Runcrate

GPU cloud for AI developers and their agents: on-demand GPUs, 200+ open models behind one OpenAI-compatible API, and an MCP server. runcrate.ai

Joined on Oct 20, 2026 https://www.runcrate.ai

Runcrate

Oct 29

torch.compile vs eager for training a 468M Llama-style model on four GPUs: step time, memory and compile cost

#pytorch #machinelearning #gpu #python

5 min read

Want to connect with Runcrate?

Create an account to connect with Runcrate. You can also sign in below to proceed if you already have an account.

Create Account

Already have an account? Sign in

Runcrate

Oct 29

Text embedding throughput per dollar: BGE small, base and large on RTX 5090, L40S, RTX A6000 and H100

#embeddings #rag #gpu #python

6 min read

Runcrate

Oct 29

Fresh Hugging Face download vs local cache: time to first token for a 7B model on a Runcrate L40S

#huggingface #llm #gpu #python

5 min read

Runcrate

Oct 28

Whisper large-v3 hours of audio per dollar on four GPUs: RTX 5090, L40S, RTX A6000 and H100, five setups measured

#whisper #speechrecognition #gpu #python

6 min read

Runcrate

Oct 28

Full fine-tune vs LoRA vs QLoRA on one L40S: Qwen2.5-1.5B, memory, speed and held-out loss, measured

#machinelearning #llm #pytorch #gpu

6 min read

Runcrate

Oct 28

Ollama Q4 vs vLLM bf16 on a 24 GB A10: 3x the single-user speed, 1.2 points of GSM8K, and where vLLM still wins

#llm #ollama #vllm #gpu

6 min read

Runcrate

Oct 27

SDXL images per dollar on four GPUs: RTX A6000, L40S, RTX 5090 and H100, eager vs torch.compile

#stablediffusion #gpu #pytorch #benchmark

6 min read

Runcrate

Oct 27

vLLM FP8 vs bf16 on an H100: 1.4x the throughput, and what it did to GSM8K accuracy

#vllm #llm #gpu #benchmark

5 min read

Runcrate

Oct 27

vLLM prefix caching on vs off: 7.5x the throughput with a 4,000-token system prompt on one L40S

#vllm #llm #gpu #performance

6 min read

Runcrate

Oct 26

4 GPUs, 2x speed: why PyTorch DDP stalled on a PCIe-only 4x RTX A6000, and two flags that took it to 3.6x

#pytorch #gpu #machinelearning #performance

4 min read

Runcrate

Oct 26

How long until a rented GPU runs CUDA? Launch to first matmul on six GPU types, timed

#gpu #cuda #pytorch #devops

5 min read

Runcrate

Oct 26

vLLM on three GPUs: Qwen2.5-7B tokens per second and cost per 1M tokens on RTX A6000, L40S and H100

#vllm #llm #gpu #benchmark

5 min read

Runcrate

Oct 25

Six open models, one question: 1.3 s for the fastest answer, 16 s for the slowest, and hidden thinking made the difference

#llm #benchmark #python #ai

5 min read

Runcrate

Oct 22

Run your LLM evals every night on a GPU that only exists for a minute

#machinelearning #llm #devops #tutorial

5 min read

Runcrate

Oct 22

Fine-tune an open model on one GPU with a job that stops itself ($0.05, 4 minutes)

#machinelearning #python #llm #tutorial

5 min read

Runcrate

Oct 21

How to compare GPU cloud prices without fooling yourself

#gpu #cloud #machinelearning #python

5 min read

Runcrate

Oct 21

Switch from OpenAI to open models in two lines (streaming, tools and JSON mode still work)

#ai #openai #llm #python

4 min read

Runcrate

Oct 20

Give your coding agent a GPU, and a budget: the Runcrate MCP server in Claude Code, Codex and Cursor

#ai #mcp #machinelearning #claude

4 min read

DEV Community

Runcrate

Badges

Writing Debut

torch.compile vs eager for training a 468M Llama-style model on four GPUs: step time, memory and compile cost

Want to connect with Runcrate?

Text embedding throughput per dollar: BGE small, base and large on RTX 5090, L40S, RTX A6000 and H100

Fresh Hugging Face download vs local cache: time to first token for a 7B model on a Runcrate L40S

Whisper large-v3 hours of audio per dollar on four GPUs: RTX 5090, L40S, RTX A6000 and H100, five setups measured

Full fine-tune vs LoRA vs QLoRA on one L40S: Qwen2.5-1.5B, memory, speed and held-out loss, measured

Ollama Q4 vs vLLM bf16 on a 24 GB A10: 3x the single-user speed, 1.2 points of GSM8K, and where vLLM still wins

SDXL images per dollar on four GPUs: RTX A6000, L40S, RTX 5090 and H100, eager vs torch.compile

vLLM FP8 vs bf16 on an H100: 1.4x the throughput, and what it did to GSM8K accuracy

vLLM prefix caching on vs off: 7.5x the throughput with a 4,000-token system prompt on one L40S

4 GPUs, 2x speed: why PyTorch DDP stalled on a PCIe-only 4x RTX A6000, and two flags that took it to 3.6x

How long until a rented GPU runs CUDA? Launch to first matmul on six GPU types, timed

vLLM on three GPUs: Qwen2.5-7B tokens per second and cost per 1M tokens on RTX A6000, L40S and H100

Six open models, one question: 1.3 s for the fastest answer, 16 s for the slowest, and hidden thinking made the difference

Run your LLM evals every night on a GPU that only exists for a minute

Fine-tune an open model on one GPU with a job that stops itself ($0.05, 4 minutes)

How to compare GPU cloud prices without fooling yourself

Switch from OpenAI to open models in two lines (streaming, tools and JSON mode still work)

Give your coding agent a GPU, and a budget: the Runcrate MCP server in Claude Code, Codex and Cursor