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
Personally, I can't say much, other than from what I know (though very basic), is that from what we've seen, considering the security vulnerabilities that exists with Intel procs, AMD seems to be eyed by many now.
Edit: That's just from a security's perspective. As a whole, there aren't that many providing services off AMD procs, as compared to Intel, so maybe go for AMD?
All the way AMD. Intel's security is fubar and this should be a great no-no for any hosting provider/dc. Plus performence and value are right now way better on AMD.
AMD no matter what!
could you explain more in details why ? or why not Intel ? for which use have you chosen AMD ?
security
diversify your portfolio
I thought the newer AMD generation offers more bang for the buck, right?
https://www.phoronix.com/scan.php?page=article&item=3900x-9900k-400&num=1
https://www.phoronix.com/scan.php?page=article&item=amd-epyc-7502-7742&num=1
Benchmarks says all. AMD offer better performance and lower power usage at same or even lower price.
Of course it's gonna be AMD. Win win for provider and consumer
Intel for single thread performance, AMD for best price/value ratio (for now).
20 Years ago. Mostly Intel for servers.
Today AMD for servers, desktop and laptops. Not overpriced, fair performance for $.
So by the looks of it, AMD is currently “winning” in terms of Security + Performance + Cost.
Tagging @jsg
Whatever is the cheaper at the moment.
My understanding is that AMD has a better ratio performance / power usage / cost.
Like being said, the recurring security issues with Intel is adding to the balance, in favor of AMD.
I am only renting server(s) to run my own sites, which are not ecomm and I am not selling services, or hosting anything, so just for my own use. In that case, I don't mind if the server has an Intel or AMD cpu, I mind more about the connectivity, amount of RAM and drives.
IF, I was a provider of VPS for example, I would prefer to get servers running with AMD CPU, "today", because of security things.
But I am no one, my opinion has very limited value
Aside from some extremely small usecases in single threaded applications (less than 1% performance difference in some old applications), AMD is basically better in every aspect right now.
Just go with ARM. And fuck all those nefarious, childish "miners".
I am going to page the person I consider the residential expert on this topic @jsg
In terms of absolute raw clockspeed power, Intel still has a slight lead I believe. However, in terms of price-performance ratio and power consumption (critical for DCs), AMD is in the lead.
If you are a VPS provider, it is a no brainer because you can have way more cores per host node, which translate to better end user experience. Assuming that the percentage of heavy CPU use remains the same as Intel, for the price Intel is asking, you can offer AMD that has more cores, so other users will have more available CPU power translating into better user experience.
I just bought a second generation 2600X (6 cores, 12 threads, up to 4.2 Ghz) myself. Why? The price of this CPU can only buy me an i3-9100 (4 cores, 4 threads, up to 4.2Ghz). Pretty much sums up the state of affairs.
thank you for your point of view, could you develop more in details why ?
Will there be a surprise deal for ikoula?
It's not about price or "power" (which is either used for nefarious activities, or to compensate for poor design/sloppy code) - it's about getting the job done while minimizing resources use (and therefore long-term costs - for which "price" is a bad proxy).
By this metric, AMD wins.
Maybe, I haven't followed up closely the news on the "less than 1W CPU" front
For now, AMD...
For me, after testing some programs, If run a huge of similar process, there are basically no restrictions, but whether the process will be crashed is another topic.
While INTEL will limit when it reaches a certain amount...
So far with new equipment AMD is killing it. We have been using quite a b it since this new batch came out.
From our desktops to dual Epyc systems we have been super happy with performance.
Haven't heard anything bad yet from customers either.
Hopefully we see them keep kicking butt in both sections as you can see even a 1K drop in Intels latest high performance desktop/workstation cpu's! This is good news for everyone.
Been using AMD since I built my first PC 8 years ago, still my fav
Sorry to disappoint those who - understandably - prefer AMD, but ...
