Home / Value / Cheapest GPU for Gemma 4 31B Instruct
Cheapest GPU for Gemma 4 31B Instruct
Q4 · 8K (default)
Ranked by lowest fresh Amazon-matched board-partner SKU price among GPUs that fully fit this exact workload. This is not a speed ranking and not a “best GPU” list. Query combinations are not indexed.
To open another model, use the value explorer or the model catalog.
Requirement
- Model
- Gemma 4 31B Instruct
- Quantization · context
- Q4 · 8K
- Required VRAM
- 28.1 GB
- Calculator
- v1.0.0
18 canonical GPUs fully fit. 3 of them have a fresh current price. 15 compatible GPUs have no current price.
Formula: MIN(live fresh Amazon-matched board-partner SKU price) WHERE compatibility = FITS_IN_VRAM. Sort: price ASC, advertised VRAM ASC, canonical GPU id ASC. Prices are the lowest fresh Amazon-matched board-partner SKU, not MSRP or a street-price average.
Among 3 canonical GPUs with a fresh matched Amazon offer currently available in RigForAI. Ranked 3. Unpriced compatible: 15. This is not the entire US market — only canonical GPUs with a fresh matched Amazon offer currently in RigForAI.
Cheapest currently available full-VRAM GPU
NVIDIA RTX 5000 32 GB
- Advertised VRAM
- 32 GB
- Usable VRAM
- 28.8 GB
- Headroom
- 0.7 GB
- Compatibility
- FITS_IN_VRAM
- Current fresh Amazon price
- $4479.96
- Winning physical SKU
- Lenovo NVIDIA RTX 5000 Ada 32 GB GDDR6LenovoASIN B0FGQG418C
- Offer fetched
- 2026-08-17 14:52 UTC
Why it ranks first
- FITS_IN_VRAM for this model, quantization, and context
- current fresh matched Amazon offer (EXACT/HIGH, not a bundle)
- lowest price among 3 eligible priced GPUs
- price ASC, advertised VRAM ASC, canonical GPU id ASC.
Next option: NVIDIA RTX 5090 32 GB $4699.99 · difference +$220.03. Build a complete PC around this option · Workstations · Rent a server · Buy vs rent. GPU-only price is not a complete-build cost.
Other currently priced full-fit GPUs
Ranked by lowest fresh Amazon-matched board-partner SKU price among GPUs that fully fit this exact workload.
| # | GPU | VRAM | Current price | Headroom | $ / advertised GB | Winning SKU |
|---|---|---|---|---|---|---|
| 2 | NVIDIA RTX 5090 32 GB | 32 GB | $4699.99 Amazon | 0.7 GB | $147 | GIGABYTE GIGABYTE AORUS GeForce RTX 5090 MASTER 32G Graphics Card - 32GB GDDR7, 512bit, PCI-E 5.0, 2655MHz Core Clock, 3 x DP 2.1a, 1 x HDMI 2.1b, GV-N5090AORUS M-32GD ASIN B0DT7GHQMD |
| 3 | NVIDIA RTX 6000 48 GB | 48 GB | $20888.00 Amazon | 15.1 GB | $435 | HP HP NVIDIA RTX 6000 Ada 48 GB 4DP Graphics ASIN B0CTP2HHBN |
Compatible — current price unavailable
These GPUs calculate as a full VRAM fit. They are not inserted into the current-price ranking because RigForAI has no fresh matched Amazon offer for them. That is not the same as too expensive or incompatible.
- NVIDIA Tesla M10 32 GB · 32 GB advertised
- NVIDIA Quadro GV100 32 GB · 32 GB advertised
- NVIDIA RTX 8000 48 GB · 48 GB advertised
- NVIDIA RTX A6000 48 GB · 48 GB advertised
- NVIDIA H100 80 GB · 80 GB advertised
- NVIDIA H100 NVL 94 GB · 94 GB advertised
- NVIDIA A100 80 GB · 80 GB advertised
- NVIDIA A100 40 GB · 40 GB advertised
- NVIDIA L40S 48 GB · 48 GB advertised
- NVIDIA L40 48 GB · 48 GB advertised
- NVIDIA Tesla V100S 32 GB · 32 GB advertised
- NVIDIA H200 141 GB · 141 GB advertised
- NVIDIA GH200 96 GB · 96 GB advertised
- NVIDIA B200 192 GB · 192 GB advertised
- NVIDIA RTX PRO 6000 Blackwell 96 GB · 96 GB advertised
Supported llama.cpp decode per current $1,000
Not ranked here: comparable published MEDIUM-or-higher-confidence llama.cpp decode predictions with a fresh price are unavailable for this workload. Raw Phase 5 estimates may still appear on GPU and comparison pages.
Open the GPU finder · Model page · Value explorer · Workstations · Servers · Buy vs rent