Hacker News new | past | comments | ask | show | jobs | submit login
Apple silicon M1 as-a-Service (scaleway.com)
362 points by edouardb on Feb 2, 2021 | hide | past | favorite | 226 comments



Here are some marketing videos: Scaleway: https://www.youtube.com/watch?v=jZJnrKjfA3s&feature=emb_titl... Amazon's Mac Mini: https://www.youtube.com/watch?v=Pn3miC_tTH0&feature=emb_titl...

Pricing:

Scaleway Mac Mini M1: ~$85/month (converted from 70Euro's)

MacStadium Mac Mini M1: $109/month

Amazon Intel Mac Mini (they don't have m1): ~$26 per day, ~$780/month. Therefore, at NO scale, does Amazon mac mini sense. Amazon's pitch is to help companies reduce their infrastructure problem. Except for every machine you use on Amazon for a month, you could buy a brand new M1 mac mini


I think the scale where it makes sense is at companies where paying $780/month to an existing provider is easier than all the paperwork and internal regulations around allowing people to provision on prem equipment or use an untrusted provider.

Consider the item from yesterday about AWS Outposts: https://ahl.medium.com/aws-outposts-2021-stories-for-folks-w... .. the markup is absolutely mind boggling, but there are companies who pay it.

AWS is the master of having a pricing model that feels nice to most indies and small/medium companies but that scales up extremely effectively to capture huge amounts of enterprise spend at the same time.


I mean, you're not wrong.

For example, I know many furry artists who open their commissions and within an hour get 20 solid offers to pay them. They can't possibly do all the work, so they just lose business.

That's, uh, not how capitalism works! Companies like Amazon know, all you have to do is charge 10x more to capture all of that value. Sure, it sucks for 18 other potential customers, but these furry artists have tens of thousands of fans who will enjoy the work regardless.


Here's a really interesting pricing model your friend might be able to use to capture more value (and not make the community mad).

https://kevinlynagh.com/notes/pricing-niche-products/


Great example of the power of auctions! It's amazing to see the Vickrey Auction come off the blackboard and enable an indie creator to connect with their community sustainably. Markets really are all about moving information so we can coordinate activity.

Wouldn't it be awesome for Kickstarter / Patreon to add this as a feature?


Reading more of it, it looks an awful lot like the business model behind YCH auctions-- Your Character Here. That's a very successful model, and it often results in prices $400+, which is about 2-3x more than what artists often charge otherwise.

That said, YCH is a more traditional auction over a single piece, not a Vickrey auction over a specific supply. According to this model, a Vickrey model would be:

- Scope the work to a specific standard that the artist specifies, in terms of colors, quality, characters, background, etc. That way, the bid is on a roughly equivalent amount of labor per piece.

- People put in (blind) what they're willing to pay

- When demand is exceeded, the nth+1 highest is what everyone pays.

Very fair!

I'd be curious if there'd be a way to price something more custom; for example, putting a price on the number of hours required for each feature (more color or shading, more characters, etc.), how would that affect such a pricing dynamic?


Does such an auction hold under the modification made for multiple participant?

Generally if you intentionally overbid you end up paying the otherwise highest amount discouraging that kind of action (ie you may as well put your max amount).

But if the top 10 pay the price of the 11th bid there’s a perverse incentive for me individually to just say I’m bidding $1 million because I know that most other players are likely to not be smart enough (at least in the short term) to bid strategically and thus I’m guaranteed to get the item at a steeper discount without having to reveal (or bother thinking about) my true max price for the item.


If you do that, there's a real risk you'll end up having to pay more than your max price. Say your max price is $2500, but you bid $1 million. Then it turns out 10 others have a max price of, and bid, $10000. You are now on the hook for $10000, significantly more than your actual max price.


That's a really good idea! I'm going to send this to the people who make https://commiss.io, thanks!


You're assuming that 2 of those 20 potential customers would pay 10x, when it could be that 20 of the 20 would balk at 1.1x.


Of course that requires some market analysis not a straight up "let's charge 20x". But when talking about limited capacity in general it makes sense that you overcharge to capture as much value as possible. Some customers are always willing to pay more so if you somehow know or can guess the distribution of offers it makes sense to capture everything from the most valuable buyer group down to whatever your capacity can cover rather than a random collection in the middle.

In the case of the M1 it might make sense to charge at these levels especially for customers that aren't interested in using anywhere near the full value of the product. So if a company or developer estimates it will only use the equivalent of 50% of the Mac's price it would be a better deal than purchasing the thing outright and also pay the associated operational costs. As always you have to know your market if you want to maximize the profit.


Not a furry artist but similar vein. There are genuinely a surprising number of people who are willing to pay 5 to 10x the market rate for commissions.


That’s first step of capitalism.

When someone makes crazy profit, you’ll have fast followers.

More artist will switch. Prices will drop. At some point some artist will drop out and a rough equilibrium will occur.

Then one of the artist will lobby a law that limits the number of furry artist.

Prices go back up due to restriction.

People complain at high cost of furry art. Licensed artist will lobby a new tax on all artist to subsidize the cost of furry art.

Now non furry artist are paying the furry artist.

Customers get confused on using tax vouchers on Furry artist.

Non furry artist who complain are now regularly attacked for not supporting the Arts.


Ouch. That's cynical. But, sadly not too far from the truth.

But, what's the alternative? Capitalism without human nature?


Everything in story was fine until government got involved.

Capitalism increases efficiency by being ruthless. Better, cheaper, more or die.

Government and regulatory capture screw it all up.


But what about the trees, treeman79?

The problem is capitalism optimizes for profit, not life.


Some regulation.

Not allowing Dumping toxic Waste in river is one thing.

Banning ride sharing In cities to protect taxi companies is another.


>AWS is the master of having a pricing model that feels nice to most indies and small/medium companies but that scales up extremely effectively to capture huge amounts of enterprise spend at the same time.

I wonder if there are any other industry that does the same. Most ( if not all ) of the time pricing dont scale to both ends. But AWS pricing seems to be the outliner here.


Taschen is a publisher that has somehow pulled this off in what you'd think would be a highly commoditized space: books.

They sell normal sized versions of most of their books for $20-$80, some slightly premium ones for up to a few hundred dollars, but then have these large coffee table sized books in limited editions selling for $10k+: https://www.taschen.com/pages/en/catalogue/photography/all/0...


