NVIDIA RTX 40

Can RTX 4090 run Gemma 4?

24 GB VRAM profile for desktop local Gemma 4 runs. The checker below ranks Gemma 4 E2B, E4B, 12B, 26B A4B, and 31B specifically for RTX 4090.

GPU
VRAM24 GB
BW~1008 GB/s
RAM16 GB
CORES16384
WebGPU
4RUNS GREAT
1RUNS WELL
2DECENT
0TIGHT FIT
0BARELY RUNS
0TOO HEAVY
Gemma 4 only
Gemma 4 E4BbestQ4_K_M

Better phones, Apple silicon laptops, and mainstream desktops.

Memory: 4.6 GBUse: 19%Speed: ~153 tok/sScore: 96/100
RUNS GREAT
96/100

Phones, tablets, old laptops, and the fastest local proof.

Memory: 3.1 GBUse: 13%Speed: ~228 tok/sScore: 95/100
RUNS GREAT
95/100

16GB+ unified-memory Macs and desktops with room for context.

Memory: 7.1 GBUse: 30%Speed: ~99 tok/sScore: 93/100
RUNS GREAT
93/100

The instruct 12B when you want chat, vision, and a 256K window on 16GB+ machines.

Memory: 7.1 GBUse: 30%Speed: ~99 tok/sScore: 93/100
RUNS GREAT
93/100

Gemma-family MoE when you want the newer diffusion decode path and multimodal prompts.

Memory: 13.8 GBUse: 58%Speed: ~51 tok/sScore: 73/100
RUNS WELL
73/100
Gemma 4 26B-A4B IT4B activeQ4_K_M

Strong desktop GPUs and disciplined 24GB+ Apple silicon setups.

Memory: 15.8 GBUse: 66%Speed: ~45 tok/sScore: 68/100
DECENT
68/100

High-end workstations where local quality matters more than convenience.

Memory: 19 GBUse: 79%Speed: ~37 tok/sScore: 60/100
DECENT
60/100

Full Model Directory

Explore 130+ Other Open-Weight Models on This Hardware

Compare Llama 3.3, Qwen 3, DeepSeek-R1, Mistral, and Phi on RTX 4090, or run a side-by-side GPU and Apple Silicon benchmark comparison.

Hardware Memory Analysis

Usable Memory Pool & Bandwidth Profile for RTX 4090

RTX 4090 provides 24 GB of dedicated graphics memory at ~1008 GB/s memory bandwidth. Keeping quantized model weights plus KV cache within ~85% of dedicated VRAM avoids silent Windows WDDM or Linux PCIe spillover into ~50 GB/s system DDR5 RAM, where throughput drops sharply.

Practical Model Budget

24 GB

Memory Bandwidth

~1008 GB/s

Recommended Checkpoint

Gemma 4 31B (Q4_K_M)

Gemma 4 Checkpoint Compatibility Matrix on RTX 4090

Evaluated at Q4_K_M baseline weights plus 0.5 GB baseline runtime overhead against RTX 4090's 24 GB usable budget. Leave at least 1.5–4 GB spare headroom for multi-turn KV cache expansion.

ModelQ4_K_M MemoryEst. SpeedGradePractical Headroom Verdict
Gemma 4 E2B3.1 GB~228 tok/sRuns greatComfortable fit (+20.9 GB spare for long context)
Gemma 4 E4B4.6 GB~153 tok/sRuns greatComfortable fit (+19.4 GB spare for long context)
Gemma 4 12B7.1 GB~99 tok/sRuns greatComfortable fit (+16.9 GB spare for long context)
Gemma 4 12B IT7.1 GB~99 tok/sRuns greatComfortable fit (+16.9 GB spare for long context)
Gemma 4 26B-A4B IT15.8 GB~45 tok/sDecentComfortable fit (+8.2 GB spare for long context)
DiffusionGemma 26B-A4B IT13.8 GB~51 tok/sRuns wellComfortable fit (+10.2 GB spare for long context)
Gemma 4 31B19 GB~37 tok/sDecentComfortable fit (+5 GB spare for long context)

NVIDIA CUDA, Flash Attention & PCIe Limits for RTX 4090

  • Run via LM Studio CUDA runtime, Ollama (`OLLAMA_FLASH_ATTENTION=1 OLLAMA_KV_CACHE_TYPE=q8_0`), or `llama.cpp` with `-ngl 99`.
  • Verify 100% GPU residency with `ollama ps` or Windows Task Manager Dedicated GPU Memory; if Shared GPU Memory spikes, step down one quant or cap context at 4K–8K.
  • If experimenting with Gemma 4 26B-A4B MoE on 12–16 GB RTX cards, test `llama-server -ot ".ffn_.*_exps.=CPU"` paired with 32 GB+ system RAM.

Frequently Asked Questions for RTX 4090

What is the best Gemma 4 model to run on RTX 4090?
Based on RTX 4090's 24 GB practical memory budget and ~1008 GB/s bandwidth, Gemma 4 31B (Q4_K_M, 19 GB) is the strongest comfortable fit (~37 tok/s).
Can RTX 4090 run Gemma 4 26B-A4B MoE locally?
Yes. At Q4_K_M (15.8 GB), Gemma 4 26B-A4B runs cleanly on RTX 4090 with 8.2 GB remaining for KV cache and runtime buffers.
How fast does RTX 4090 generate tokens with Gemma 4?
With ~1008 GB/s memory bandwidth, Gemma 4 31B at Q4_K_M averages roughly 37 tokens per second when fully resident without swap or PCIe spillover.
Which quantization format works best on RTX 4090?
Start with Q4_K_M (~4.5 bits/weight) to keep memory headroom free for 8K+ context windows. Step up to Q6_K or Q8_0 only if Q4_K_M consumes well under 60% of available VRAM or unified memory.