Assuming you mean 99.9999%; your hyperscaler isn't giving you that. MTBF is comparable.

It's hardware at the end of the day, the VM hypervisor isn't giving you anything on GPU instances because those GPU instances aren't possible to live-migrate. (even normal VMs are really tricky).

In a country with a decent power grid and a UPS (or if you use a colo-provider) you're going to get the same availability guarantee of a machine, maybe even slightly higher because less moving parts.

I think this "cloud is god" mentality betrays the fact that server hardware is actually hugely reliable once it's working; and the cloud model literally depends on this fact. The reliability of cloud is simply the reliability of hardware; they only provided an abstraction on management not on reliability.

I worked for a payments company (think credit cards). We designed the system to maintain very high availability in the payment flow. Multi-region, multi-AZ in AWS. But all other flows such as user registration, customer care or even bill settlement had to stop during that one incident when our main datacenter lost power after a testing switch. The outage lasted for three hours and it happened exactly once in five years.

In that specific case, investing into higher availability by architecting in more redundancy would not be worth it. We had more downtime caused by bad code and not well thought out deployments. But that risk equation will be different for everyone.

Once management is convinced by sales people or consultants any technical argument can be brushed away as not seeing the strategic big picture of managing enterprise infrastructure.

Our industry has really lost sight of reality and the goals we're trying to achieve.

Sufficient scalability, sufficient performance, and as much developer productivity as we can manage given the other two constraints.

That is the goal, not a bunch of cargo-culty complex infra. If you can achieve it with a single machine, fucking do it.

A monolith-ish app, running on e.g. an Epyc with 192 cores and a couple TB of RAM???? Are you kidding me? That is so much computing power, to the point where for a lot of scenarios it can replace giant chunks of complex cloud infrastructure.

And for something approaching a majority of businesses it can probably replace all of it.

(Yes, I know you need at least one other "big honkin server", located elsewhere, for failover. And yes, this doesn't work for all sets of requirements, etc)

From what I hear, Nvidia has an exceptionally high failure rate.

I manage an infrastructure with tens of thousands of VMs and everyone is obsessed with auto scaling and clustering and every other thing the vendor sales dept shoved down their throats while simultaneously failing to realize that they could spend <5% of what we currently do and just use the datacenter cages we _already have_ and a big fat rack of 2S 9754 1U servers.

The kicker? These VMs are never more than 8 cores a piece, and applications never scale to more than 3 or 4 in a set. With sub 40% CPU utilization each. Most arguments against cloud abuse like this get ignored because VPs see Microsoft (Azure in this case) as some holy grail for everything and I frankly don't have it in me to keep fighting application dev teams that don't know anything about server admin.

And that's without getting into absolutely asinine price/perf SaaS offerings like Cosmos DB.

Some of the cloud providers don't even do live-migration. They adhere to the cloud mantra of "oh well, its up to the customer to spin up and carry on elsewhere".

I have it on good authority that some of them don't even take A+B feeds to their DC suites - and then have the chutzpah to shout at the DC provider when their only feed goes down, but that's another story... :)

This is not the first time that "low yield" karma comments have sporadic changes to their votes.

It seems unlikely at the rate of change (roughly 3-5 point changes per hour) that two people would simultaneously (within a minute) have the same desire to flag a comment, so I can only speculate that:

A) Some people's flag is worth -2

B) Some people, passionate about this topic, have multiple accounts

C) There's bots that try to remain undetected by making only small adjustments to the conversation periodically.

I'm aware that some peoples job very strongly depends on the cloud, but nothing I said could be considered off topic or controversial: Cloud for GPU compute relies on hardware reliability just like everything else does. This is fact. Regardless of this, the voting behaviour on my comments such as this are extremely suspicious.

This isn't really true. I mean it's true in the sense that you could get the same reliability on-premise given a couple decades of engineer hours, but the vast majority of on-premise deployments I have seen have significantly lower reliability than clouds and have few plans to build out those capabilities.

E.g. if I exclude public cloud operator employers, I've never worked for a company that could mimick an AZ failover on-prem and I've worked for a couple F500s. As far as I can recall, none of them have even segmented their network beyond the management plane having its own hardware. The rest of the DC network was centralized; I recall one of them in specific because an STP loop screwed up half of it at one point.

Part of paying for the cloud is centralizing the costs of thinking up and implementing platform-level reliability features. Some of those things are enormously expensive and not really practical for smaller economies of scale.

Just one random example is tracking hardware-level points of failure and exposing that to the scheduler. E.g. if a particular datacenter has 4 supplies from mains and each rack is only connected to a single one of those supplies, when I schedule 4 jobs to run there it will try to put each job in a rack with a separate power supply to minimize the impact of losing a mains. Ditto with network, storage, fire suppression, generators, etc, etc, etc.

That kind of thing makes 0 economic sense for an individual company to implement, but it starts to make a lot of sense for a company who does basically nothing other than manage hardware failures.

Yeah we've already had about a day's worth of downtime this year on office 365 and Microsoft is definitely a hyperscaler. So that's 99.3% at best.

For what it's worth, GCP routinely live-migrates customer VMs to schedule hardware for maintenance/decommissioning when hardware sensors start indicating trouble. It's standard everyday basic functionality by now, but only for the vendors who built the feature in from the beginning.

TLDR, VC money, is being burnt/lost

I know that's basically impossible to answer generically, especially given that the recurring cost is likely already zero, given that the GPUs are already paid...

