One of the issues is availability. Lambdalabs and Paperspace never have instances available. AWS and GCP have the per-account quota set to 0 by default for GPU instances and deny requests to raise it, even to 1. (One of my AWS accounts was granted 1 GPU after some back-and-forth with support.) I'm not upset, I understand that demand is high, but how will oblivus handle availability?

Thank you for your question!

Firstly, unlike AWS, GCP, or Azure, there are no restrictions on the number of GPUs you can deploy with us. As long as you have sufficient account balance, you can deploy hundreds of GPUs simultaneously.

In terms of availability, we maintain a diverse range of resources from various vendors in our on-demand stock. While we have our own infrastructure, we also leverage infrastructure from other vendors to meet the growing demand.

Currently, we have more than 3000 GPUs in stock, with over 2600 of them available for deployment. You can find more detailed information on our availability page at https://oblivus.com/availability/.

I hope this helps!

> you can deploy hundreds of GPUs simultaneously.

> with over 2600 of them available for deployment.

guessing you mean 2600 in total.

FWIW we ran a workload recently on AWS that required a few thousand g4 instances in a single AWS region. We ended up scavenging and using g3s as well due to capacity constraints.

That's quite an impressive workload!

If you're using our on-demand service and intend to terminate the machines once your work is completed, we currently have 2600 available GPUs. However, if you have an ongoing need for these machines, we also have reserved instances with additional stock, which brings our total capacity to an estimated 7000 GPUs as of today.

But of course these numbers could easily change in the future.

I think they mean that almost 2600 are literally available right now for you to rent - or if you requested them, then you would get them right now, and then there would - for a while - be 0 that are available, because you have them all.

Out of interest, was this workload for training or serving?

Lots of people seem to think they have no choice but to rent GPUs from cloud service providers.

Remember retail GPUs are much more powerful, much cheaper and much more available than cloud GPUs.

Just go to the store and buy one.

Then I'd only have one. Instead, I can borrow a dozen of them and use them in parallel and I'll return them when I'm done. I don't know how many I'll need, or for how long, and cloud enables my bad habit of not planning.

> retail GPUs are much more powerful

I agree with your basic point about building vs renting, but retail GPUs aren't simply more powerful. Retail GPUs are just as fast but have a lot less VRAM and memory bandwidth. The RTX4090 has 24 GB of VRAM with 1 TB/s of bandwidth. The A100 has 80 GB at 2 TB/s. For some tasks, the memory and bandwidth are the strongest constraints.

AI being a primary example, and that's probably the thing most people are renting these for nowadays anyways.

Thank you for sharing your thoughts.

Acquiring the latest technology, such as an A100 GPU, individually from a store can indeed be quite expensive, with prices around $10,000 per unit. Additionally, setting up and maintaining a home lab to scale from one GPU to thousands can be a significant investment in terms of both cost and resources.

In contrast, our platform provides access to the latest GPU technology without the need for users to individually purchase and manage the hardware. We offer the scalability to deploy multiple GPUs as needed and scale back down when required, making it a more cost-effective and flexible solution compared to setting up and maintaining a personal lab.

Furthermore, it's important to note that our GPUs are directly dedicated to each virtual machine, ensuring that the power and performance are not compromised by sharing resources. This ensures that our GPUs provide the full capability and performance expected, making them just as powerful as individual units.

I have just bought 4 servers with 10 GPU (Titan X Pascal) and 384 GB of RAM for the price of a macbook. I found a solar powered CoLo facility to put them in. The economics of owning vs renting from AWS are worth it. My server set up is the same price as running one model on AWS.

If memory serves, NVidia's drivers can only be used for Tesla / Ampere family, if you are using them in a datacenter. Titan / GeForce are not allowed. So, there's that (but maybe I'm missing something).

