Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: Oblivus GPU Cloud – Affordable and scalable GPU servers from $0.29/hr

oblivus.com

Post reply on HN
Print view View original

131–135 of 135 posts

Re: Show HN: Oblivus GPU Cloud – Affordable and scalable GPU servers from $0.29/hr

#131

Post by lostmsu »

YetAnotherNick wrote: post #38
pitched wrote: post #19

Earlier quoted context omitted.

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.nv…

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.

Re: Show HN: Oblivus GPU Cloud – Affordable and scalable GPU servers from $0.29/hr

#132

Post by 71a54xd »

oblivuslimited wrote: post #108
71a54xd wrote: post #98

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…

Thanks!

Re: Show HN: Oblivus GPU Cloud – Affordable and scalable GPU servers from $0.29/hr

#134

Post by oblivuslimited »

michele_f wrote: post #133

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.

Re: Show HN: Oblivus GPU Cloud – Affordable and scalable GPU servers from $0.29/hr

#135

Post by 0x008 »

crabbone wrote: post #115
dougSF70 wrote: post #113

Earlier quoted context omitted.

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…

NVLink was discontinued with 40 series nvidia consumer GPUs.
Post reply on HN