Normally when I shop around for other cloud providers, I see AWS’s “pay us more but everything’s under one roof” strategy in action: Other cloud/infra providers may beat them on price, but IME it’s often like 10-20% at most. Maybe 2x in some cases.

It could be that there are companies for whom a contract with MacStadium is too much to handle, but for anyone who isn’t a supergiant enterprise, I just can’t see how an 8x markup over the competition makes any sense.

I’m sure at this point some enterprise customers are just hard-wired to reach for the AWS solution on everything. But I also wonder if this way of thinking has an actual hard limit somewhere.


I'm not particularly experienced at the enterprise level but I sometimes roleplay the conversations I expect they have in my head ;-)

I can just picture a department needing to do some CI on a Mac, having access to a certain (high) budget on AWS, and going with Amazon's offering rather than put together the business case to authorize the MacStadium spend. Or if it's under their approval amount, figuring out who's responsible for the thing when inevitably staff change around, etc. If the company doesn't bat an eyelid at another $800/mo on the AWS bill, I can see why many people would be tempted to click the button and get on with their work day :-)

I would appreciate people who live in this enterprisey world to fact check me on this as my experiences in this matter are rather limited..


I can confirm that if you have a large contract already setup it is much easier to add to it than go through the approval process for outside spend.

I can click a button that costs us $5k and nobody blinks an eye, but I have to request permission to spend $100 to do the same elsewhere.


If you have a giant pile of data already in AWS, that you need to use on a Mac for a short time, it may make more sense to rent a Mac next to all your existing data, rather than paying exorbitant transfer fees to get the data to a Mac elsewhere.

AWS's transfer fees are a massive lock-in.


except this would not be the case. we just need M1 instance from Scaleway or whoever provide it as runner / job executor. not much data egress for those traffic.


That's a big assumption on how EVERY company that needs an M1 and uses AWS uses it for


The most common use case for a Mac in a business setting is for building iOS applications.


You may have to define business.

Account managers, web developers, python developers, Julia developers, sysadmin, netsec.

All those areas are riddled with Mac.


But none of that code runs on Mac alone. Even if they're developing on a Max their python or Julia or whatever code that they want to run on the cloud will run on Linux. The only real use case out there is building/testing iOS/Mac applications. The only other one I can think of is a Mac only video editing or animation software that you'd prefer to render somewhere besides your laptop.


Although to me-and-you and to an extent SME's it may seem crazy to buy one over running it on-prem or at other companies to large organisations it makes perfect sense and the cost is just a drop in the ocean compared to their normal AWS monthly bill. Lets not forget AWS will only invest time if there are customers who will actually use it.

A few points from the top of my head:

* Large enterprises won't pay the costs listed publicly, they will each have their own contract written up with AWS.

* Contracts - it can take YEARS for new companies to be added to the approved vendors list.

* Frameworks will already be in place for how resources are managed and accessed securely.

* Enterprise support agreements will already be in place.

* Physical security that is regularly audited


Given the ridiculously low power requirement of m1 chips if anyone wants to use these probably in small scale capacities they should just buy it. The power and maintenance won't be that much if a hassle (10 of these would use less than 300w) and you would get warranty too. I think the mac mini m1 was designed (lack of) by apple to replace current mac mini racks but ideally this thing can be as tiny as a 3rd gen apple tv and a rack of 10-20 could be as small as a mac pro ...


Well thats what we need from Apple next. The Mac Rack of 10 M1 chips (aka. Mack). Not sure how useful that would be for me personally though. My reasoning being, if 1 chip takes 30W, why not double the number of chips or do something to the chip to get more from it.


I wouldn't be surprised if Mac Pro version of the M1 has everything turned up to 11.


it's going to be interesting with the 16 inch macbook pro the imac and mac pro. of course apple has the full licence with ARM to do almost anything they want but based on my understanding the X ARM cores the 7 series and 5 series can be clustered. My guess would be apple moves to a cluster based design eventually like in m1 but with more cores for higher end machines but right now i'd say just a chiplet version of current m1 chips would still be a great option with higher frequency. A 4 or 8 chiplet design would still be only 150w or so with gpus disabled or somehow using all gpus as one. This is the strategy amd has gone with obviously with chiplet without gpus but I think this also makes sense for apple to reuse its designs.


Or maybe it's just a big I/O & power box, with a bunch of slots on top that accept a Mac mini. Like a more stylish SNES.


don't hold your breath. There are good reasons the X-Serve line doesn't exist any more.


Do those reasons still exist in 2021? What are/were some of them? (Genuinely curious, not playing devil's advocate)


I have no idea why it was canned, but it was a very expensive server with a very expensive SAN solution. I think it just didn't compete well with other enterprise class machines on price.

People are willing to pay more for a Mac Laptop/ desktop because the overall experience is nice. The screen, the construction, keyboard (save the shitty one), trackpad, Retina display, high speed SSD, etc all significantly affect user experience.

When the primary use case is sitting in a closet and serving up bits quickly, it quickly boils down to bang for the buck where the Mac has always lagged. The M1 could very well change this though. Particularly since power use is significantly better.

Also, when the Xserve was around, there weren't nearly as many iPhone developers looking for CI solutions. Seems like now there would be a pretty healthy demand for that alone.


If you have, for example, a profitable cross-platform mobile app, but your developers primarily use windows/linux and test on Android, then it might make sense. If all you need is to spin up a mac for the occasional build/test/deploy, then paying $85 per month is easier than having to maintain an M1 in the office that no one is going to regularly use, especially if your team is distributed and/or remote.


In many large companies, operational costs are often budgeted differently from capital expenditures.

The workflow from approving a $85/month cost can be significantly easier than approving a $1,099 purchase.

Also, renting an machine from a 3rd party might allow you to avoid dealing with a dysfunctional IT department.


AWS’s margin is Scaleway’s opportunity, you might say.


Normally yes, it should be, but I don't think it's going to happen. People have been (1) miscalculating the costs of AWS, (2) realizing they are relatively high but deciding to stick to them for various reasons for a long time now. And because Amazon managed to capture many companies with their marketing, in many cases it's practically impossible to get out.


Yeah. It was just a snarky play on the famous Bezos quote, "Your margin is my opportunity."


This comment is on most AWS stories. Bandwidth charges alone are usually significantly higher than you can get elsewhere.