Now, "running" may also mean different things. For example, you may want to do things s.a. performance diagnostics (in order to understand if your code uses resources efficiently), and then you'd need stuff like NVML, DCGM and co. Consumer-grade hardware might not be supported, or might not in principle support diagnostics collection / instrumentation. Or, if you have multiple GPU-dependent workloads that cannot saturate your resources -- you might think of MIG as being a way to address that... and, again, consumer-grade GPUs won't help you here...

I'm not saying you shouldn't try self-hosting. I'm actually all for it. But, you also need to be mindful of the pros and cons. NVidia must have some reason to pitch the datacenter family of GPUs to, well, datacenters. They aren't just blowing up prices.

Also, to give some sense of comparison:

Titan X ~= 3.5K CUDA cores.

V100 ~= 5K CUDA cores.

A100 ~= 7K CUDA cores.

H100 ~= 18.5K CUDA cores.

It probably doesn't translate directly into H100 being six times as fast as Titan X, but, I hear that these GPU workloads might be lengthy...

I bought these from a Chinese internet firm. They dismantled their data center and I am buying it.

Also, Datacenter GPUs only have passive cooling, presumably to allow for more Cuda cores. I get that what I have is older tech but for the cost of one H100 GPU I can have 60 of these servers (600 GPUs) plus some change to pay for CoLo fees.

I'm not trying to discourage, and I don't have concrete numbers on hand, but there are other factors, beside the price of h/w. Like, obviously, electricity use, data locality / moving it around (with more smaller units you'd have to move it more, also, not sure if consumer-grade GPUs support NVLink, and even if they do, then at what bandwidth?)

For some of these, you could obviously pay with your time. Sometimes that time is very valuable, and sometimes you have a lot to spare.

Also, the amount of VRAM (3x)... Sometimes having too little of it means having to re-write the program, or it could dramatically impact the speed. Similarly for bandwidth (8x). But, again, if the kind of workload you have isn't constrained by either, then you could probably win by running on more smaller / older GPUs.

NVLink was discontinued with 40 series nvidia consumer GPUs.

I estimate $1.5k of electricity per GPU for a 5y lifetime assuming 100% load. Not including other components.

Uhh… can you share details… this seems like price point of things when they’ve fallen off the back of a truck.

Can you please share where to buy this from? I would like to purchase one. Thanks!

In the tests I’ve done, retail GPUs are much faster than data center GPUs.

They are, but data center GPU (ampere and above) are twice as good in performance per watt.

Aside from being green and that, why would a consumer care about a leased processor's performance per watt?

The retail price of renting the GPU includes, among other things, the cost of the power bill that the cloud service pays (or else they'd lose money.) The more efficient GPU will, all else the same, be cheaper to lease.

Given a relative amount of benefit, with a ratio of Watts/Benefit, a lower ratio of Watts/Benefit should cost less to the consumer.

So the consumer would care because the service is lower cost.

I'm truly confused by your argument and @pjlegato's ...

It's more expensive to rent a GPU than to buy it.

This whole comment thread started because GP wrote:

>Remember retail GPUs are much more powerful, much cheaper and much more available than cloud GPUs.

So, still, why would a consumer care about performance per watt?

Because you'll have to include its price when you buy it, but not when you rent it. And how the economics plays out then is not obvious without maths.

I can assure you it's cheaper to buy. Otherwise they wouldn't be leasing it.

(ofc assuming you're not going to use it for like a day)

Lower wattage allows more GPUs per system. For distributed training, that’s allows 8 GPUs per system which would otherwise be very difficult with retail GPUs.

And GPU-GPU interconnects are important for this type of training so putting as many GPUs as possible close together is necessary.

We use the same graphic buses, such as SXM4 and PCI Express 4.0 x 16, as you would find in your system. Theoretically, there should be no difference in performance or compatibility at all.

If you have conducted your tests on shared instances and have observed such differences, then it's understandable. However, I want to emphasize once again that we provide dedicated virtual machines, which means that the resources are exclusively allocated to each user and not shared.

> Theoretically, there should be no difference in performance or compatibility at all.

