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
X64 : everything works out of the box, on the software side.
armV8: adventure time/ student of technology arts / learning opportunities. All the best
ARM UnixBench score 407 (testing 1 core) https://www.lowendtalk.com/discussion/comment/2194602/#Comment_2194602
x64 UnixBench score 744 (testing 1 core too)
https://www.lowendtalk.com/discussion/comment/2194635/#Comment_2194635
82% faster -- almost 2x.
I don't see how they can be seriously trying to sell these at the same price tbh.
Yea that's what I don't understand.
I guess there is no reason to get ARM unless you actually need it for some reason.
Exactly, ARM should be priced at 1.99
You pick obviously x86, more benchmarks:
https://wiki.x8e.net/doku.php?id=dedicated_benchmarks
ok x64 is out of stock in both FR and NL lol
I use ARM. It's stable and behaves good. I really had no issue so far. The only problem would be when hosting special compiled software, but so far I sticked with Debian/Ubuntu repositories and had no issue.
My only problem is IPv6 implementation, but on Scaleway you can't reserve IPv6 to begin with.
So they seem to of double the amount of cores today for their lower plans for no extra charge and added some higher end ones like the intensive. It seems suspicious they never got back to me with proof that arm was faster then x64 all of sudden went radio silent.
Here's the blog for anyone interested. https://blog.online.net/2017/06/22/armv8-scaleway-new-high-core-variants/
Yeah the 2-core ARM was about exactly 2x slower than 2-core x86, both at 2.99 EUR. Now they should be about even, assuming you have a task which can use multiple cores well.
x86 still has the benefit of having the AES acceleration.
However with x86 I experienced a high amount of CPU steal on some nodes (oversold CPU!!! -- up to 50-60% steal at full load; 10-20% now on a luckier node). I wonder if this is present on ARM instances as well.
In fact I have to wonder if they actually reduced the instance density to accommodate for more cores on the two cheap plans, or is it just more oversold.
That's the main concern, if you're pushing any kind of https traffic.
Yeah that was pretty much what I showed them when I asked on Twitter. They said to me that had proof/benchmarks that proved otherwise and thought I meant the C1.
Yeah I noticed a lot of steal time on certain ones too
I still have a huge problem with the idea of IPv6 lost on server start/stop. I simply can't use Scaleway.
Yes, however the difference is not a big as you might think: https://www.servethehome.com/cavium-thunderx-micro-benchmarks-enterprise-arm-developers-need-machines/
Don't they auto-assign a static dns name for your server, regardless of what ipv6 you get provisioned?
That doesn't appear to show the actual AES stream encryption benchmark -- only RSA sign/verify, for whatever reason.
ARMv8 does have AES instuctions
http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.ddi0514g/way1395175472464.html
The article you linked says that the Cortex A57 supports those AES instructions. The Scaleway CPU is a Cavium ThunderX (not a Cortex), but this appears to have a cryptography engine too.
Thats how they are selling both of them in same price?
It's not as relevant what it has, the question is, is that engine supported today in the kernel, OpenSSL and apps that you use. The VIA Nano CPU had its own AES engine too, but it was a major pain to get it working (recompiling OpenSSL yourself), and it seemed to be incompatible with some applications entirely (Tor).
You make a very good point. I imagine you need at least quite a new compiler to deal with it, although even that might not be enough. Maybe someone who has one of these servers can show us what kind of AES throughput they get.
I haven't checked these CPUs specifically, but many ARM SoCs don't have AES acceleration in the CPU itself, rather it's a separate engine on the SoC. Perhaps see if there's a /dev/crypto or similar device available.
Compilers generally don't help, the program itself needs to be coded to use it.
to blow this thread back alive:
`Dear cloud rider,
We're glad to announce both a huge performance increase for our current ARM64 lineup and bigger, more powerful, variants of the ARM64 servers.
The ARM64-2GB server gets a 100% core count increase and the ARM64-4GB variant gets a 50% core count increase. All of this for the exact same price as before.
Starting at €2.99 per month, these variants offer the best compute-price ratio on the market.
We're also excited to announce the new ARMv8 high core lineup. This lineup provides an incredible amount of memory and CPU for intensive workloads. These new servers provide up to 128GB of RAM, 64 cores and 1Gbit/s bandwidth with unlimited transfer.
To learn more about the performance increase and the new ARMv8 lineup, checkout the announcement blog post: `