the default. The supply would always be limited if you weren't purchasing extra hardware, but at least it gives some people a chance to get their hands on a 3.5 whenever someone cancels theirs, at the moment it seems like they're being wasted because people can no longer buy them.
  • The faster CPU speeds were the only reason I went with Vultr. I just rebooted my Sydney instance for a kernel update and the performance has dropped by at least 25%. It's absurd that they made this change with no warning or options available. I would gladly pay extra to keep the higher CPU speed since that's what made Vultr unique. Now I'm looking at other providers.
  • @lambdadelta you mean rebooting existing instance dropped from 3.4+Ghz to 2.4Ghz cpu ? or you mean same clocked cpu ?
  • edited June 2015
    @lambdadelta I rebooted an instance about 20 days ago, and as I suspected...

    cpu MHz: 3392.152

    I doubt they'd change you to a new node on reboot. It would take ages to transfer the files to a new host node, especially if it was a large instance.

    Are you sure we're not confusing this with the placebo effect? What's your current CPU clock speed?
  • edited June 2015
    I submitted a support ticket about this a couple of weeks ago. They said they'd never change the clock speed of existing nodes, but eventually all newly created ones may be affected.

    There wasn't any explanation of why this change went unannounced.
  • @SneskyDave thanks for that .. indeed curious about @lambdadelta's cpu clock speed
  • It would not change on a reboot, the only way your physical node would change is if the node was degraded and we live migrated you to prevent downtime in the event the host node failed.
  • edited July 2015
    So out of pure interest, I made a single thread benchmark script, to compare 2.4, 3.4 and 3.6GHz instances
    Results (lower is better!):

    Vultr (LAX) 3.6GHz edition: 3191 E3? (Virtual CPU e7da7129d3ee)
    Vultr (SYD) 3.4GHz edition: 3501 E3? (Vultr Virtual CPU 2)
    Vultr (LAX) 2.4GHz edition: 5991 E5? (Virtual CPU e7da7129d3ee)

    For good measure, here's the results for DO..

    DO (LON) 2.4GHz standard: 6521 E5-2630L
    DO (SFO) 2.4GHz standard: 7829 E5-2630L v2

    So we're still reasonably faster than DO, fortunately, but there is no denying those 3.4/3.6GHz cores were amazing..!

    Fortunately I am lucky enough to have a 3.4 in Sydney! (I deployed two temporary instances in Los Angles, one being 3.6, the other being 2.4, all three being used for benchmarks)

    Hmm.. I might just hold on to that 3.6.. :-)
  • I've done the same ben, removed three instances and stacked them all onto the one 3+ghz instance.
    Ah well.
  • Running a 3.6Ghz in Sydney which i am able to run everything i need to on it. Prior to this i had 2x 2.4Ghz instances....

    +1 for paying a premium for the 3.6Ghz instances, be good to see a 3 core option too
  • I just got a 3.6GHz instance in New Jersey
  • I've been deploying a lot in the past week and most of the time I have ended up with 3.6GHz cores in New Jersey, also got a 3.4GHz just now and only twice have I gotten a 2.4GHz core. I'll be keeping my 2 CPU @ 3.6GHz virtual machine because it's a BEAST!
  • Then why pay for dedicated instances to get these cores?
Sign In or Register to comment.