At least in the past there was a tendency by NVIDIA to run their professional GPUs at a lower frequency, offering reliability and correctness guarantees as tradeoff. Meanwhile most of the retail/gaming GPUs came overclocked and almost outright guaranteeing that they would start to crash and fail if you tried to run them 24/7.

Even so, an A100 is not actually faster than an RTX 4090, right? So basically the reason you would want to use something like this instead is because you need the VRAM?

Yes, 0x008 makes a valid point regarding this topic. Data center GPUs and RTX Series use different technologies and optimizations for different purposes.

Our intention is not to discourage anyone from purchasing their own GPUs. We offer our GPU cloud services as an alternative for those who may not have the resources or prefer the convenience and flexibility of renting GPUs on-demand. We apologize if the discussion took a different direction, and we appreciate your understanding.

That’s not the case. It is significantly faster, at least for deep learning workloads. Check out Lambda Lab’s benchmarks.

https://lambdalabs.com/gpu-benchmarks for those who are curious, ends up being quite the difference no matter which way you slice it, up to 60% better. Of course, this is probably due to the increased interconnect bandwidth and memory rather than raw compute horsepower, but for the workloads in question that's relevant.

Would you be so kind to post these results? How much faster, which models etc.

I found some charts here showing A100 roughly doubling a 3090: https://bizon-tech.com/blog/best-gpu-for-deep-learning-rtx-3...

With the 4090 being a fair amount more powerful, it might give the a100 a challenge for about 10% of the cost. That’s still about 10k hours of renting one with this service though.

Sure if you are using tiny net(by today's standard) and small batch sizes, A100 is just twice as fast compared to 3090 due to overhead of framework and memory movement. A100 has 312 FP16 TFlops[1], compared to 40 and 82.6 of 3090 Ti and 4090 respectively[2], and has 3x memory. Also it has 10x inter GPU communication bandwidth.

[1]: https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Cent...

[2]: https://images.nvidia.com/aem-dam/Solutions/Data-Center/l4/n...

I believe you are wrong. The relevant metric is tensor flops and 4090 has 330. I believe 312 you quoted for A100 is tensor flops.

As someone whose native language is not English, I find it helpful to use AI to grammatically correct and improve my texts. There's no shame in asking for help or using tools to improve your language skills at all.

Same here and don't interpret it as an accusation, more of an "huh, that's interesting", mostly cause I had the suspicion that this was partly written by ChatGPT and was able to detect it, I'm not entirely sure what made me realize it.

Retail GPUs do not have the maxed out VRAM that is desirable for machine learning. And it's cheaper to run them, but it's not free.

Presuming you don't want an A100 for just a few hours.

I agree. Even just a bunch of RTX 3060s go a long way[1]. Also, in the case of my startup we can use our local storage vs uploading dozens of terabytes of data to the cloud and deal with data privacy issues.

[1] https://timdettmers.com/2023/01/16/which-gpu-for-deep-learni...

This is the actual answer. Why rent when you can own. The beefy GPUs of today will still carry their weight years into the future. If you take it even remotely seriously you'll save money and have resale value

Because you’ll probably be only be using it for a small fraction of the time you have it, so sharing can be much more cost effective depending on how much you use it

I dropped a couple hundred bucks on a GPU years ago. It's done a lot of work for me and has years to go. True if you don't plan on using it no point in buying it. Same goes for literally anything

True. As with most rental models, there is a usage threshold from wich on it’s more cost effective.

> Why rent when you can own.

This is a very silly question... the answer is the same as with virtually anything else: real estate, transportation, wedding dresses...

I mean, you need to compare the price that will ultimately depend on the kind of thing you are doing... that's all. Sometimes it makes sense to rent, other times it makes sense to buy. There's no one right answer.

Still pretty expensive IMO.

I'm building a GPU accelerated signal processing web-app, and I'm currently planning to deploy it on a $180 Nvidia Jetson. Performances are a little bit low, but mostly enough to make the app usable.

