Gemma 4 26B A4B · 17 GPUs measured first-party · llama.cpp Q4_K_M · Updated October 2026
What GPU Do You Need for Gemma 4 26B A4B?
We ran Gemma 4 26B A4B (Q4_K_M, llama.cpp) on 17 GPUs, from budget cards to datacenter parts, and measured generation speed, prompt processing, power draw and peak VRAM on every one. The fastest was NVIDIA GeForce RTX 5090 at 264.7 tok/s.
Benchmarked weights: bartowski/google_gemma-4-26B-A4B-it-GGUF
NVIDIA GeForce RTX 5090
264.7 tok/s on Gemma 4 26B A4B, the ceiling. Measured on our bench. 32GB of VRAM, $1,999 at launch.
- 264.7 tok/s on Gemma 4 26B A4B
- 32GB, clears the Gemma 4 26B A4B floor
- 575W board rating
NVIDIA GeForce RTX 4090
188.2 tok/s on Gemma 4 26B A4B, fastest card you can buy at retail. Measured on our bench. 24GB of VRAM, $1,599 at launch.
- 188.2 tok/s on Gemma 4 26B A4B
- 24GB, clears the Gemma 4 26B A4B floor
- 450W board rating
NVIDIA GeForce RTX 3090
154.3 tok/s on Gemma 4 26B A4B, lowest launch price that still fits. Measured on our bench. 24GB of VRAM, $1,499 at launch.
- 154.3 tok/s on Gemma 4 26B A4B
- 24GB, clears the Gemma 4 26B A4B floor
- 350W board rating
NVIDIA GeForce RTX 3090 Ti
168.0 tok/s on Gemma 4 26B A4B, most speed per dollar. Measured on our bench. 24GB of VRAM, $1,999 at launch. That is 84.03 tok/s per $1,000 of launch price.
- 168.0 tok/s on Gemma 4 26B A4B
- 24GB, clears the Gemma 4 26B A4B floor
- 450W board rating
What GPU Do You Need for Gemma 4 26B A4B?, tok/s by GPU
Top 15 shown; 2 more cards in the full table below.
Efficiency: tok/s per 100W drawn
Top 15 shown; 2 more cards in the full table below.
Power is the average pulled during the run, sampled at 1Hz. The fastest card is often not the one here, and for anything left running this is the number that shows up on the bill.
Value: tok/s per $1,000 of MSRP
Top 15 shown; 2 more cards in the full table below.
Launch price, not street price, so it ages. A speed leaderboard always crowns the most expensive card; this is the counterweight.
Gemma 4 26B A4B. Measured generation speed by GPU
| GPU | tok/s | Prompt t/s | tok/W | Avg power |
|---|---|---|---|---|
| NVIDIA GeForce RTX 5090 | 264.7 | 12450.3 | 1.23 | 214.5 W |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 230.3 | 11048.9 | 1.49 | 154.7 W |
| NVIDIA H100 80GB HBM3 | 218.4 | 7041 | 1.29 | 168.9 W |
| NVIDIA GeForce RTX 4090 | 188.2 | 9283.6 | 1.35 | 139.2 W |
| NVIDIA GeForce RTX 3090 Ti | 168 | 5314.3 | 0.82 | 205.7 W |
| NVIDIA L40S | 156.6 | 7606.5 | 1.15 | 136.6 W |
| NVIDIA GeForce RTX 3090 | 154.3 | 4461.2 | 0.67 | 231.2 W |
| NVIDIA RTX A6000 | 138.1 | 4181.1 | 0.88 | 156.3 W |
| NVIDIA A100 80GB SXM4 | 136.8 | 3877.7 | 1.12 | 122.2 W |
| NVIDIA A100 40GB PCIe | 136.7 | 4069 | 1.26 | 108.5 W |
| NVIDIA RTX PRO 4000 Blackwell | 134.5 | 4170.2 | 1.6 | 83.9 W |
| NVIDIA RTX A5000 | 128.7 | 3588.1 | 0.86 | 150.1 W |
| NVIDIA Titan RTX | 119.9 | 2796.6 | 0.74 | 162.4 W |
| NVIDIA RTX 4500 Ada Generation | 104.3 | 4876.2 | 1.41 | 73.7 W |
| NVIDIA A10G | 101.9 | 2890.8 | 0.97 | 105.1 W |
| NVIDIA Quadro RTX 8000 | 96.49 | 2417.3 | 0.84 | 115.0 W |
| NVIDIA L4 | 68.51 | 2460.9 | 1.28 | 53.7 W |
What the numbers show. Across 17 GPUs measured on our own bench, RTX 5090 is fastest at 265 tok/s. RTX PRO 4000 Blackwell is the most efficient, 135 tok/s at 84W (1.603 tok/W). The measured peak was ~18GB, so it runs on 24GB cards and up. That floor is a property of the model, so it applies to every GPU, measured or not.
How it compares. H100 80GB HBM3: Gemma 4 26B A4B 218.4 tok/s, Qwen-AgentWorld-35B-A3B 219.1, Ornith-1.0-9B 216.9, gpt-oss-120b 220.4 (117B), Ornith 1.5 9B 221.4 (10B). 3 of 4 beat Gemma 4 26B A4B here.