My own guess - a bit of a niche product not super easy to scale do they set a rediculous price. If the governance wins are there - they may still sell some - they rarely do stuff w out customer demand. At least in govt contracting - if you are willing to jump through all the hoops, some insane markups exist.


I would assume that Amazon's offering is targeted at customers who already using other AWS services. So even if a Mac Mini costs $500 per month more, it's really not that big of a deal considering how much the bureaucracy of signing up with a new hosting provider would cost in a big company (e.g., legal, infosec, ...).


Ohh yes. We had built a small tool that business people had to use, in form of a questionnaire, that went through about 20 topics. Depending on the question/answer it was an immediate "nope, must-have, don't sign a contract".

Ironically this (nope-case) was always the case if the provider ran on AWS.


Few things on AWS make sense financially. They're just leveraging the power of lock-in™. Why save money on AWS when you can save money on employee salaries?


But that would be an asset, not an expense. It depreciates over time and your accounting department must track it somehow in the books.

Ok, maybe you could sign a leasing for the mini; and still be cheaper than AMZ but... your company would need some credit scoring and your finance department should talk with someone...

But for $26/day it gets buried for free in a bill that needs a cloud architect to decipher.


Makes sense if you only need it for an hour every day to do a build and/or run a QA automation test. The billing on it is per-second.

Worth noting that they require 1 day minimum to play ball, so kicking the tires is not a per-hour option. There is also no spot market for them, and no reserved instance pricing either.

It seems like AWS has really replicated the Apple experience in more ways than hardware.


I understand that the 1 day requirement is from Apple and thus AWS requires that minimum too.


Why did Apple require a 1 day minimum?


>https://www.youtube.com/watch?v=jZJnrKjfA3s&feature=emb_titl...

Start watching the video, thinking no baguette?? 1:03 here it is!! Thanks France, love your sense of humor!!


Those are MRPs. Amazon will be giving heavy discount to large customers anyway


Anybody knows if Scaleway it's better than the parent company?

I was an online.net costumer for a lot of years for a personal project, and a happy one, but then they changed the contract unilaterally, which included the billing going up almost 75% with another changes in the ToS. And look I get it, sometimes prices rises, and you can always change providers, so the problem was not that... but the customer service and poor communication.

I only notice the change, after my credit card notified my of a higher charge than the ones I'm used to. There was not notification about it in the dashboard. And after a lot of tickets with representatives, they told me under the excuse of "ram / hdd prices are going up" the notice was send through email, and probably gmail mark it as spam... and then the nightmare to cancel begin.

After telling them to cancel the account, they told that under the new contract (that I didn't accept) they could only do it if I got charged another extra month, even if from the start you always payed one month in advance, plus paying the setup fee at the beginning.

Look I get it, hardware it's messy and prices can go up to stay in business, but this attitude of "I am altering the deal, pray I don’t alter it any further" was the one that made me look for another solutions, and be skeptical about Scaleway despite in paper being a good alternative to Digital Ocean (that let me tell you, DO customer service it's gold in comparison to what I got from online).


If you are looking for low-priced servers I can highly reccomend HETZNER[0].

They are even cheaper then scaleway and I had no problems so far.

[0]https://www.hetzner.com/cloud


Would have loved to use Hetzner, sadly they rejected a picture of my ID card/passport (with some details blacked out) a few years ago when i tried registering and verifying my identity, i don't think there was any process for appealing to have someone else look at it again, thus my account ended up being blocked. Of course, this is just 1 data point and i'm sure that the experience of other people is more positive in general!

That said, the prices actually look pretty competetive and are indeed cheaper than Scaleway's offerings. However, Scaleway also recently came out with more cost effective instances, called Stardust (though the specifications are somewhat limited): https://blog.scaleway.com/a-star-is-born-as-scaleway-launche...

Personally, i use Time4VPS for my VPSes, which is a somewhat small daughter company of a larger Lithuanian telecom (Interneto vizija). Affiliate link, if anyone is interested in having a look: https://www.time4vps.com/?affid=5294

I've been using them for a few years, the uptimes and the performance both are generally pretty good, can even self-host an e-mail server and the few IPv4 addresses i've had from them haven't been blacklisted anywhere. However, they don't directly compete with SaaS or PaaS like AWS's or Scaleway's managed offerings.


Why wouldn't they reject a ID/passport with some details blacked out?

They have proper support lines where one can call and ask. (At least this was so a couple of years ago, no need to call support since then..).


> Why wouldn't they reject a ID/passport with some details blacked out?

A better question is why would they? Sending an unredacted copy of my ID to some foreign company just to buy some hosting would seem crazy to me.

Here's a quote from their authentication email when I signed up a few years ago:

> As a new customer, we kindly request that you provide us with a copy (scan/photo) of your passport or ID card for authentication purposes. You can blur or block out information that is not necessary for authentication.

I sent in a picture of my driver's license with everything except my photo, name, and address blacked out, and they approved it. (In hindsight I'm not even sure why I didn't black out the photo part.)


I stand corrected with them mentioning blur/block in the authentication email (this was not the case earlier). But anyhow, what is for you "some foreign company" and "some hosting" is for them "some customer" which most likely is nice/friendly but there are exceptions.


> Why wouldn't they reject a ID/passport with some details blacked out?

In my subjective opinion, a better question would be: Why would they reject an ID/passport with some details blacked out?

The ID card and the passport both have serial numbers on them, as well as sufficient information (issuer country and the institution within the country), that any decent implementation of validating such data should be able to look it up.

As for the blacked out fields - they contain information pertaining to my social identification code, one that should be kept private under most circumstances (analogous to social security number in the US). It feels like that information could:

- provide absolutely no benefit to them

- compromise my privacy, should it be leaked due to improper handling of that data on their end

If not providing such sensitive information is grounds for denying account verification and termination of an account, then i'm perfectly fine with that and will simply look elsewhere, for other services. Other people may disagree with that and provide any and all information, the handling of which they deem acceptable, which is okay (hence, i mentioned that others' experience might be more positive/successful in that regard).

Maybe someone who works in the industry (not Hetzner in particular) and has worked with similar systems in the past may comment on what the mechanisms for validating ID cards and such are, and which bits of data are necessary and which aren't?