I tried to find GPU instances for < $50/month, but couldn't find any. The only alternative would have been to rent Apple M1 instances at Scaleway, but it's still way more expensive (€80/month) than hosting my 10 watts Jetson at home.

Does anybody know cheaper GPU instances? I don't need much computing power, about 0.5Tflops to 1TFlops

Lambda Cloud has A100 40GB for $1.10/hr—less than half the price of OP). We also just launched H100s for $2.40/hr.

https://lambdalabs.com/service/gpu-cloud

Engaging in self-promotion without taking the time to understand the context is not helpful. Moreover, the estimated cost of approximately $800 per month would still apply, which may not be within the desired budget.

How dare someone self-promote in the midst of your self-promotion?

I am answering to the version before you edited it.

You are absolutely right, and I apologize if I came across as dismissive. We may have different viewpoints, and it's important to respect and acknowledge each other's perspectives. I appreciate your input, and I'm here to listen and address any further questions or concerns.

Let's keep the conversation friendly and open.

what do you mean with "the estimated cost of approximately $800 per month"?

Parent is saying $1.10/hr = $792/mo of GPU leasing. GGP is looking for $50/mo. Obviously $792 is a lot higher than the GGP's request for $50/mo.

Whether or not $50/mo is an unreasonable ask (it is), they have a budget and feel that even $85/mo is too far out of budget for their application.

> what do you mean with "the estimated cost of approximately $800 per month"?

Presumably they mean

$1.1/hr * 24hr/day * 30day/month = $792/month

> We also just launched H100s for $2.40/hr.

What's the availability like? I had tried it once and had problem getting any GPU.

We have lots of H100s available right now.

Thanks! Is there a status page like https://oblivus.com/availability/ for Lambda Cloud? I got very excited about building a product on top of Lambda cloud for cocalc.com, but after read lots of docs, when I tried Lambda cloud, I just got the message: "We are currently out of capacity for all instances. Please check again in a few hours." This was on April 27. I immediately thought: "There is no possible way I can build a product on this for my users, but it may be very useful for other tasks, e.g., training." If there were something like https://oblivus.com/availability/, especially with historical data, it would be very useful to appropriately set my expectations about how Lambda Cloud can best be used.

By the way, Lambda Stack is VERY impressive. Thanks for maintaining that!

Lambda availability is awful.

On vast.ai I rent 3090s for $0.12 an hour. Nothing comes close in price.

They operate a model known as a "marketplace" or "community" cloud. Unlike us, they utilize servers hosted by individuals in their homes. This is something completely different.

While they may offer lower prices, the infrastructure may have limitations in terms of reliability, scalability, and security compared to data center-based providers. Additionally, the performance and availability of the services vary depending on the hosts' hardware and internet connection.

No, it's not just individuals, and not just at home.

Biggest problem with vast.ai is data security. It is running on somebody's PC. Anybody can get hold of your data.

Thank you for your feedback!

If you're working with Jetson, the RTX 4000 may be more powerful than you need right now, and it's the least expensive we've got. It's worth exploring if there are cloud service providers that offer lower-end GPUs, as that could potentially be a better fit for your requirements and budget.

Less than 50$ will be really hard, at least in any form of professional setup (so not hosted in a random basement ;) ).

Our lowest at Genesis Cloud at this time are instances with an RTX 3060Ti for 0.20$/hour which adds up to 146$/month ( https://www.genesiscloud.com/pricing#nvidia3060ti ) Though, this includes free storage, no egress fees and has a lot more power than a Jetson.

If you need to optimize for low cost hosting, did you already check whether you actually must have a GPU for your use case? Modern CPU have some impressing capabilities.

Here we have another instance of self-promotion that does not align with what GGP mentioned. Thank you, Genesis and Lambda, for promoting yourselves in a startup thread. Given your long-standing presence in this industry, I would expect better engagement from you.

You were doing so well, self promoting and engaging with the community on your post earlier. I didn't expect to see you stoop to this level of commenting.

