Does GLM-5.3-Flash AJ-IQ2_XXS fit in the NVIDIA RTX 3090's memory?

Estimate: the AJ-IQ2_XXS file is 87.3 GB and needs about 90.3 GB with an ~8K context; the NVIDIA RTX 3090 has about 23.5 GB usable, so it does not fully fit (part of it runs on the CPU).

Home › Models › GLM-5.3-Flash › NVIDIA RTX 3090

GLM-5.3-Flash on the NVIDIA RTX 3090

GLM-5.3-Flash runs at about 11.6 tokens/s on the NVIDIA RTX 3090 (measured, AJ-IQ2_XXS, CUDA, llama-bench, part of the model on CPU).

Best measured generation11.6tok/s

AJ-IQ2_XXS · CUDA

Prompt processing37tok/s

same run

For comparison, our size-scaling estimate for this combination is 66.7 tok/s.

Every run

Quant	Backend	Generate tok/s	Prompt tok/s	Test	Build / by / notes	Source
AJ-IQ2_XXS	CUDA	11.6	37	tg128	Single RTX 3090 with 40 of 42 expert layers in system RAM (-ncmoe 40); llama.cpp from upstream PR #27752. File size 87,346,006,560 bytes as stated by the source.; model file is larger than this card's VRAM, so part of it ran on the CPU/system RAM	aj9o9/GLM-5.3-Flash-GGUF model card (Hugging Face)

tg128 = generating 128 tokens; pp512 = reading a 512-token prompt (llama-bench). LocalScore averages several prompt and output lengths, so its numbers run lower than tg128. "decode, llama-server" = single-stream generation speed measured against a running llama-server (no speculative decoding).

Check prices

Amazon Newegg

As an Amazon Associate I earn from qualifying purchases.

Doesn't fit in this card's VRAM

Part of this model ran on the CPU here. A rented 48-80 GB GPU keeps it all in VRAM.

Rent a bigger GPU on RunPod

Affiliate link: if you sign up, RigBench may earn a referral credit. It never affects rankings or numbers.

All results for the NVIDIA RTX 3090 · GLM-5.3-Flash on other hardware

Embed this result

HTML

<a href="https://rigbench.pages.dev/model/glm-5-3-flash/on/nvidia-rtx-3090/"><img src="https://rigbench.pages.dev/badge/glm-5-3-flash--nvidia-rtx-3090.svg" alt="GLM-5.3-Flash · NVIDIA RTX 3090: 11.6 tok/s (AJ-IQ2_XXS, measured)"></a>

Markdown (GitHub README, forums)

[![GLM-5.3-Flash · NVIDIA RTX 3090: 11.6 tok/s (AJ-IQ2_XXS, measured)](https://rigbench.pages.dev/badge/glm-5-3-flash--nvidia-rtx-3090.svg)](https://rigbench.pages.dev/model/glm-5-3-flash/on/nvidia-rtx-3090/)