I might go a further step, to:

1. Why does some webhost think it gets to ask for your passport?

Lol the second they get hacked you have a world class identity theft nightmare

Go to a host where asking for your most protected personal information isn't part of doing business. This is absolutely ridiculous


No kidding. Linode does a good job for cheap VPS machines, as do DigitalOcean and Vultr, and they don't feel the need to ask for identity documents. I haven't heard of big cloud providers (AWS, Azure, etc.) doing that either. Perhaps this is some quirk of German law? In any case, that would be an immediate reason not to do business with them.


There are bad actors. I don't know details but suspect that passport requirements help with deterring them. If Hetzner gets a bad reputation it affects all their customers.

Nice that you have options which meet your wishes. But I like the way Hetzner is doing business (not to mention the great quality-price ratio which is hard to find elsewhere).


When I ran a hosting provider, a small US-based one, we had to block most international signups simply because every one of them turned into a spammer. Abuse is a very, very big problem. They may have discovered that individuals from specific locations are generally troublesome and require stricter verification.


I've got one of Hetzner's machine from the "auction" side (https://www.hetzner.com/sb) for various things that has been running along nicely for a while. Paying a pretty keen price for 3x3T drives (unless the storage is on a RAID backed SAN I never get a single drive remote machine), 32G RAM and a fairly decent (though far from leading edge) CPU.

For cheaper (and other options in the same range or more expensive too) in a large array of configurations https://www.serverhunter.com/ and https://en.metadedi.net/ are useful resources, though obviously do a little research before signing up for a host based on it being cheap.


Hetzner Cloud is nice! When I benchmarked SCW cloud and HZN cloud, Hetzner was the clear winner. SCW was a bit cheaper at the time. Now Hetzner is cheaper and faster! You just don't get as much bandwith.


Hetzner is great for cloud but especially for dedicated instances. If you have your infrastructure as code, buying dedicated servers there is likely much cheaper and easier than using a large cloud provider and trying to scale.

Only issue I have with them is that there's no way to get 10gbit/s (not even burstable) and that you don't get datacenters outside of Europe.

Not affiliated with them, just a very happy customer.


I haven't used their 10G uplinks in years, but it seems they're still easily available at €1/TB over 20TB:

https://docs.hetzner.com/robot/dedicated-server/general-info...


you can order 10g uplinks for most of hetzner's dedicated servers (for ~€45/m).


I second this. We use Hetzner Cloud for all our staging servers and have never had an outage, and I had a client on a bare metal Hetzner server for about 12 years before the hard drive eventually gave up.


I have been using Hetzner and they are good and cheaper than AWS.


I will check them out, the prices doesn't seem bad at all. Will try for a small projects, and if everything goes well maybe migrate bigger ones.


Hetzner was one of the major sources of attack I experienced at my last job. Hackers thought they were a good value, too. (Was 5-ish years ago, not sure if that’s the same case now.)


I mean... yes? Cheap servers are cheap servers no matter who's buying them or why


It's the same, there's no "parent company", Online rebranded as Scaleway. Just a 75% increase ? You've been lucky. They also raised the prices two times since then on the Scaleway IaaS VMs.

Stay away from them, I don't see how they could be trusted with pricing again


Oh, I use to think Online.net was the parent brand. Thanks for clarify. Didn't knew was a full rebranding. My impression was that was sub-brand something like soyoustart to OVH, in that case, I will keep away from scaleway too.


It used to be the case. Online was doing dedicated servers/housing, Scaleway was IaaS. Now they went full "Scaleway", with Scaleway Dedibox being the dedicated part, and Scaleway Elements being IaaS!


Online rebranded as Scaleway, but there is a parent company called Iliad which is also the parent company of Free.


Yes, but the parent was talking about Online (“I was an online.net customer”)


Yeah, something similar happened to me @ Scaleway (so, the part before the rebrand/merger/whatever they did). Incredible price increase for my usecase. Even with the limited 25% discount, it grew out of hand. I'm in the process of migrating to Hetzner, but unfortunately some stuff is IP tied and it takes some time to fix everything.

Even before that, the constant changing of plans for Cloud was anoying. One of the reasons I chose Scaleway was because they promised easy upgrade/increase of resources as I needed it on the VPS. But then they discontinued the line I was using and I had to create a completely new one anyway. Then I had to move that one to the higher priced models again. A lot of fuss for something I would just like to keep running as long as possible with as little involvment.


I am in the same boat. they increases the price out of the blue and their instance availability for scaling up or down is not great.(M1 is out of stock now)


I'm a former Scaleway employee so take this as a informed guess: DC4 space is divided in very small, secure rooms (which you can see in the promotional video) with space for like less than a dozen racks at a time.

They probably filled a room with mac minis and will need to fill another room/racks to provide more hw to customers.


I've had pretty terrible experiences with Online/Scaleway customer support and them just pulling the plug on services with no transition path (like C14). I've decided that I'd rather just pay Amazon or DigitalOcean more and not have to deal with Scaleway.


I recently set up one of their "stardust" instance, which is the absolute cheapest one I found at ~2€/mo. Works as intended , not much more to say about it.


Sounds great, but I will not go for this the second time, Scaleway.

Two years ago I purchased a server there, configured everything and started handling workload. Then I received the following email:

Our support team created a new ticket associated to your account. Hello, Your instance 'REDACTED' is running on a hypervisor that encountered a critical failure. We are not able to power on the hypervisor again. We were not able to recover your local files located on your LSSD. Your node has been stopped. If you created snapshots of the server's volumes or if you halted your node recently, you will recover your volumes at their latest good state. We are sorry for the inconvenience. Scaleway Team

¯\_(ツ)_/¯


Seriously problems happens, and you should not held Scaleway responsible for THIS.

However-- you can definitely hold Scaleway responsible for the crazy price hike (that happened two or three times), very bad support (it improved a bit lately).


For the ones downvoting me, please say why?


I didn't downvote but IMO it is because disagreement is often shown with votes instead of comments. Ability to downvote is a broken system, but saying that will also give downvotes, especially if you also point out a big site that have tried the up/downvote system already and shown have broken it really is.


I just stated facts tho. They did have raised their pricing multiple times. It could be argued on the support quality, ok, but a comment with details is always better for this...

What's also intringuing me is that my comment was at 5 or 6 points. And then in a couple of minutes went directly to -1.


If it were on Reddit I'd guess vote manipulation but on HN? Strange.

I can't see anything wrong with your comment either.


A bit late to follow-up, but by the time you commented, my comment was back from -3 to 6 and now 9.

So who nows? But it's not the first time I raised my issues with Scaleway price rises, and every time they get downvoted quickly. So that makes me think, yes.