Maybe it's time to step away from the keyboard for a while?

I appreciate and respect every user who contributes by asking questions, providing feedback, or sharing suggestions. However, it is disappointing and unreasonable to witness self-promotion from companies that have been established in this industry for a considerable period of time under a startup thread.

Moreover, the fact that their self-promotion does not align with the intention of the original discussion and GGP explains their purpose. Their primary goal is not genuinely assisting or finding a solution.

In such cases, as you can imagine, it's challenging for me to maintain respect.

If you're finding it too challenging, might I reiterate my suggestion to take a break from the keyboard? it's not a good look.

You make a valid point, and I appreciate your suggestion.

I apologize if my previous comment came across as dismissive. I believe I have expressed my viewpoint clearly, but I'm open to further discussion.

Let's continue this conversation in a friendly and respectful manner, but after I come back from my break as you have suggested. :)

> 9. No quotas or complex verification processes. Whether you represent a company, an institution, or you're a researcher, you have full access to our infrastructure without any limitations.

Given how easy it is to shoot oneself in the foot with stolen credentials or bad Terraform code, I'd think it should be possible for users to set their own quota that is locked for, say, 24 hours to at least put a cap on abuse scenario costs.

Thank you for sharing your idea with us!

Just wanted to clarify; currently, our platform operates on a pre-paid system where users are required to make a deposit to initiate server usage. We do offer the option for users to deposit smaller amounts, starting from $5, which helps minimize potential losses in the event of stolen credentials. Additionally, we have an Auto Top-Up system that can automatically replenish your balance based on the settings you specify. We also send notification emails to ensure that any transaction made is intentional and authorized.

As a result, even if you have full access to our platform, you would need an account balance to deploy or maintain virtual machines.

But we genuinely appreciate your suggestion and will give serious consideration to incorporating such a security feature into our platform. Thank you once again for your valuable input!

Great time to be "selling umbrellas" :) good luck on your business! Exciting stuff!

Thank you so much! :)

I feel like I'm missing some things on your pricing page. In the GPU Instances table, what are the units of "System RAM"? In the CPU Instances table, what does "the cost of RAM is included in the per vCPU price" mean, in terms of how much RAM?

In the Comparison table, some of the savings percentages seem to be calculated incorrectly, as the percentage of the price left instead of the percentage of the price saved (e.g. 3.99/5.88 instead of (5.88-3.99)/5.88).

The pricing is transparent but complicated. I think you need calculator that lets you select different options and tells you the price. Maybe you have such a thing? I didn't find it.

Thank you for your questions, and I apologize for any confusion caused.

We offer two types of virtual machines: GPU-based and CPU-based. The pricing structure differs slightly for each.

For GPU-based VMs: The vCPU/GPU and RAM/GPU ratio is set at a minimum of 1-to-1. This means that if you deploy a VM with one GPU, you will need at least 1 vCPU and 1GB of RAM. As our platform is fully customizable, we bill you for the specific amount of RAM ($0.006/hr per GB) and the number of vCPUs ($0.011/hr per vCPU) you select.

For CPU-based VMs: The RAM/vCPU ratio is fixed at 4. In this case, only the vCPU is customizable, and you will only be billed for the number of vCPUs you choose. The cost of RAM is already included in the price you pay for the vCPU.

Here's an example: A GPU-based VM with 1x Quadro RTX 4000 ($0.27/hr), 1 vCPU ($0.011/hr), 1GB of RAM ($0.006/hr), and a 40GB NVMe drive (40 x $0.00011/hr).

Another example: A CPU-based VM with 1x AMD EPYC Rome ($0.033/hr), 4GB of RAM (FREE/INCLUDED), and a 40GB NVMe drive (40 x $0.00011/hr).

We also have a calculator available on the VM deployment page, where you can see detailed information about the specific configuration you want to deploy. Once you register on our console, you can access the calculator at https://console.oblivus.com/dashboard/oblivuscloud/deploy/.

