Compatibility check
Can you run DeepSeek R1 Distill Llama 8B on the RTX 5090?
Yes. DeepSeek R1 Distill Llama 8B fits on the RTX 5090 (32 GB) in full FP16/BF16 precision, using about 19 GB including a 0.5 GB KV cache at 4,096 tokens. You have comfortable headroom for longer prompts and modest batching.
Memory breakdown
Weights plus a 0.5 GB KV cache at 4,096tokens, against the card's 32 GB. Verdicts leave ~10% headroom for activations and fragmentation.
| Precision | Weights | KV cache | Total | % of 32 GB | Fit |
|---|---|---|---|---|---|
| FP16 / BF16full quality | 18.5 GB | 0.5 GB | 19 GB | 59% | Fits |
| INT8 (8-bit)near-full quality | 9.2 GB | 0.5 GB | 9.7 GB | 30% | Fits |
| INT4 (4-bit)GPTQ / AWQ / GGUF Q4 | 4.6 GB | 0.5 GB | 5.1 GB | 16% | Fits |
Planning estimates, not a substitute for profiling. Real usage varies with the inference runtime, batch size, and how much context you actually use — the KV cache grows linearly with prompt length.
GPUs that run DeepSeek R1 Distill Llama 8B
Cards where this model fits (at its best precision):
Models that fit the RTX 5090
Other popular models that run on this card:
Go deeper
Frequently asked questions
Can the RTX 5090 run DeepSeek R1 Distill Llama 8B?
Yes. In FP16 it uses about 19 GB, which fits the RTX 5090's 32 GB.
How much VRAM does DeepSeek R1 Distill Llama 8B need?
Approximately 18.5 GB in FP16, 9.2 GB in INT8, and 4.6 GB in 4-bit for the weights, plus a KV cache of about 0.5 GB at 4,096 tokens.
Does quantization let DeepSeek R1 Distill Llama 8B fit on the RTX 5090?
Yes. Dropping to FP16 / BF16 brings total usage to about 19 GB, which fits the 32 GB card with headroom for the KV cache.
What happens to memory with longer context?
The KV cache grows linearly with prompt length. At 4,096 tokens it is about 0.5 GB here; doubling the context roughly doubles that term, so long-context use can push a tight fit over the edge.