You may want to know that this also happens in more expensive providers like AWS.

As a counterpoint, in my experience of a bunch of years with Scaleway they have always worked quite well. I left them due to the price getting out of hand.


What did you move to?


I can recommend Hetzner. For non-production stuff I highly recommend checking lowendtalk.com / lowendbox.com to see offers from different providers.


In cases like this, I would like them to offer a paid option to take the hardware to a data recovery specialist.

For example, the email could say:

"As the plan your instance was on did not include redundant storage, we regret to inform you we have no plans to recover the data on it. We will hold the server hardware for 7 days, and forward it to one of our approved data recovery experts on request. We can also send the data recovery expert encryption keys used for your account that they will need to recover data. There is a charge of $X for this service".


I really liked this idea, until I realized that the net effect would be a post here: "their inferior hardware lost my data and then they tried to extort 5000EUR to get it back."

While on paper it seems crazy that offering a customer an extra option could never be a negative, in practice it can.


I had a similar issue (two years ago as well!) and support was pretty disappointing. It was a self hosted email server, all content just vanished.

The only alternative they gave to me was to upgrade to another type of machine, as the previous one was not available anymore. It was noticeably more expensive of course.

So in the end they had a failure and I ended up having to pay more (I found another solution at another provider)


Problems happen. Did you contact support and ask for a refund? How did they respond?

If the PSU failed, there's not a lot they can do to prevent a critical failure. (Sounds like the drive failed in this case.)


> Did you contact support and ask for a refund? How did they respond?

How this was resolved is the most important part in my opinion


>If the PSU failed

The PSU? As in one (1)? Because then, to quote Adam Savage, Well there's your problem!


Yeah that caught my eye as well. My home media server is a repurposed Dell tower server with redundant PSUs, if I can swing that for non-critical movies and TV shows, surely a large scale cloud provider will have not only redundant PSUs on each node, but their storage will be redundant and rebuildable as well? Else what are their customers actually paying for?


The way I see it, Scaleway vs. AWS is the old tradeoff: enterprise-grade hardware vs. commodity hardware with resilient software.

You can set up one AWS EC2 backed by an RDS.

Or you can set up two Scaleway servers in failover, backed by three CockroachDB Scaleway nodes.

In the second approach, you need more cheap things, but the total price can still be lower. On the other hand, you should not expect the same reliability from a single Scaleway node, compared to a single EC2.


EC2 isn't enterprise hardware, it's all commodity and, more lately, custom-designed silicon. It certainly fails, and Amazon's own recommendations call for redundancy using things like EC2 auto-scaling groups and RDS multi-AZ.

EC2 has a slight leg up because EBS stores data redundantly, so a single disk failure won't knock out your data. But failure is still possible, and a durable storage strategy should involve backups.


I left in 2018 when they increased all their prices 60% (€16/mo boxes turning €24/mo). I wouldn't rely on them for anything critical, but €2.40 for 24h is unbeaten right now. I have some software I need to build for Apple M1s and as far as I can see it's this or MacStadium (or buying a device).


SSD $/GB/mo was almost half what it is on AWS last time I checked. Run a clustered replicated storage system or ephemeral workload and you are fine.


What's the problem exactly? hardware dies.


I was using Scalway's ARM (Cavium) 2core/2GB instance for a web application (HTML+Minimal Vanilla JS+Go+PSQL) until they suddenly pulled the plug on their ARM instances. I didn't like the way they did it and so I migrated to AWS x86 T2.micro 1core/1GB, but surprisingly the application seems faster here although I think Scaleway had greater network bandwidth.

It may be that I didn't use ARM optimized version of PostgreSQL or some other variable I'm missing. But anyways Apple M1 could be more performant than any of the low-mid x86 offering considering the work is optimized for ARM.


I recieved plenty of emails in good notice and saw warnings throughout the console about the disconinuation of arm. moving away from arm was easy too.


Received the first email in April telling, technical support would end in July and servers pulled out by December. No detail or technical documentation on why they had to do it and just pointed their customers to the migration document.

What about those customers who had spent extensive resources on optimising their work for ARM, when they forced a x86 migration?

A cloud provider can do better than just 3 months notice. Naturally it was not just me who was unhappy[1].

https://news.ycombinator.com/item?id=22866216


I’ve been a customer for years and enjoy having access to their team via slack.

I recommend giving them a spin— nice to have players other than the big three cloud providers.


Came here to say much the same thing. They're a joy to deal with on Slack.


Hetzner have also just announced 5950X servers at €99/mo ($120 USD/mo), bringing Zen 3 to the data center early.

The raw power you can get these days for peanuts/mo if you go for bare metal is impressive.


The sheer amount of know-how required to not shoot yourself in the foot is intimidating to me at least - i.e. I am much more comfortable working on a compiler than setting up a server in a way that it doesn't get pwned

I want to get some kind of CI going that can spot cache miss hotspots, and other things like that (ideally implementing Intel TMA) - the price is reasonable, the software is hard but solved, but I literally have no idea how to run (or better, how not to run) a bare metal server safely.


I'm in no way a person who run some cluster on bare metal servers, but managing them are not any different than cloud compute instance. Only extra thing you need is to have some monitoring for SMART status of your drive.

> I am much more comfortable working on a compiler than setting up a server in a way that it doesn't get pwned

99% of the time all you need is to have firewall up, password login off and unattended upgrades enabled. It's will literally never get pwned then.


I add fail2ban as good measure too.

I don't know about literally, but it helps.


It's can certainly be useful if you use anything with password auth or just want to avoid logs full of bots, but otherwise just change default SSH port.


If you only allow incomming traffic to sshd, and require ssh-key login (no password login), fail2ban will likely only add complexity, not security.


Hosting for most companies should be a commodity.

I've seen entire businesses supporting double digit employees run off single $20/mo DO droplets, with just the builtin backup/snapshot capability used for redundancy. One of these had 50,000 concurrent realtime API users on such a droplet.

Ultimately what is most important is knowing your needs and how quickly you can recover from disaster, not how much you spend on infrastructure.

Simulating the "Ok we came in to the office this morning and Digital Ocean no longer existed" scenario is more valuable to most small businesses than just throwing money at over-engineered AWS solutions.


People 10-20 years ago could do it just fine, I'm sure you can, too :)


On AWS for the same price you get nothing but crap.