Please note that I have forwarded the comparison calculations internally, and they should be fixed soon. Thank you for bringing it to our attention!

I think you should have the calculator on the pricing page for everyone, near the top. It would be a good marketing tool to help people decide to sign up.

Definitely a great idea, we will do it in the near future. Thank you once again!

So this is about 30% cheaper than GCP. But GCP sustained use discounts give you a 30% discount if used for the whole month.

Azure and AWS pricing for GPU’s is insane.

lambdalabs seems to be the cheapest but I haven’t used it. Anyone has experience with it?

The issue with lambda is that they're always booked out in my experience.

Nobody goes there any more, it’s impossible to get a table

Hello, thank you for providing valuable feedback.

If you require sustained usage, we offer long-term reservations with discounts of up to 50% off our regular on-demand prices. For assistance in selecting the most suitable option for your specific needs, please feel free to email us at business@oblivus.com. We'll be happy to assist you.

I suggest using vast.ai instead, much cheaper than this.

"oblivus": 1x RTX A5000 for $0.84/hr vast: 4x RTX A5000 for $1.08/hr, 1x RTX A5000 for $0.20/hr, and much better and larger selection.

This is a duplicate, so I'm pasting my previous answer here as well.

"They operate a model known as a "marketplace" or "community" cloud. Unlike us, they utilize servers hosted by individuals in their homes. This is something completely different.

While they may offer lower prices, the infrastructure may have limitations in terms of reliability, scalability, and security compared to data center-based providers. Additionally, the performance and availability of the services vary depending on the hosts' hardware and internet connection."

Once again, no, not just individuals in their homes

Okay, then let me paraphrase it differently... While it's possible for individuals or companies to host servers in data centers or actively participate in the system, this doesn't alter the fact that Vast.ai operates as a community cloud. The essential point remains unchanged: they offer a distinct and different service compared to us.

On equivalent instances it seems a good bit more expensive than e.g. lambdalabs [1], but in turn you get a lot more flexibility. Both in terms of having more GPU models to choose from, and having much more flexible machine configurations. Seems like a worthwhile tradeoff.

1: https://lambdalabs.com/service/gpu-cloud#pricing

https://oblivus.com/pricing/

I am using an 8*A100 machine from Lambda Labs, but I honestly cannot understand how it can be so cheap. I haven't found a cheaper one than this. It is even cheaper than most of spot-like interruptible offerings.

Agreed. Even by their on demand price, cost per A100 per year is <$10k. Even if they somehow make maintenance and electricity cost to close to 0(which likely won't be the case), and their GPU usage 100% they are running it with operational loss for the first year.

Thank you for your valuable feedback!

We often receive comparisons to Lambda, but it ultimately boils down to your specific requirements and preferences. As you mentioned, our platform is tailored for personalized virtual machine configurations, distinguishing us from other cloud service providers that offer pre-set configurations. We strongly believe that this level of flexibility is a key differentiator and delivers enhanced value to our customers.

Interesting name choice. I read it as "oblivious" even after looking at it a few times.

Hahaha, yeah,

Even though Oblivus doesn't have a specific meaning, we initially didn't make the connection with "oblivious". When we attended Cloudfest recently, many people also pronounced it as "oblivous," so you're not alone in that regard, I suppose. :)

Definitely submit a PR to add Oblivus to https://cloud-gpus.com/

Will do! Thank you very much for the suggestion!

In a world of shared GPUs, what tools do people use to profile their GPU workloads. Are there code level profilers like for CPUs? I typically hear that people don’t often have issues with the code running on the GPU itself but rather ensuring it is fed information quickly enough to stay saturated. How do people measure this and find out what to improve, especially in a shared setting?

What is your hypervisor technology? How do you ensure isolation between tenants?

PS: I see that your VMs are created in CoreWeave datacenters. I wonder the relationship between your company and CoreWeave. Are you just a reseller?

