General ML research and experiments.
Most published research uses RTX consumer cards. Spinning up a single 24GB GPU for an afternoon of experiments is cheap and sufficient.
Top GPUs for this workload.
Ranked by suitability — higher fitness scores mean the card handles this workload more comfortably.
| GPU | Tier | VRAM | Fit | |
|---|---|---|---|---|
| consumer | 32GB | 80 | Compare → | |
| consumer | 24GB | 80 | Compare → | |
| workstation | 48GB | 75 | Compare → | |
| consumer | 24GB | 70 | Compare → | |
| workstation | 32GB | 70 | Compare → | |
| workstation | 24GB | 70 | Compare → | |
| workstation | 16GB | 65 | Compare → | |
| consumer | 16GB | 65 | Compare → | |
| datacenter | 24GB | 65 | Compare → | |
| datacenter | 16GB | 50 | Compare → |
Top models for this workload.
OLMo 3 7B
Allen AI's latest fully-open OLMo — model + training data + checkpoints.
Claude Opus 4.7
Frontier reasoning and long-form coding from Anthropic.
OLMo 7B
Qwen 3 235B
Alibaba's frontier MoE — 235B total / 22B active.
DeepSeek R1
DeepSeek's reasoning model — RL-trained, frontier-class, MIT-licensed.
Llama 3.1 405B
Meta's largest open-weight LLM — dense 405B, frontier-class at launch.
Gemini 2.5 Pro
Google's frontier reasoning model with native 1M-token context.
GPT-5
OpenAI's frontier multimodal reasoning model.
GPT-OSS 120B
OpenAI's first open-weight LLM in years — Apache-licensed MoE.
DeepSeek R1 Distill Llama 70B
70B Llama distilled from DeepSeek R1's reasoning traces.
DeepSeek V3
DeepSeek's flagship MoE — 671B total, 37B active, frontier-class.
Llama 3.3 70B
Meta's best-in-class open-weight LLM — 70B class.