Where can i find the cheapest on-demand GPUs with more than 80GB ram
- User
Where can i find the cheapest on-demand GPUs with more than 80GB ram
- Assistant
Finding the absolute cheapest on-demand GPU with over 80GB of VRAM involves looking at both major cloud providers (hyperscalers) and specialized GPU cloud companies. The most common GPUs meeting this requirement are the NVIDIA A100 80GB and the NVIDIA H100 80GB (or higher variants like H200 141GB)[1][2][3]. Here's a breakdown of options and pricing, keeping in mind that prices fluctuate and may vary by region and specific instance configuration:
Specialized GPU Cloud Providers (Generally Cheaper):
- RunPod: Often cited as a cost-effective option[4]. They offer various GPUs, including:
- A100 80GB (PCIe): Starting from 1.64/hr (Secure Cloud).
- A100 80GB (SXM): Starting from $1.89/hr (Secure Cloud)[3].
- H100 80GB (PCIe): Starting from 2.39/hr (Secure Cloud)[3].
- H100 80GB (SXM): Starting from 2.99/hr (Secure Cloud)[3].
- H100 NVL (94GB): Starting from 2.79/hr (Secure Cloud)[3].
- H200 SXM (141GB): Starting from 3.99/hr (Secure Cloud)[3].
- RunPod bills by the minute and has no ingress/egress fees[3]. Community Cloud instances might be less reliable than Secure Cloud[1][3].
- Lambda Labs: Known for simplicity and ML focus[5][6].
- A100 80GB (SXM): $1.79/hr per GPU (typically offered in 8x configurations)[5][7].
- H100 80GB (SXM): $2.99/hr per GPU (typically offered in 8x configurations)[5].
- They offer per-minute billing with no egress fees[5]. Availability can sometimes be inconsistent[8][6].
- TensorDock: Promotes cost-effectiveness[1][2].
- A100 80GB: From $1.63/hr[2].
- H100 80GB (SXM5): Starting from 3/hr[1].
- Vast.ai: Uses a marketplace model, prices vary but can be very low[9][10]. Reliability can be a concern[9][6].
- A100 80GB: Around $1.35/hr (median)[7].
- H100 SXM: 2.67/hr reported in April 2025[11]. Interruptible instances can be cheaper[4].
- FluidStack: Another option often compared[8].
- A100 80GB: $1.49/hr[7].
- H100: $2.00/hr[7].
- Hyperstack:
- A100 80GB: $1.35/hr.
- H100 80GB (SXM): $2.40/hr[12].
- DataCrunch.io:
- A100 80GB: $1.75/hr.
- H100: 3.35/hr[13]).
- CoreWeave:
- A100 80GB: $2.21/hr.
- H100: $4.25/hr[7].
Major Cloud Providers (Hyperscalers - Generally More Expensive On-Demand):
- Google Cloud Platform (GCP): Offers A100 80GB and H100 80GB[13][14].
- A100 80GB (a2-ultragpu-*): Starting around 6.25/hr for a single GPU instance[7][13].
- H100 80GB (a3-highgpu-*): Starting around $11.06/hr for a single GPU instance[13].
- GCP might be slightly cheaper than AWS and Azure for comparable instances[13].
- Microsoft Azure: Offers A100 80GB and H100[13][15].
- A100 80GB: Starting around $3.67/hr[7][13].
- H100: Starting around $6.98/hr[13].
- Amazon Web Services (AWS): Currently does not appear to offer A100 80GB or H100 on-demand, focusing on 40GB A100s or large multi-GPU H100 configurations[13][16]. Their listed p4d/p4de instances use A100 40GB[13][9].
Summary & Considerations:
- Cheapest Potential: Specialized providers like RunPod (Community Cloud), Vast.ai (marketplace/interruptible), Hyperstack, and FluidStack seem to offer the lowest starting on-demand prices for A100 80GB GPUs, often below 1.99-$3/hr[1][3].
- Reliability vs. Cost: Cheaper options like RunPod's Community Cloud or Vast.ai's marketplace might have lower reliability compared to Secure Cloud options or established providers like Lambda Labs or the major hyperscalers[17][9][1].
- Availability: High-demand GPUs like the A100 80GB and H100 can have limited availability, especially on cheaper platforms[8][6].
- Billing Granularity: Providers like RunPod and Lambda Labs offer per-minute billing, which can be more cost-effective for short tasks compared to hourly billing[17][5][3].
- Hidden Costs: Factor in potential costs for data storage, network egress (data transfer out), and specific instance configurations (CPU, RAM), although some providers like RunPod and Lambda Labs advertise no egress fees[9][5][3]. Hyperscalers often have significant data egress fees[17].
For the absolute lowest on-demand cost for a >80GB GPU, explore RunPod (Community Cloud), Vast.ai, FluidStack, and Hyperstack, carefully considering the trade-offs in reliability and availability[17][9][7]. Lambda Labs and TensorDock are also strong contenders with competitive pricing[1][5][2].
- RunPod: Often cited as a cost-effective option[4]. They offer various GPUs, including:
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.