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.
Why are E3/E5 v2 CPUs still popular today?
It's been almost a decade since Xeon E3/E5 v2 came into the market, and I'm wondering why it's still much popular than v3 models and even the v0s in low-end market.
Here are some of my thoughts:
1. v2 models are much cheaper in the flee second-hand market than v3 ones;
2. v3 doesn't offer significant single thread performance gain compared to v2, especially in the area of integer/scalar computing;
3. v2 models outperform v0 models by ~13% in single-threaded workloads and by ~40% in multi-threaded scenario with more cores.
Are these presumptions making any sense?
I'm always interested in cheap plans supporting AVX2/FMA, for I sometimes test SIMD code processing floats on servers.
Comments
If you aint payin for the power, why care?
Providers that offer v1/v2 are ones who already have it for years.
If you take a look at providers that are actually deploying new nodes on fresh aquired then pretty much nobody goes below v4.
GreenCloud can be such example - they recently got DE nodes with $15/yr offer and its on v4.
Heck, even BoomerHost is on v3.
RackNerd is on v2, but thats because they own these servers for years.
Thrre is quite a jump between v2 v3 and v4 in terms of TLS/AES-GCM performance.
Doubledigit performance diff each generation.
I think not many people know about it, but that makes huge difference for nginx.
many providers using V2, probably rent instead of colocate.
Because the burst performance and the instruction set, say AVX2/FMA3, really matter in certain situation.
Let me give an example here.
Recently I worked hard for a vector search engine, and I had to mix all kinds of tricks to achieve a better performance than some famous open-source project, among which tuning AVX2/FMA3 instructions and taking care of L1 cache made awesome contribution.
That is awsome. Thanks for that point, and I never realized that.
But we could see a lot of giants, including OVH and online.net, selling v2 bare metals. I prefer to think @AXYZE makes the point:)
yes i had meant to say resellers as vps providers. get from colocrossing and OVH and they are cheaper to rent than V3/V4 using DDR4 ram.
datacenters which own them probably might have fully depreciated these assets and they cost almost next to nothing along with a pool of old spares, priced accordingly with some recurring costs, power, etc.
and who would care about app performance as long as they make profit. consumers who care about them will pay for Ryzen / EPYC.
I think it also has to do with platform compatibility since v1/v2 generally don't work in v3/v4 systems. I don't believe most hosts are going to upgrade from v1/v2 to v3/v4. They likely skipped across to Xeon Scalable (or EPYC), if they upgraded at all. A lot of the hosts that entered at v3 are also not likely to go through all the hassle, downtime, and work to upgrade to v4. This is especially true if these systems are racked, configured, running, ROI'd, and still have happy paying customers on them.
They just resell colocrossing, they don't own any of the hardware.
Main reason for people using e5v2 is that it's still "fast enough" and it's the most core dense x86 platform using ddr3 still. Ddr3 is currently pretty cheap since it's ewaste, so you can pretty easily get a lot of ram and a few e5 cpus and cram a node full of customers. As for big providers, they already own the hardware and it's easier to just continue offering it, the only actual costs are power/bandwidth.
I mainly use them as I either got them for a deal or they've been ROI'ed already. I try to squeeze the most of the servers til they won't lease anymore.
Correct
Our pre-built dedicated servers we rent out are v2 and v3 Xeons too
We have custom as well with much newer stuff.
But guess what. Customers rather buy the v2 and pay few Euro less monthly then going for v3/v4. And they still get loads of bang.
(I still don't buy much of this anymore)
We host few other hosting companies. Same here.
Well, I don't know if they rent it, but surely they also use other providers, for example: Racknerd LA-02 VPS is Multacom, shared hosting is from Velia. Racknerd AMS is indeed ColoCrossing, don't know about other locations as I didn't used them.
Yep, I have maybe 2 dozen e5v2 boxes (lots of cheap blades) and their compute power is honestly fine. Can get pretty dense too with the right chips, but I really like the 2650v2. Pretty good balance of performance + TDP + compute density. I actually run multiple prod workloads entirely on dual 2650v2 systems, performance is excellent. Main issue is power draw, but power is still pretty cheap outside of the EU.
All the colocrossing hosts always have one or two locations that arent CC, just to mix things up a bit. They have some diversity in LA, but that's all resold hardware. They use Velia for their shared hosting since besides being in Amsterdam, CC doesn't leave the US and I assume they wanted APAC locations, etc (plus every CC IP is blacklisted to hell and back so you can forget about mail deliverability). Dedis are resold CC + WebNX. I don't use them for this reason (and a lot of other ones) but it's decently smart to run a business this way since you don't have to deal with deprecating hardware. Downside is less overall control, but I'm sure considering they're currently one of the bigger CC customers they get decently well taken care of.
Totally agreed. And if I were a provider, I would have skipped Xeon Scalable and sticked with AMD (Ryzen maybe), since the latter offers about twice the former's single-threaded performance (again, for float-intensive work).
Damn right. As long as (1) Xeon Scalable and AMD 7000 series (+DDR5) remain fairly expensive, and (2) killer Apps, such as WP, don't make aggresive upgrade, we'll have to wait for quite some time before v2 retires.
I did some tests on a E5v2 vm from central.so, mostly floating point ops, and the performance is quite impressive.
Some E5's do have nearly 1k single core G5 bench.
If power is cheap, who cares.
Power in Dallas is stupidly cheap.
You could ask the question why are E3's still popular in general.
Regardless if its the original or v3 (or even later series including v6)
I can only speculate, but it may have something to do with it being a known quantity. Or just being a well recognized "name" in general.
That said, logically you would pick a E-2234 over an E3-1230v3 at the same price point. However in terms of interest, i can observe what appears to be a higher interest in the e3 models.
Pretty much all E3 CPUs are ultra low power and offer a decent spec for most low price points.
E3 1230v6 gives you 8160 cpubenchmark points and a Single Thread Rating of 2221 for a tdp of 72 watts that they do not reach often. It is old and used and available for cheap.
E 2234 is 9900 points and a single thread rating of 2871 for 71 watts that will be reached much more often. It is relatively new and availability is a concern.
So performance wise E3 is still in the game and it can be much cheaper as a dedicated server.