MODEL ARCHITECTURE SPEC โข GOOGLE26.4B MoE
Google Gemma 4 26B MoE VRAM & inference speed
26.4B Total Params4.2B Active (MoE)46 Layers โข 4 KV Heads32K Default Ctx (Max 128K)Ollama: gemma4:26b
Accurate memory requirements, KV cache footprint, and tokens/sec generation benchmarks for running Google Gemma 4 26B MoE across local GPUs, Apple Silicon Macs, and cloud instances.
โ
Runs 100% in GPU VRAM (Optimal Headroom)20.23 GB (84%)~208 t/s
Google Gemma 4 26B MoE on NVIDIA GeForce RTX 4090 (24GB GDDR6X)
โ Fits Comfortably in VRAM
โ FITS IN VRAM
Required VRAM
20.23 GB
84% of hardware capacity (24 GB)Est. Decode Speed
~208 tok/s
4.2B active params @ 1008 GB/sAvailable Headroom
+3.77 GB
Safe for OS display & full contextMemory Allocation & Headroom
20.23 GB allocated of 24 GB capacity
84% utilized
Weights: 15.84 GB
KV: (32K): 3.09 GB
Overhead: 1.3 GB
Free Headroom: +3.77 GB
โก Architecture & Throughput Transparency: Single-user decode speed (~208 tok/s) is memory-bandwidth bound at batch size = 1 on NVIDIA GeForce RTX 4090 (24GB GDDR6X) (1008 GB/s). Prompt prefill / TTFT (~3,016 tok/s) is compute-bound and utilizes GPU Tensor cores. Actual latency varies with FlashAttention-2 and runtime engine (Ollama/vLLM/llama.cpp).
OLLAMA_FLASH_ATTENTION=1 OLLAMA_KV_CACHE_TYPE=q8_0 ollama run gemma4:26bCloud Instance Options for Google Gemma 4 26B MoE
On-DemandDeploy this model on an on-demand cloud GPU without local hardware constraints or power usage:
RunPod Cloud GPUsBest 1-Click Ollama & vLLM
$0.34/hr (RTX 4070 Ti) โ $2.39/hr (H100)Vast.ai GPU MarketplaceCheapest Dual RTX 3090 / 4090
$0.18/hr (RTX 3060) โ $0.55/hr (RTX 4090)Lambda GPU CloudEnterprise 8x H100 Clusters
$1.99/hr (GH200 / H100)Quantization Precision Matrix
32K ctx โข KV: Q8_0| Quantization | Weights | KV Cache | Total VRAM | Est. Speed | Fit Status |
|---|---|---|---|---|---|
| FP1616 bpw | 52.8 GB | 3.09 GB | 58.49 GB (244%) | 0 t/s | OOM |
| Q8_08.5 bpw | 27.98 GB | 3.09 GB | 32.8 GB (137%) | 0 t/s | OOM |
| Q6_K6.56 bpw | 21.65 GB | 3.09 GB | 26.25 GB (109%) | ~36 t/s | PARTIAL |
| Q5_K_M5.68 bpw | 18.74 GB | 3.09 GB | 23.24 GB (97%) | ~183 t/s | FITS |
| Q4_K_M4.8 bpw | 15.84 GB | 3.09 GB | 20.23 GB (84%) | ~208 t/s | FITS |
| FP44.25 bpw | 13.99 GB | 3.09 GB | 18.32 GB (76%) | ~228 t/s | FITS |
| IQ3_XS3.3 bpw | 10.82 GB | 3.09 GB | 15.04 GB (63%) | ~275 t/s | FITS |
Export Full Architecture & Infrastructure Spec
Get an executive-ready PDF report with exact memory math, hardware benchmarks, and cost comparisons.
Hardware Compatibility Matrix for Google Gemma 4 26B MoE
Standard Q4_K_M precision with 32K context window and Q8_0 KV cache
| Hardware System | VRAM Envelope | Required VRAM | Headroom | Est. Decode Speed | Compatibility | Test Fit |
|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5090 (32GB GDDR7) Consumer GPU โข 1792 GB/s | 32 GB | 20.23 GB | +11.77 GB | ~370 tok/s | โ FITS | Test Fit โ |
| NVIDIA GeForce RTX 4090 (24GB GDDR6X) Consumer GPU โข 1008 GB/s | 24 GB | 20.23 GB | +3.77 GB | ~208 tok/s | โ FITS | Test Fit โ |
| 2x NVIDIA RTX 3090 Dual Rig (48GB Total VRAM) Multi-GPU Rig โข 936 GB/s | 48 GB | 20.23 GB | +27.77 GB | ~176 tok/s | โ FITS | Test Fit โ |
| NVIDIA GeForce RTX 5080 (16GB GDDR7) Consumer GPU โข 960 GB/s | 16 GB | 20.23 GB | -4.23 GB | ~44 tok/s | โก PARTIAL | Test Fit โ |
| NVIDIA RTX 4070 Ti Super (16GB) Consumer GPU โข 672 GB/s | 16 GB | 20.23 GB | -4.23 GB | ~30 tok/s | โก PARTIAL | Test Fit โ |
| NVIDIA GeForce RTX 3060 (12GB) Consumer GPU โข 360 GB/s | 12 GB | 20.23 GB | -8.23 GB | 0 (OOM) | โ OOM | Test Fit โ |
| Apple Mac Mini M4 Pro (64GB Unified Memory) Apple Silicon โข 273 GB/s | 54 GB | 20.23 GB | +33.77 GB | ~52 tok/s | โ FITS | Test Fit โ |
| Apple Mac Studio M4 Max (128GB Unified Memory) Apple Silicon โข 546 GB/s | 112 GB | 20.23 GB | +91.77 GB | ~106 tok/s | โ FITS | Test Fit โ |
| Apple Mac Studio M5 Ultra (256GB Unified Memory) Apple Silicon โข 890 GB/s | 236 GB | 20.23 GB | +215.77 GB | ~173 tok/s | โ FITS | Test Fit โ |
| AMD Strix Halo Mini-PC (Ryzen AI Max+ 395 128GB) Unified APU / Mini-PC โข 256 GB/s | 110 GB | 20.23 GB | +89.77 GB | ~44 tok/s | โ FITS | Test Fit โ |
| NVIDIA DGX Spark (128GB Grace-Blackwell Unified) AI Appliance โข 273 GB/s | 118 GB | 20.23 GB | +97.77 GB | ~56 tok/s | โ FITS | Test Fit โ |
| NVIDIA H100 SXM (80GB HBM3 Cloud GPU) Datacenter Cloud GPU โข 3350 GB/s | 80 GB | 20.23 GB | +59.77 GB | ~733 tok/s | โ FITS | Test Fit โ |