AMD doesn't really have a big product spread. Just have a look at my BanditHost review. Those high core count Epycs do not offer an enormous speed advantage. Don't be fooled by all those crazy benchmarks numbers out there. Most of them were done by gamers, overclockers, etc. and those do usually not look with the eyes of a DC.
The outstanding high performance numbers (e.g. seen in my NexusBytes benchmark) are Ryzens
Yes, AMD seems to have a lead in terms of performance/power consumption and pricing - but it's not overwhelming enough to make large customers jump and switch away from intel and/or their system providers.
But: companies don't tick like end users and especially not like gamers. intel knows that and AMD does, too. That's why AMD re-entered the market with very very low prices. We should understand that they didn't do that because they are nice guys but because they absolutely needed to. intels reaction? Not at all surprising, just turn around and look at intels new Xeon pricing. It's lower than just some months ago, brutally lower, but still higher than AMD Zen. Why? Because intel can.
As an end user/consumer I don't care. I just see the low price, the performance, and the attractive gadgetry (like PCIe 4) and the lower vulnerability - but then, what's my risk? Next to none. For a company that's very different. Also add the classical inertia of companies to the equation.
And again, don't mix up Ryzens and Epycs! Most consumers will go for a 4 - 8 core Ryzen which also means that AMD has a generous el. power budget. One can push a few cores to quite high clock rates (and hence performance) with something like 90 - 120 W. In the DC the picture is very different, there it's largely about performance per Watt, plus those Watts are limited, both by the sockets and electronics and by rack power density. Keep in mind that power per rack is limited plus one of the most important cost factors in hosting.
The result: AMD too has to go low on clock rates with their 24+ cores Epycs ... and the performance results is? Well a bit better than the Xeons. And keep in mind that bang for the Watt still is one of intels strengths.
In a DC the game is almost never performance per core. It's performance per rack. Yes, AMD seems to be better overall. They have the gate size advantage (7 vs 10 nm), but for how long? They offer more performance, but (a) not that much more that companies will hurry to switch, and (b) for how long? And so on. Plus they highly likely have higher manufacturing cost, the need to regain immense amounts of research costs and to then make a profit. They have one (1) single supplier and at least as of now no alternative, etc.
And then there is the BIG issue in the low end hosting market which is largely based on 2nd lifetime or at the very least on very long lifetime systems (and we simply do not have any experience yet re. how well those Epys age ...)
So, how and where is e.g. @Ikoula supposed to find those systems?
In the DC? Considering all factors: for a few corner cases. IF you can find and get those systems and processors and IF you really want new systems.
Desktop is an entirely different story. There the answer is simple. GO Ryzen or TR!
Nuh. Better wait for RiscV. Arm many of the same problems as the x86 world and I do not see any significant advantage in terms of safety/security. The ARM domain is small stuff, that's where they shine (yes, yes, I know about diverse many many cores Arms based systems which all seem to end up in some niches). Plus the vast majority still wants to run x86 code in a DC.
AMADEUZ.
You are right that there are some problems with avability on most expensive models like Ryzen 3950x, but in my country (Poland) you can buy other Ryzen CPUs without a problem.
Strongly disagre on, take a look at link below + few other sites that benchmark CPUs like phoronix. These tests are not made by gamers, cpu miners or overlockers:
https://www.servethehome.com/amd-epyc-7002-series-rome-delivers-a-knockout/
https://www.theverge.com/2019/5/7/18535078/worlds-fastest-exascale-supercomputer-frontier-amd-cray-doe-oak-ridge-national-laboratory
https://aws.amazon.com/ec2/amd/
https://azure.microsoft.com/en-us/blog/announcing-new-amd-epyc-based-azure-virtual-machines/
Amazon jumps to epyc, microsoft also. Ofc they dont dump they old hardware and switch to epyc, but if CPU market trends will be similar to trends today probably AMD is going to gain bigger and bigger market share everywhere.
https://www.phoronix.com/scan.php?page=article&item=amd-epyc-7502-7742&num=10
@dodheimsgard
I disagree.