You get what you pay for. Optimize your architecture and you can do r6gd.4xlarge at $0.3174/hour or ~189€/month.

I don’t understand comparing AWS to Hetzner. If you want just bare-metal there are much better clouds.


The thing is though for many workloads (I'd even say most) applications get I/O constrained way quicker than CPU, so scaling CPU will not give you the boost that you expect on AWS and the other big cloud providers.

Keep in mind that PCIe4 NVMes can do 5-7GB/sec (gigabytes - not gigabits) read and write, with up to 500K IOPS.

Compare that to RDS on AWS which charges a comical $0.10 per IOPS-month, so just for IO it would be $50k/month (and for many DBs IOPS are more important than CPU) to meet the performance of one NVMe SSD, which can be bought retail for probably $200.

I don't think I've ever seen such ridiculous pricing. I do understand AWS can be more expensive, but it is just outrageous margin for IO performance (and external bandwidth, but that's a whole other story!).


You can't really compare an ARM AWS vCPU to a bare-metal Ryzen 9 core though.

And the AWS offer is missing storage and bandwidth, which are extremely expensive if you want something comparable.


That's why I choose the `d` EC2. They come with NVMe SSD ephemeral storage and in terms of bandwidth, AWS will beat it anyway. R6 is around 10Gbps if I remember correctly.


Hetzner gives you 20TB/month outgoing bw + 1euro/TB if you add 39 euro for 10gbps uplink - or un-metered 1gbps bw.

Even if you have "up to 10 gbps" bw from aws - it's going to be much more expensive! And if you just need low latency (and "no" data), 1 gbps would probably serve you just as well?

So at those prices (essentially no bandwidth) - you could get two physical servers from hetzner...

Fwiw i agree comparing aws to dedicated doesn't make much sense - but unless flexibility is important - I don't see aws comming out on top from such a simple comparison.


With 8TB of not throttled IO ?


How are you going to throttle local storage on bare-metal exactly? Ephemeral isn't over network, it's directly attached. So yes, even 8TB.


Yes the local storage is not throttled on bare-metal. Maybe these new AWS instance with ephemeral storage are not throttled as well, I can't find the information.

Anyway, it's not 8TB but 1TB and on the AWS price calculator it costs 650€ for this instance. You may have access to special prices.


> I don’t understand comparing AWS to Hetzner. If you want just bare-metal there are much better clouds.

Share, please? I'd love to get off of Hetzner, but their prices absolutely destroy everything else


That there is a product coup. Impressive.

They've still got all the good qualities of a startup but in most meaningful senses they now seem to have the scale and sophistication of offering their global competitors have. Today I'm on the cusp of launching something on there. I used them first for the cheap ARM instances (3eur a month or so) and now for their Kapsule offering. I avoided Amazon, Microsoft and Google out of a sense of moral indignation about their tax affairs.

Declaration of interest: I've been using Scaleway for a few years now and really, really like them.


For what it's worth, I've been using Scaleway for a few years and have had a really great experience.

I would absolutely recommend them.


A product landingpage without any price information is imo a No-Go.


That's exactly how I feel, which is why I was pleasantly surprised to see "0.10€/hour M1 Mac Mini" right there on the front page.


That isn’t shown on mobile. Seems like poor design.


It's also part of an svg. Is it accessible to screen readers?


Aren't SVG text? I imagine they can be made accessible.


Don't plan on this price for too long tho. They are notorious for raising prices and then raising them again and again.


What’s the minimum period?


It's 24h, as requested by Apple's EULA


A static product landing page that shows a progress bar as it's loading is a No-Go as well


Funny, because the price is in the top most hero section on my computer. €0.10/hour


The price is in the bubble near the macbook image. 0.10€/h


So what's the main value of this? How people are using it? Is it used like a remote DEV machine that you use through remote desktop/SSH? Is it used to host software developed for Mac? Some use case where you need very high bandwidth for your app?

It looks cool but I'm having hard time to justify the reason for the existence of the service probably because I never used "Mac on the cloud" and I don't know what I am mission out. Can anyone enlighten me here?


I assume for development and testing of your Mac app. E.g. if you are not primarily a Mac user and just want to port your Windows app.


The price feels high for that, yes? Hardward is relatively inexpensive and falling.

Put another way, with this "lease" after a year you could have bought and owned the vehicle. Instead, you're committed to another year.

That said, I assume there are security advantages to this model. Better control of access, etc.


Primarily for CI/CD servers for mobile apps.


this got me thinking. instead of buying a new mac every 2 years (about 600$+ a year + transition labor I guess), I would rather just use my current mac, and stream into a VM with the latest mac running on really good hardware and use that as a saas. is this a thing?


Not possible really. MacOS has terrible remote desktop software which is really laggy and generally crap even on gigabit FTTH.

They all seem to be based off VNC, instead of something closer to Windows RDP which seems to be a lot smarter in how it sends it over (I cannot notice I'm on RDP most of the time). It also has loads more features (for example, it can change resolution dynamically as you go into full screen instead of being static), handles audio well, etc.

I could be wrong on this - would be interested to hear if anyone has had a better experience.


I wouldn't agree. I often connect to my GF's Macbook to help her out with a code or something like that. You can't watch a video but it is phenomenal to work together on the same thing because it's very easy to connect and has a phenomenal audio link that it feels like you are in the same room. There's almost nothing to mess up, unlike the 3rd party conferencing and desktop sharing software that every time someone is not getting the sound/can't share the desktop/can't unmute the mic.

BTW where I live currently, 4Mbps upstream is considered good and the remote desktop works fine on this connection.


This is what I mean though. You can't watch a video. This works flawlessly on RDP on Windows (will be a little compressed if you are on lower bandwidth). It's a fairly major problem if you're using it as your primary machine.

Have you tried RDP? It is miles ahead of anything else I've seen.


I have not but I guess if I want to watch a video, why not stream it or download it on my actual machine?


What if you use the system as your only computer, using other devices as thin clients? Windows/Linux has tons of different ways to optimize for this (RDP with acceleration, Citrix, LTSP/X forwarding), whereas the Mac doesn’t. Which is fair, Apple didn’t design the OS to be used that way, but it’s an annoying drawback.


I assume the difference will be more prominent if you go into other applications such as cloud gaming.

As an addendum, I ran Cyberpunk on my beaten down 4 year old ThinkPad 420 via an EC2 instance (specifically to test out the hypothesis) and it was faster than my usual usage on it.