Hello,

We utilize the KVM hypervisor to ensure that all resources are dedicatedly allocated to the virtual machine.

They are among the vendors we collaborate with within the United States. To reiterate what I previously mentioned to another user, our on-demand stock comprises a diverse range of resources obtained from multiple vendors. In addition to our own infrastructure, we leverage resources from these vendors to effectively address the increasing demand for our services.

Re the pricing I see all prices given per hour (which I think is sensible), but I can't find what the granularity of usage measurement is.

Assume I try something out on a GPU server and I realize I made some wrong assumption in my code and I need to think things over, so I shut down the instance after 2h 18min 24s of usage. What do I pay for?

Ed. Nevermind, I found the answer here https://docs.oblivus.com/billing/payment-plans, seems to be per minute, so I assume 2h 19min.

Hello,

Yes, we bill our services on a minute basis, so you will only be charged for the actual usage duration, which in this case would be 139 minutes (2 hours and 19 minutes). You can find detailed information about the resources you have used and their respective costs on the Billing > Invoices page.

I hope this helps!

I always thought there were license restrictions from Nvidia preventing the use of their consumer grade GPUs in datacenters / cloud environments. Is this not the case?

Hello, thank you for the question!

To clarify, we don't offer any consumer cards (RTX 3080, RTX 4090, etc) mentioned in the thread. Instead, we provide professional cards from the RTX Series such as the RTX 4000 and RTX 5000.

While vast.ai (see other comments) may not be a direct competitor, Runpod is. And Runpod is significantly cheaper https://www.runpod.io/gpu-instance/pricing

Runpod: A100 80GB for $1.990/hr

vs

Oblivus: A100 80GB for $2.41/hr

How do you justify being significantly more expensive?

Hello, thank you for your question!

It's important to note that our platform offers full customizability, while they only offer pre-set configurations. This level of flexibility comes with a trade-off in terms of pricing.

Regarding their infrastructure and system, I don't have specific information. However, I couldn't find any details about long-term reservations on their website. In comparison, -for example- we offer reserved instances with A100s starting as low as $1.2 per hour.

There are also other factors to consider. We prioritize investing in data center infrastructure, security, quality, and reliability. For instance, we provide up to 40Gbps public, 200Gbps private network connectivity for each virtual machine and offer 3x replication for storage. These features come with their own trade-offs, and each company has its own pricing structure.

I hope this clarifies the situation!

Can I put an instance to sleep and just pay for the storage ?Could not find the info on your site.

Also for Windows instance, what do people use to connect. RDP seems slow. My primary use case is software development trying to learn CUDA programing.

Hello, thank you for your questions!

At the moment, when you shut down a server, you are only billed for the storage and the IP address associated with it. However, we are working on implementing a feature in the near future that will allow you to detach the IP address, enabling you to pay solely for the storage usage.

Regarding Windows servers, the default software for remote access is RDP. However, you have the freedom to install and use any other remote access software of your choice on our platform.

Hope this clarifies things for you!

How does cloud gaming work from a customer perspective? I see it mentioned on the main page but I haven't found any additional information about it (on your mobile page).

Thank you for your question!

We offer Windows (BYOL) images on our virtual machines. If you have a good latency to our servers and sufficient bandwidth to handle the streaming, you can enjoy cloud gaming using any software of your choice.

To get started, you can deploy a CPU-based Windows virtual machine for as low as $0.019/hr. Once you have installed your games, you can stop the virtual machine and modify it into a GPU-based VM for as low as $0.29/hr. While your server is running, you will be billed for all the components. However, if you shut down the server, you will only be billed for the storage and the IP address. Your data will remain safe as long as you don't delete the server, allowing you to start playing again whenever you want.

In the coming days, we will be implementing a feature that allows you to detach the IP address, resulting in more cost-effective billing, where you will only be billed for the storage.

I hope this information helps!

Neat - thanks! What I would find helpful is a calculator to estimate the total cost per day, including storage, ram, etc. It would be great if it allowed me to easily answers questions like this one.