Btw, why do you think that the low power CPUs are more expensive although their performance is lower than the non 'L' processors? Because those allow a provider to put more money making cores into a real rack and because their lower power consumption pays back the higher cost over time due to reduced (high) DC power cost.
To avoid misunderstandings: I am a big AMD fan and I'm extremely pleased with the Ryzens, Threadrippers and Epycs. But the question here was what a mid size provider should (and can!) put into their racks.
AMD EPYC and Ryzens have a disadvantage out the gate for standard distro usage with default Linux Kernels as 1st generation EPYC and 1st/2nd gen Ryzen need at least 4.15+ Kernel to shine performance wise and 2nd gen AMD EPYC needs at least 5.1+ Kernel to shine - ideally 5.3+. You're not really experiencing the full performance benefit of AMD EPYC/Ryzen otherwise.
But I agree from web host provider perspective, some don't really factor in performance for end users in their choice of cpus used. How else would you explain how 6-10+ yr old Intel Xeon processors still make up some web host provider's offerings these days
Curious about how discussion has moved to DC workloads, why original poster I assume is thinking about choice of servers to offer for web hosting = end user web hosting work loads ?
Personally, my eyes have been opened by my current benchmark comparison testing between the following servers for CentOS 7 with both default 3.10 and 5.3 Kernels
I'll maybe have access to Intel 9900KS 8C/16T and AMD Ryzen 3950X 16C/32T later too.
For me workloads = traditional web hosting related work loads like serving Nginx HTTP/2 and HTTP/3, PHP, MySQL, disk I/O, compression, crytographic loads etc. Guess which cpu(s) are leading the way ? Still doing testing 2+ months in, so will be a while before I go through the raw data for review/benchmark write ups and charting the various results
@eva2000
I largely agree, but in the given context performance is overrated IMO. Sure if asked "would you like an Epyc 7xx2" based VPS" almost everybody would say "Yes! Yay!" but that is not the relevant question. The relevant question is "Would you like an Epyc 7xx2" based VPS for $20/mo -or- would you like an E5-26xx v3 based VPS for $5/mo -or- would you like an 56xx based VPS for $3/mo?" - and frankly, the vast majority of VPSs will not have a significant and relevant advantage from the higher end nodes.
Plus there are quite some caveats, some of which you mention, e.g. kernel version but there are others too like e.g. the binaries coming with the distributions which of course have to address (almost) the whole space of processors out there.
Concrete example: I (the developer) could gain considerably more performance in most of my code (largely crypto related) ... but ... I usually have to compile for Nehalem because my software has to support the vast majority of processors the clients use.
Hell, I would love to use at least 256 bit AVX as it would make a lot of computations very much faster - but I can't, we simply can't provide and support 5 different x86-64 binaries.
+1 on cpu specific compiler optimisations. Can be problematic if you're doing the building of binaries on your end i.e. distro provider/repo provided binaries as opposed to on end user/server end built.
That's part of my AMD/Intel cpu testing work I do too as Centmin Mod LEMP stack can auto optimise the compilation of binaries for Nginx and PHP-FPM based on the cpu architecture and cpu instructions supported/detected on the server it's running as well as the web host detected (dedicated vs cloud provider). So I can eventually optimise specifically for Intel or AMD cpus.
It takes the workload off my end so I don't have to build the binaries myself for every cpu out there. End user's install will automatically compile/build optimal binaries for Nginx and PHP according to cpu detected
But GCC has FMV - Function Multi-Versioning which could compile one binary with support for multiple cpus/instructions https://lwn.net/Articles/691932/. AFAIK, that is what Intel Clear Linux built binaries do to optimise for different Intel cpu's differing cpu instructions supported. But my knowledge isn't enough to grasp how to use FMV yet.
If other Linux distro providers besides Intel Clear Linux started using FMV, that would be nice.