Dumb question - on assorted Linux or BSD OSs you can run xrdp against a VNC server. Would that work on a mac?


I don't think so, but I could be wrong. RDP on windows is accelerated (uses the clients graphics card) which makes it much more faster than VNC. VNC just basically sends a load of images of the screen in quick succession. It doesn't know really anything about the contents of them.


I think technically speaking Mac Stadium does that. Pretty pricy vs just buying and restoring from time machine. It’s arguable that the needed Mac gains to be worth it are probably more like every 3-4 years and you are throwing most of that away in Remoting in. The M1 is an inflection point too, I would just argue to buy a new Mac mini each time they update it (once every 4 years or so).


Nice to see them on hackernews, I've been using their service for the last 2 months no issue so far.

They even have an issue tracker answered by humans for all my inquires.

Intuitive UI! Hope they keep growing, it'd be nice if they had some cluster outside europe as well


72 EUR/month. At that price, why not keep them in house ? It would be cheaper.


You want dusty Mac Mini laying under desk somewhere which if fails will put your entire operation on pause for couple of days?

There might be valid business uses for that, but for a company that invested a huge amount of money in a team of developers and then pays them through the nose for every hour whether they work or not and maybe have clients that count on updates, that is simply not acceptable.

Also many companies choose to have their entire infra in the cloud and then it doesn't make sense to spend effort to integrate one physical machine.


The failure rate of single machine setups is way overrated (also, you could have several machines of course).

We need decentralization, smaller providers like Hetzner and also in-house IT.


It is not failure rate, it is the cost if it happens.

I work for large corporations and the calculations are entirely different from what a small, cash strapped startup would be doing.

For example, if you work on a new product and it would take 1 day to get Mac Mini for free vs you would be paying for cloud solution 1000 dollars daily but get it one day sooner, it might still be well worthwhile to get cloud version.

Because if you think about it, it is possible that Mac Mini would be holding up your product development for one day, then you calculate how much you hope to earn from that product daily and how much you are loosing while you haven't launched and then it is no brainer.

As I said, it depends on business case. For some, the cost of this M1 VM is not going to be any factor in the decision.

Have you seen how expensive GPUs are in the cloud? Have you seen how many clients are that pay for huge farms of these?


At that price it takes about a year to pay for the hardware. Factoring in the cost of hosting the hardware it seems reasonable (macstadium charges 30-70 EUR a month for colocation alone).

It's still not cheap if you can ignore the hosting cost and need it for several years but it's way cheaper than using AWS for intel mac minis.


You can probably stop renting on the weekend and if you have an app with infrequent updates you don't need to pay every day. With the right workflow it is definitively cheaper than buying.


AWS charges $26 per day. Someone buys those.


What you pay for in AWS is access to the entire ecosystem, and cheap internal bandwidth to the rest of the AWS customers.

Scaleway offers neither. Their competition is Linode and Digital Ocean.

But at 72eur/month for M1, you could buy it in a year so it’s actually a reasonable price. Unlike AWS if all you need is compute.


> cheap internal bandwidth

This is vendor lockin that AWS put you in and now you mention like it's a pro.


It’s network effect, not vendor lock-in. If you are doing B2B, chances are good your peers are also in AWS, regardless of the lock-in effects of other Amazon services (which are real, and like any lock-in are always negative).

I am old enough to have collocates at Exodus in 2000 for the exact same reason: everyone I needed to work with collocates with them. (The later went bankrupt and became Savvis, no idea what they are called today or if they still exist)


Even if 90% of your peers are in Amazon, the egress fees on the other 10% cost more than sending all of your data over a competitively priced network connection.

And that's the best case, where all your connections to them start and end in the same availability zone. Just inter-AZ fees cost more than you'd normally pay for transit.

It's almost entirely lock-in.


network effect, literally.


GitHub Actions charge $0.08 per minute ($115 per day).


Scaleway have an API that could realistically be driven through a CI process. Assuming the idea is to do automated testing, and to destroy instances when done, the cost is likely to be pretty low.


Apple does not allow renting Macs less than 24 hours.


The page specifies the price is hourly and doesn't seem to list that restriction. Could you please link to where it is specified?

Edit: I don't think you're right. I've checked the Scaleway console and it appears the minimum unit of billing for this is one hour. I'd check, but they're out of stock.

Second edit: I missed the part where it says 24 hour minimum period on the instance setup page:

"As required by Apple License, you must keep this instance at least 24 hours. You will be able to delete it only after 24 hours."


Here[1] is the license for BigSur. The relevant section is 3.A.(ii). This clause was added for BigSur so I _think_ that if you're leasing a VPS with macOS Catalina or older you can legally rent it for less than 24 hours.

[1] https://www.apple.com/legal/sla/docs/macOSBigSur.pdf

Edit: It seems that before BigSur the legallity of subleasing macOS was a "gray area". I guess as long as it doesn't hurt Apple they didn't come after you.


Catalina is not available for M1, though.


Makes me wonder how they draw the line between renting a Mac and providing a service which happens to run on a Mac. I guess I know the answer (arbitrarily).


You could get a rackmount kit and then ship it to a hosting company: https://www.sonnettech.com/product/mac-mini-mounting-solutio...


I'm working for a customer whose IT only really deals with Microsoft. We as iOS developers had an internally set up build server but it's unused right now, because everybody works remote and it's behind the corporate mote. This Scaleway offering would work much better.


Getting in-house IT to host a machine in their(our?) on-site data center would cost my department a lot more than 72eur/month, and that is before I pay for the hardware.


I am interested in hosting my own server, but I am not sure how to go about setting one up. Does anyone know a guide on this topic?


I was evaluating scaleway a couple of years ago. I had issues setting up firewall, some kernel errors were thrown when I tried to reconfigure iptables.

It seemed like such a basic thing that I completely lost trust in them then.


I'm curious about how the hardware side of it all looks. Does Apple sell chips to these cloud providers or they are indeed using real/modified Mac Minis?


Using actual Macs is the only way. Apple does not care about helping cloud providers in any way (e.g. even just an exception to the 24h rental requirement would be huge for them, but not even AWS got that).


I saw a video of a mac only data center and they made customized racks to optimally fit real consumer Macs into those racks.

They had those for the trashcan/cheese-grator macs, mac minis and well as the all-in-one macs for special use cases (yes they put those all in ones in a rack)


I believe it is Mac Minis but I would think they strip them of the case.


