This is as much an indictment of AWS compute as it is anything else.
I benchmarked I4i at ~2GB/s read, so let's say I7i gets 3GB/s. The Verge benchmarked the 256GB Neo at 1.7GB/s read, and I'd expect the 512GB SSD to be faster than that.
Of course, an application specific workload will have its own characteristics, but this has to be a win for a $700 device.
It's hard to find a comparable AWS instance, and any general comparison is meaningless because everybody is looking at different aspects of performance and convenience. The cheapest I* is $125/mo on-demand, $55/mo if you pay for three years up front, $30/mo if you can work with spot instances. i8g.large is 468GB NVMe, 16GB, 2 vCPUs (proper cores on graviton instances, Intel/AMD instance headline numbers include hyperthreading).
> Here's the thing: if you are running Big Data workloads on your laptop every day, you probably shouldn't get the MacBook Neo.
> All that said, if you run DuckDB in the cloud and primarily use your laptop as a client, this is a great device
It’s staggering. Jaw dropping. Bandwidth is even worse, like 10000X markup.
Yet cloud is how we do things. There’s a generation or maybe two now of developers who know nothing but cloud SaaS.
I watched everyone fall for it in real time.
You're either underestimating how big cloud instances can get or overestimating how much it costs to rent a cloud instance that would beat an M1 Max at any multi-core processing.
According to Geekbench, the M1 Max macbook pro has a single-core performance of 2374 and multicore of 12257; AWS's c8i.4xlarge (16 vCPUs) has 2034 and 12807, so relatively equivalent.
That c8i.4xlarge would cost you $246/mo at current spot pricing of $0.3425/hr, which is, what, 20% of the cost of that M1 Max MBP?
As discussed recently in https://news.ycombinator.com/item?id=47291906, Geekbench is underestimating the multi-core performance of very large machines for parallelizable tasks -- the benchmark's performance peaks at around 12x single-core performance. (I might've picked a different benchmark but I couldn't find another benchmark that had results for both the M1 Max and the Xeon Scalable 6 family.)
If your tasks are _not_ like that, then even a mid-range cloud instance like a 64-vCPU c8i.16xlarge (which currently costs $0.95/hour on the spot market) will handily beat the M1 Max, by a factor of about 4. The largest cloud instances from AWS have 896 vCPUs, so I'd expect they'd outperform the M1 Max by about 50-to-1 for trivially parallelizable workloads. Even if you stay away from the exotic instances like the `u7i-12tb.224xlarge` and stick to the standard c/m/r families, the c8i.96xlarge has 384 vCPUs (so at least 24x the compute power of that M1 Max) and costs $3.76/hr.
A 5 month ROI on a hardware investment would be excellent, so not sure what you're trying to say here?
The tooling — K8S with all its YAML, Terraform, Docker, cloud CLI tools, etc. — is pretty hideously ugly and complicated. I watch people struggle to beat it into shape just like they did with sysadmin automation tools like Puppet and Chef a decade or more ago. We have not removed complexity, only moved it.
The auto scaling thing is a half truth. It can do this if you deploy correctly but the zero downtime promise is only true maybe half the time. It also does this at greatly inflated cost.
Today you can scale with bare metal. Nobody except huge companies physically racks anymore. Companies like Hetzner and DataPacket have APIs to bring boxes up. There’s a delay, but you solve that by a bit of over provisioning. Very very few companies have work loads that are so bursty and irregular that they need full limitless up and down scaling. That’s one of those niche problems everyone thinks they have.
The uptime promise is false in my experience. Cloud goes down for cluster upgrades and any myriad other reasons just as often as self managed stuff. I’ve seen serious unplanned outages with cloud too. I don’t have hard numbers but I would definitely wager that if cloud is better for uptime at all it’s not enough of an improvement to justify that gigantic markup.
For what cloud charges I should, as the deploying user, receive five nines without having to think about it ever. It does not deliver that, and it makes me think about it a lot with all the complexity.
The only technical promise it makes good on, and it does do this well, is not losing data. They’ve clearly put more thought into that than any other aspect of the internal architecture. But there’s other ways to not lose data that don’t require you to pay a 10X markup on compute and a 10000X markup on transfer.
I think the real selling point of cloud is blame.
When cloud goes down, it’s not your fault. You can blame the cloud provider.
IT people like it, and it’s usually not their money anyway. Companies like it. They’re paying through the nose for the ability to tell the customer that the outage is Amazon’s fault.
Cloud took over during the ZIRP era anyway when money was infinite. If you have growth raise more. COGS doesn’t matter.
Maybe cloud is ZIRPslop.
If your application won't ever require more resources than a single server or two, then you are better off looking at other alternatives.
If the metal dies in a catastrophic way (multiple nodes at once and loss of quorum, catastrophic DC outage, etc.) you spin it up in AWS.