Show HN: San Francisco Compute – 512 H100s at <$2/hr for research and startups (sfcompute.org)

Article summary

The article discusses San Francisco Compute, a service offering 512 H100s at less than $2/hour for research and startups. The comments reveal that a similar service, TPU Research Cloud, has become less accessible to hobbyists and tinkerers due to high demand and limited resources. The discussion focuses on the challenges of accessing powerful computing resources for research and development. The service's availability and reliability are debated among commenters.

Main themes

• access to computing resources
• research and development
• optimism vs realism
• risk-taking and caution
• TPU Research Cloud
• compute availability and reliability

What commenters say

• Some commenters believe that TPU Research Cloud has become less supportive of hobbyists and tinkerers due to internal priorities and limited resources.
• Others argue that the program is still going strong and has expanded its compute pool to include Cloud TPU v4 Pod slices.
• The reliability and yield of TPUs are questioned, with some attributing issues to hardware problems and others to preemption by paying customers.
• There is a debate about the importance of optimism versus realism in pursuing innovative projects, with some arguing that optimism is necessary for success and others advocating for a more realistic approach.
• The discussion also touches on the trade-offs between risk-taking and caution, with some commenters emphasizing the need for persistence in the face of uncertainty.
• Some commenters express frustration with the limited access to resources, while others share their positive experiences with the TPU Research Cloud program.
• The role of Google's internal priorities and the impact of preemption on research projects are also discussed.