The aluminum case is a massive heat sink, no reason to remove it.


If it's already in a cooled data center, removing it could mean being able to fit more Minis in a rack with little downside to thermals. The aluminum case makes sense when you're not in a climate-controlled environment.


They use real retail macs. As for the modifications i don't know.


I am getting an “ Oops, an error occurred!” message.

Update: I opened it with Chrome and it works fine. It’s not a “hug of death” as I initially thought.

It’s on Safari (iPhone 12 Pro iOS 14.4) that it’s always leading to the error above. In case it’s helpful for the website owners if they see this.


Works flawless for me, both before and after you commented.


Firefox Android: it works only if I request the desktop site.


I don't see an error message and can view the page but if a hosting company's website can be "hugged to death" then that doesn't exactly inspire confidence in them.


You’re right, everything besides AWS and GCS sucks. Thanks for warning the rest of us /s

If it’s only failing in one browser it’s obviously not a capacity issue. Works on my iPhone btw.


Just bought a Macbook pro last week, and now I'm wondering if I was a giant moron for steering away from the M1. Just seemed like it was going to cause me a lot of headaches when my target instances are all on x86.


Rosetta is pretty effective, I don't see any significant difference in performance.


I wonder who precicely this is targeted for. The concept is interesting, but doesn't everyone who develops for macos already have a mac? Or is this actually an economic path with respect to cloud computing?


> I wonder who precicely this is targeted for. The concept is interesting, but doesn't everyone who develops for macos already have a mac?

Maybe, but is everyone who develops for OSX going to have an M1 in 2021-2022? I just bought a refurbished Intel-based Macbook Air (for just $50 shy of what a new M1 Air would have cost) due to concerns about virtualization software compatibility. People developing software to run on OSX may not want to invest in M1 machines this early, and small-to-midsized companies who rely on their Mac farms to run test suites etc. may not want to duplicate their infrastructure just to validate their software on M1, when they can perhaps meet their needs by using elastic M1 services.


Our startup relies on server-side HTML rendering in Headless Chrome, and some ImageMagick image manipulation. Due to how optimised these machines are for this use-case, I do plan to evaluate them to replace the much pricier AWS instances.


I don’t see how those things need a Mac


They don’t and currently work on Linux. It just might turn out to be much cheaper to use this one


A dedicated x86 machine with equivalent performance for those two tasks is a lot more expensive.


(if i get it correctly it's mostly cause it's a gpu bound task? (hardware acceleration of the browser engine?)) Back in 2017 hetzner offered gtx1080s for around 70€ a month with 64gb ram and and a 4+ghz quad Intel CPU, n 2x 512gb ssd

Im sure these days it's not hard enough to find smth with equilavent processing power in that price range

Maybe even direct contact with hetzner could help you out these days(they stopped offering them cause of abuse)

And if it's not a gpu bound task but badly concurrently cpu tasks Then maybe a game server specific dedicated hosting (they use consumer CPUs with uo to 5+ghz single core performance) wouldn't also be a lot of money

N if its a good concurrency task then the amd CPU offering of hetzner beats all(high memory, insanely fast iops, lots of core) , and has a offering for 5950x these days


5950x is 117€ / month on Hetzner.

The M1 actually beats it in single core performance, and is quite a bit faster at running JavaScript for that headless browser due to design optimizations.


But they could have a preference for the platform, no?


Have you tried browserless.io?


Thanks, might check it out


I would guess it's for people who need to run automated tests that their Mac M1 software works. Or maybe there's some program that only works on M1 and they need to run it in an automated manner (a compiler?).


Its almost always cheaper to just buy a mac mini than to rent one out 24/7 though. I think Apple also made a tos change banning renting them by the hour as well.


If you're a small operation, maybe. If you're a big company you probably want to have a service, not extra hardware you have to manage, and the rental cost is a rounding error for your software development costs.


Will take a year for the rent to cover the price of a mini, and it includes bandwidth, IP assignment and power. Not a bad deal at all.


I'm doing some work for a company that develops apps on behalf of clients. There are dozens and likely soon, hundreds of apps that sometimes all need to be updated at once. The process is automated using Fastlane but it still takes significant time per app, esp. if new screenshots are needed. We've got this running on a MacStadium-hosted Mac mini. Seems like a perfect use case. Scaleway is a little cheaper but it looks like they don't have any M1s available currently.


A customer of mine runs a static and dynamic vulnerability analysis service for Android and iOS apps (their customers are the companies developing those apps.) They have Mac Minis to perform some processing that Apple wants to be done only on a Mac.


One of the obstacles to docker coming to m1 was being able to set up their test pipeline with m1 instances.

Beyond that sort of thing I am mystified


For Scaleway and Amazon, does Apple ship these Mac Minis in the same packaging as retail, or do they strip it down to something less wasteful at that scale?


from the images that i saw on the news, aws just shove those M1 mini into the rack.


This is amazing!


I signed up to test this and they are already out of capacity. I hope they can keep ordering more quickly...


Interesting.

However I would be way happier if they added openSUSE to their distro offer and present some plan what they are planning to do about centos situation.

So far they only have Fedora and CentOS as RPM distros. I do not consider Fedora a server distro and CentOS 8 is finished. Their support is silent for now...


Can anyone recommend a database that already works on the M1 chips?


Not used it on M1 yet but Postgres seems to: https://info.crunchydata.com/blog/postgresql-benchmarks-appl...


So what sort of virtualisation do they run on M1 macs? QEMU?


none, it's baremetal servers


I clicked this and same time, got an email notification from Scaleway about the exact same thing, what are the odds :)


Interesting to see a Flash-style loading bar make a return for these sorts of animated UI experiences.


No Nitro... Just a Mac mini

this nitro https://aws.amazon.com/ec2/nitro/

( ͡° ͜ʖ ͡°)


Isn't nitro mostly AWS marketing? They profit from it as they burn fewer CPU cycles on virtualization for shared hardware. But on dedicated hardware you own everything anyway, so why bother if AWS virtualization runs there?

The big difference is support for AWS ecosystem (also partly realized via nitro). You won't get that but for that price difference it's probably still ok.


Is Nitro a problem? It's there to stop users destroying the hardware, and has minimal performance impact.




Consider applying for YC's Spring batch! Applications are open till Feb 11.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: