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.
Comments
CPU can run in two modes, with and without hyper-threading (what does AMD call it nowdays?) enabled, there are significant trade offs to hyper-threading and many choose to disable it (especially in certain industries).
Regardless of whatever shenanigans are being played, HyperThreading/HT is Intel BS. It’s one core, two threads at best.
Yep, at best.
I also don’t see how people just accept that the definition of a core is the same on CPU spec sheets and on dedicated servers but on a few VPS companies, it’s something else..
The common oversight by most users that 2 HT are not totally independent of each other for same physical core.
I ran into this issue with @HostEONS dedicated 1 vCore VDS and learned this hard way
dat's hard core-n corn porn
@bdl lmao, thread lighly, otherwise, mods may ban you of cores
The original cause for this mix up is probably that HT simply doesn't translate to VPS. You can't add it to your VM and if you try it looks like a core. The only almost correct solution would be (if HT is enabled) to pin 2 threads from 1 core to the VM for every core listed. Without pinning you can never know if the VM actually got a full core or not. I've always felt that calling the result of giving the VM access to a thread vCore was an acceptable compromise.
Honestly, it's all really complex. Especially with Intel's P and E cores as well..
But yeah, vCore or vCPU or thread could all be fine I think.
I don't even demand pinning here, I just think that a host that criticizes other hosts for their CPU practices shouldn't call whatever vCPU/vCore/thread is a core.
Good point. Those will be a total headache. Soon we'll probably have vCores, premium vCores and economy vCores...
I skimmed that thread now; IMHO Avoro does false advertising in proposing dedicated cores but delivering shared cores with multiple tenants; a hyper-threading core, if dedicated, has only one tenant instead. Aeza more correctly markets this tenancy as "dedicated resources" instead of "dedicated cores", which allows 100% usage of some unspecified share of the cpu.
With that reasoning: Would you say calling a thread a core is false too?
The whole point of the thread, except for shit posting, is this:
You can either say that one may market using words/slang that we generally as a community understand, while it technically isn't really true, or you may say that doing so is wrong.
You (not you @davide, more in a general sense) can't say that Avoro not pinning threads/cores and/or not hard limiting threads/cores on 'Dedicated' CPU plans is wrong, but maintain that HostHatch calling threads 'Cores' is right. Both fall under the same category of somewhat abused slang terms that we have just come to accept.
I didn't find what Avoro was doing particularly wrong, nor do I find what HostHatch is doing particularly wrong. But it's in the same league. Let's not pretend that one is trash and one is god-tier.
It's all really not true, given the actual meaning of the words.
not 'apt get', its 'apt-get'
thats why you cant find your missing cores:D
one thing for sure
this is such a fun week for LET
Definitely agree with this.
In some stretched way, it's like comparing two ponzis, one is still paying out, and the other has already collapsed. It's difficult to call out a ponzi when participators at a time are generally happy, infact it often leads to abuse (ever attempted to talk anyone out of early days MLM for example) but ultimately you've still got a ponzi, if everyone comes for their cores it's all over, just like phpfriends/avoro reputation.
Yep, cores and hyper-threading cores are hardware components, threads are software. Ignoring the implementation details of the hardware, some chips present themselves to the OS as having 2x the number of their cores; the reasons for this are implementation details, but "2x cores" is an attribute of the hardware nonetheless.
Non-hyper-threading cores aren't discrete quantities either, some chips have fewer ALUs than their number of non-hyper-threading cores, with the ALUs contended among the cores; these chips also present themselves to the OS as having N cores, but we might interpret them as having only a fraction of N complete cores. These again are implementation details of the hardware.
Catch this 🖕 Avoro
Let's take the AMD EPYC 7443P for example. It's advertised as 24 cores, 48 threads. When HostHatch gives you 1 core 100 % dedicated, do you think they deploy sub 24 of this, or sub 48 or this?
Edit: Misunderstood you. I agree. Both are technically false. Or honestly I'm not sure if you think what HH does is wrong or right.
To me, HT = several threads per core. When given 100 % dedicated access to one core, therefore, I should gain 100 % dedicated access to the threads that runs on said core. HT is basically, as you said (I think), presenting itself as something it's not. But even with say 8 cores and HT, product specs don't list them as 16 cores, and neither should VPS providers.
I also use HT and SMT interchangeably even though obviously the AMD mentioned won't have HT.
I'm gonna lawyer up for this, you're scaring me now.
From the OS perspective, one core is one hyper-threading core. If the box inside which the CPU is sold didn't inform the buyer that the CPU has N cores, but internally only N/2 structures of some kind exist, and if hyper-threading was hard-coded in the hardware and not exposed to the BIOS for the user to be amused by the knowledge of its existence, we wouldn't even complain about what VPS providers consider a hardware core.
Philosophy apart, hardware cores are hardware implementation details, threads or vCpus are marketing notions, or a scam in case of Avoro🖕™
I think I know why I was confused, when you say "Hyper-Threading Core" you mean what I'd just call a thread, right?
I first thought that it was a core with HT capabilities.
I believe that the main issue here is that for basically all applications, 2 threads or logical cores on the same HT-enabled physical core won't perform even close to 2 threads on separate physical otherwise idle cores. That's why 48 logical cores isn't advertised as 48 cores on the packaging, because it can't perform like 48 physical cores would. It's more like optimizing how the 24 physical cores handle processes.
As long as the manufacturer won't claim 2x the cores, I don't think the provider should.
Maybe I'm not knowledgeable enough to have this argument, as I may be missing something, but that's my understanding at least.
I love to see my inspiration in how you think, talk & write.
for me, if the performance is good for the thing I use and does not have cpu steal more than > 1%, I am a happy person.
but I agree, they need to stop shitting on each other and praising themselves as the best of the best.
I'm honestly not sure if this means you're inspired or if you're saying that I'm inspired, nevertheless, thanks & I felt like I was actually trying to say was kind of close to what you said about providers offering fairly similar stuff going at each other like they're so special themselves. Although we may not agree about Crunchbits, it's a highly competitive and cut-throat market. Words will be used to their limits. Unlimited will be limited, unmetered will be metered, dedicated will be shared, cores will be threads, there's probably quite a lot more examples of words in this industry not meaning what they mean outside of it.
I'm not even sure that I'm against that, maybe I am. I don't feel strongly about it anyway. And it's not like providers are particularly greedy either, their profit margins really aren't generally that good.
I would probably have joined in on the Avoro hate-train if they were unique in any way, as deep down, I know that they use words in a way that's not really compatible with their true meaning, but that goes for a lot of providers. Even outside of LET.
Google does it very nicely, they use "vCPU" and explains what that means across their product series here: https://cloud.google.com/compute/docs/cpu-platforms
But Google can't be compared in price to a standard VPS provider, so that dosen't mean much.
I guess, at the end of the day, to me at least, the sustainable long-term performance/price is what actually matters. People seemed to think that the MOST honorable thing to do would be to do what Cherry Servers did, and pin threads for their VDS line, and maybe it is, but my take-away is that the performance/price that they can offer while doing that, won't come close to the performance/price that can be offered by Crunchbits, Hetzner, Avoro, PHP-Friends, HostHatch, Netcup, and so many more providers on here.
Taking Cherry Servers vs. Avoro for example, you can get 14 threads of AMD EPYC with Avoro for the price of 4 vCores on Cherry Servers, how often will Avoro fail so badly that the 14 Avoro threads will perform worse than the 4 Cherry threads? Probably almost never. Again, same thing likely goes for Crunchbits, Hetzner, PHP-Friends, HostHatch and Netcup, or any other VPS provider known to offer higher performance for less.
I meant your entire personality is borrowed from me down to your writing style & syntax. I love to see my inspiration in you.
I wouldn't say our personalities are that much alike. I'm not even remotely as toxic. Although, I agree that my take here was relatively close to yours, but it wasn't really the same. We didn't agree about Crunchbits for example.
Quit biting nerd.
Things like this. We’re not the same. Although, speaking of completely stolen personalities, remember @CheepCluck? Why did you suddenly try to be him?
Take your schizo pills.
For my specific workload I get 80% more throughput on a E3-1245 v3 mounted in my workstation with hyper-threading enabled; but this workload is peculiar: frequent branches in hot paths, 0.5% cpu cycles spent on LLC misses, no SIMD. I know that most other workloads get a smaller improvement from hyper-threading. This guy bench-marked a 33% increment running off-the shelf GCC processes in parallel on an Intel Atom.
I've googled just now that the performance increment is only ~15% for other types of benchmarks. But still, this is not a marketing gimmick to falsely advertise an inflated number of cores. Or at least it's not primarily a gimmick, because it does yeld a performance increment.
VDS providers use the ambiguity between logical and non-logical cores to their advantage, but both definitions are valid IMHO. In the case of Avoro🖕™ though, the definition of core (not vCore!) was entirely made up.
Mentally strong people disable hyper-threading.
It gives more predictable performance.
Obscene
Hyper-threading does improve throughput for me. But basically I only run that one 8-thread process on the workstation and my benchmarks are biased to the specifics of the case.