In rural areas or even with low population it takes forever to get fiber to roll out and if your selling access to your hardware infrastructure then you really want to get a direct connection to the nearest IX so you can offer customers the best speed for accessing data and the IX would probably be one of the few places you might be able to get 400G or higher direct fiber. But if your hooking up to a IX chances are your not an end user but a autonomous system and already are shoving moving and signing NDA's to be a peer with other Autonomous Systems in the exchange and be able to bgp announce.

(Source - my old highschool networking class where I got sick of my shitty internet and looked into how I could get fiber from an exchange. I'm probably mistaken on stuff here as it was years ago and its either wrong or outdated from all those years ago.)

In NW Washington state at least, the rural counties (Whatcom, Island, Skagit, etc) have had a robust market in dark fiber for over two decades.

The normal telcos weren't responsive to need, so private carriers picked up the slack. When I was last involved in this market, you could get a P2P strand, including reasonable buildout, for less than a cost of a T1 line with a two-year commit.

The tiny four-branch credit union I worked for had dedicated fiber loops between all our locations, no big deal. It was great.

Surely someone in the trillion dollar datacenter industry can figure out a way to take waste heat and use it in a profitable way, right?

At the highest power settings, H100s consume 400 W. Add another 200 W for CPU/RAM. Assume you have an incredibly inefficient cooling system, so you also need 600 W of cooling.

Google tells me US energy prices average around 17 cents/kWh - even if you don't locate your data centre somewhere with cheap electricity.

17 cents/kWh * 1200 watts * 1 hour is only 20.4 cents/hour.

Factor in space, networking, cooling, security, etc., and $2 really do seem undoable.

Imagine I own a factory, and I've just spent $50k on a widget-making machine. The machine has a useful life of 25,000 widgets.

In addition to the cost of the machine, each widget needs $0.20 of raw materials and operator time. So $5k over the life of the machine - if I choose to run the machine.

But it turns out the widget-making machine was a bad investment. The market price of widgets is now only $2.

If I throw the machine in the trash on day 1 without having produced a single widget, I've spent $50k and earned $0 so I've lost $50k.

If I buy $5k of raw materials and produce 25k widgets which sell for $50k, I've spent $55k and earned $50k so I've lost $5k. It's still a loss, sure, but a much smaller one.

Rackspace. Networking. Physical safety. Physical security. Sales staff. Support staff. Legal. Finance. HR. Support staff for those folks.

That’s just off the top of my head. Sitting down for a couple days at the very least, like a business should, would likely reveal significant depths that $2 won’t cover.

The market doesn't care how much you're losing, it will set a price and it's up to you to take it, or leave it.

Depending on how fast their value depreciates, selling them might recoup more money then renting them away. And being exposed to 3y of various risks.

Selling now at a 40% loss gets you back the equivalent of 60c/h over three years, and without having other costs (DC, power, network, security) and risks.

Well yes, because for GPU datacentres fixed/capital costs make up a much higher fraction than power and other expenses than for CPUs. To such an extent that power usage barely even matters. A $20k that uses 1 kW ( which is way more than it would in reality ) 24x7 would cost $1.3k to run per year at 0.15$ per kWh, that's almost insignificant compared to depreciation.

The premise is that nobody could make any money by renting H100s for 2$ even if they got them for free unless they only had free power. That makes no sense whatsoever when you can get 2x AMD EPYC™ 9454P servers at 2x408 W (for full system) for around $0.70 in a German data center.

There are very few data centers left that can do 45kW+ rack density, which translates to 32 H100/MI300x GPUs in a rack.

Most datacenters, you're looking at 1 or 2 boxes of 8 GPU, a rack. As a result, it isn't just the price of power, it is whatever the data center wants to charge you.

Then you factor in cooling on top of that...

That’s essentially what the OP says. But once you’ve already invested in the H100s you’re still better off renting them out for $2 per hour rather than having them idle at $0 per hour.

For datacentre GPUs the energy, infrastructure and other variable costs seem to be relatively insignificant to fixed capital costs. Nvidia's GPUs are just extremely expensive relative to how much power they use (compared to CPUs).

> H100s you’re still better off renting them out for $2 per hour rather than having them idle at $0 per hour.

If you're barely breaking even at $2 then immediately selling them would seem like the only sensible option (depreciation alone is significantly higher than the cost power of running a H100 24x365 at 100% utilization).

If you can then probably yes. But why would someone else buy them (at the price you want), when they can rent at $2 per hour instead?

I think that's the point. Trying to buy and run H100s now either for yourself or for someone else to rent it is a terrible investment because of oversupply.

And prices you can get for compute are not enough to cover the costs.

If you read the article, such prices happen because a lot of companies bought hardware reservations for the next few years. Instead of keeping the hardware idle (since they pay for it anyway), they rent it out on the cheap to recoup something.

This company TensorWave covered by TechCrunch [0] this week sounds very similar, I almost thought it was the same! Anyway, best of luck, we need more AMD GPU compute.

[0] https://techcrunch.com/2024/10/08/tensorwave-claims-its-amd-...

What do you mean by "risking your business on unreliable compute"? Is there a reason not to use one of these to train whatever neural nets one's business needs?

I think I'll take that risk over paying more for your allegedly more reliable GPUs anytime.

I've said it before and I've said it again....

Read the cloud provider small-print before you go around boasting about how great their SLAs are.

Most of the time they are not worth the paper they are written on.

https://azure.microsoft.com/files/Features/Reliability/Azure...

Launching any multitenant system is HARD. Many of them are held together with bubble gum and good intentions….