Assuming that:

- I game 2 hours a day,

- I need 50 GB for Windows and another 200 GB for games,

- I need 16 GB of ram and an equivalent of RTX 4080,

what would my total cost per day would be?

We have plans to add a calculator in the coming days to make it easier for you to estimate costs. Currently, the console deploy page is the only place where you can see the hourly pricing.

As for finding a GPU equivalent to the RTX 4080, it's challenging as it is a new release and we don't have benchmarks for it yet. However, we can provide pricing estimates for the RTX 5000 and RTX 4000 GPUs.

For the RTX 4000, the approximate cost would be around $24, while for the RTX 5000, it would be around $45. Please keep in mind that these are estimates and the actual pricing may vary. Additionally, the optimization we are planning to implement for IP addresses in the near future will further improve the overall pricing.

Curious how this compares to Tensordock? I've had great experiences with their servers + support. Vastly preferable to other first party / distributed options.

Hi there, thank you for your inquiry!

Upon review of their system, it appears that although there may be similarities, there are some notable differences between their service and ours.

Specifically, it seems that their CPU-based virtual machines come with preset configurations, lacking full customizability. Additionally, I was unable to locate an additional storage solution, reserved instances, organization and auto top-up system on their platform.

On the other hand, we do not plan on offering anything similar to their "marketplace cloud" with individual hosts.

Thanks!

I’m new to this cloud gpu thing, how do I use this?

Do I get access to a vm with powerful gpus?

I would love to be able to execute dockerfiles over the cli that runs gpu intensive executables.

Hello!

To begin, simply create an account at https://console.oblivus.com and add a minimum of $5 to your account balance. Alternatively, you can utilize the promo code HN_1 on the same page for testing purposes.

Once you deploy your server, you'll have access to a dedicated virtual machine, providing you with the capability to perform any tasks you would typically carry out on your own computer. You also have the freedom to execute Docker within your virtual machine.

Please have a look at our documentation at https://docs.oblivus.com, where we tried to explain everything with images directly from our console. If you have any other questions, I'd be happy to answer them!

That page is not mobile friendly FYI :-)

Thank you for bringing this to our attention.

Our team will look into this matter internally to ensure that everything is functioning properly. We appreciate your feedback!

Looks really cool. Congrats on the launch! A quick question. Do you allow to build custom public images? We’d love to integrate dstack.ai with Oblivus.

Thank you very much!

We are currently working on developing a feature that will allow users to create snapshots of their system disks and use those images to create multiple instances. We expect this feature to be available to everyone in a few weeks.

If you don't mind, could you please send an email to doruk@oblivus.com so that we can discuss this further?

Hey, we produce AI tutorials on youtube - it'd be cool if we could maybe produce a video showcasing your GPU cloud?

Hello!

That sounds like an excellent suggestion. Would you mind sending an email to doruk@oblivus.com with a brief description of the video and details about your channel?

Thanks!

what is the differentiation from RunPod? (they have pretty much same prices but with more services..)

Hello, thank you for your question.

The main difference between our platform and theirs is that we offer fully customizable configurations. From vCPU to RAM, Disk, and GPU, you have the flexibility to customize each aspect according to your specific needs. Unlike pre-set configurations offered by other providers, our platform allows you to tailor your virtual machines to your exact requirements.

Is there any chance you’re going to offer the service in EU?

Hello, thank you so much for the question!

We have plans to expand to the EU, potentially in the third quarter of 2023.

Building our current infrastructure in the EU is quite difficult, especially with the high electricity prices in EU data centers. However, we have been in discussions with various providers to establish an infrastructure that meets our requirements.

Additionally, we are actively working on a solution called Oblivus Edge Deployment, which aims to offer low-latency and high-bandwidth connectivity to EU users, even if the server is located in the US. We expect to release this technology in the next few weeks.

Great to see another option in the space! Good luck!

Thank you so much! :)