Cost on a rented GPU. 1M generated tokens of Gemma 4 26B A4B: $0.22 on a RTX 3090 ($0.12/hr, 108 min), $0.41 on a RTX 5090 ($0.39/hr, 63 min, 1.9x the cost).
Gemma 4 26B A4B: cost per 1M generated tokens on rented GPUs
| GPU | Cheapest rate | Speed (tok/s) | Cost per 1M generated tokens |
|---|---|---|---|
| NVIDIA GeForce RTX 3090 | $0.12/hr | 154.3 | $0.22 |
| NVIDIA Titan RTX | $0.15/hr | 119.9 | $0.35 |
| NVIDIA RTX A5000 | $0.16/hr | 128.7 | $0.35 |
| NVIDIA GeForce RTX 5090 | $0.39/hr | 264.7 | $0.41 |
| NVIDIA RTX PRO 4000 Blackwell | $0.20/hr | 134.5 | $0.42 |
| NVIDIA GeForce RTX 3090 Ti | $0.27/hr | 168 | $0.45 |
| NVIDIA GeForce RTX 4090 | $0.34/hr | 188.2 | $0.50 |
| NVIDIA RTX A6000 | $0.33/hr | 138.1 | $0.66 |
| NVIDIA Quadro RTX 8000 | $0.26/hr | 96.49 | $0.73 |
| NVIDIA A100 40GB PCIe | $0.45/hr | 136.7 | $0.91 |
| NVIDIA RTX 4500 Ada Generation | $0.36/hr | 104.3 | $0.96 |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | $1.08/hr | 230.3 | $1.30 |
| NVIDIA L40S | $0.79/hr | 156.6 | $1.40 |
| NVIDIA L4 | $0.44/hr | 68.51 | $1.78 |
| NVIDIA A100 80GB SXM4 | $0.95/hr | 136.8 | $1.92 |
| NVIDIA H100 80GB HBM3 | $2.14/hr | 218.4 | $2.72 |
Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.
Speed tiers for Gemma 4 26B A4B. 30+ tok/s: 17 (RTX 5090, RTX 4090, RTX 3090 Ti). 30 tok/s is roughly where replies outpace reading.
Reading your prompt. Before Gemma 4 26B A4B writes anything it reads the input: 12450.3 tok/s on the RTX 5090 (0.3s for a 4,000-token prompt), 9283.6 on the RTX 4090 (0.4s), 2417.3 on the Quadro RTX 8000 (1.7s). Long documents and big code files feel this number more than the generation speed.
VRAM for Gemma 4 26B A4B. Measured peak 16.7GB, so 24GB is the smallest common card size; smallest card it ran on: RTX 4090 (24GB). With long context: Q4_K_M 20GB (tested), Q2_K 13GB, Q3_K_M 16GB, Q5_K_M 23GB, Q6_K 27GB.
Power on Gemma 4 26B A4B. Most efficient: RTX PRO 4000 Blackwell, 84W, 0.17 kWh per 1M generated tokens. Hungriest: RTX 3090, 231W, 0.42 kWh. At $0.15/kWh: $0.026 per 1M generated tokens.
Our verdict
Fastest on Gemma 4 26B A4B: NVIDIA GeForce RTX 5090, 264.7 tok/s. Cheapest consumer card that ran it: NVIDIA GeForce RTX 3090 ($1,499, 154.3 tok/s). Cheapest to rent per job: NVIDIA GeForce RTX 3090, $0.22 per 1M generated tokens.
FAQ
- What GPU do I need to run Gemma 4 26B A4B?
- About 18GB. Cheapest consumer card that ran it: NVIDIA GeForce RTX 3090 (24GB, 154.3 tok/s).
- How fast is Gemma 4 26B A4B on the NVIDIA GeForce RTX 5090?
- 264.7 tok/s at 214W, faster than every datacenter card we ran it on.
- How much does it cost to run Gemma 4 26B A4B in the cloud?
- $0.22 per 1M generated tokens on a NVIDIA GeForce RTX 3090 at $0.12/hr, cheapest of 16 rentable cards we measured.
- Can I run Gemma 4 26B A4B on a 12GB, 16GB or 24GB card?
- It used 16.7GB at the precision we tested. 12GB: no; 16GB: only at Q3_K_M (16GB); 24GB: yes.
- Is the RTX 4090 or the RTX 3090 faster for Gemma 4 26B A4B?
- The RTX 4090: 188.2 vs 154.3 tok/s, 22% faster on our bench.
- Should I buy or rent a RTX 5090 for Gemma 4 26B A4B?
- Its $1,999 launch price buys 5,139 rented hours at $0.39/hr, enough for about 4,897x 1M generated tokens of Gemma 4 26B A4B. Buy only if you'll run more than that.
How we test
llama.cpp llama-bench at Q4_K_M, 512-token prompt and 128 generated tokens, three runs after a warmup, full GPU offload, with power and VRAM sampled throughout. Token generation is memory-bandwidth-bound, so the ranking tracks bandwidth closely, which makes it a fair guide to cards we haven't run yet.