aliteq.

Type to search every guide, tool and GPU page.

How much VRAM Gemma 2 27B needs, and what it costs to run

Some links on this page are referral links: we may earn a commission, at no cost to you.

Trying to decide what to buy? The best GPU for Gemma 2 27B ↓ · Got a card already? See what your PC can run →

27.2B params8k max contextgemma

Cheapest way to rent it right now

$0.04/hr

NVIDIA L4 · Q4_K_M · spot · ≈ $9/mo at 8h/day

Live price captured Thu, 19 Nov 2026 21:20:10 GMT.It never changes which card is cheapest.

Memory needed, by quantisation

At 16k context. Weights + KV cache + ~0.8 GB overhead. Memory is in binary GB (1,024³ bytes), the unit GPU memory is sold in; Hugging Face and Ollama list downloads in decimal GB, about 7% more.

QuantWeightsKV cacheTotalQuality
F1650.7 GB5.8 GB57.2 GBFull precision. Reference quality, twice the size of Q8 for no practical gain in most chat use.
Q8_026.9 GB5.8 GB33.5 GBEffectively lossless. Use when VRAM is not the constraint.
Q6_K20.8 GB5.8 GB27.3 GBVery close to Q8 at meaningfully less memory. A safe high-quality pick.
Q5_K_M18.0 GB5.8 GB24.5 GBSmall, hard-to-notice quality loss. Good balance.
Q5_017.6 GB5.8 GB24.1 GBOlder-style 5-bit. Q5_K_M is usually the better pick at the same size.
Q4_K_M15.4 GB5.8 GB21.9 GBThe community default. Best quality-per-gigabyte for most people.
Q4_014.4 GB5.8 GB21.0 GBOlder-style 4-bit, measurably worse than Q4_K_M at a similar size. Avoid unless required.
IQ4_XS13.5 GB5.8 GB20.0 GBNewer 4-bit, smaller than Q4_K_M with comparable quality. Needs a recent llama.cpp.
Q3_K_M12.4 GB5.8 GB18.9 GBNoticeable degradation. Use to fit a larger model that would otherwise not run.
Q2_K10.6 GB5.8 GB17.1 GBHeavy degradation. Almost always better to run a smaller model at Q4_K_M instead.
IQ3_XS10.5 GB5.8 GB17.0 GBAggressive. Usually better than a smaller model at Q4, but test before trusting it.

Cards that can run it

Best quantisation each card fits at 16k context, with live rental price where we track one.

GPUVRAMBest fitUses~tok/sRent from
Apple M3 Ultra512 GBF1657 GB——
Apple M5 Ultra512 GBF1657 GB——
Apple M2 Ultra192 GBF1657 GB——
Apple M5 Max (18-core CPU / 40-core GPU)128 GBF1657 GB——
NVIDIA DGX Spark (GB10 Grace Blackwell)128 GBF1657 GB—$0.309/hrVast.ai
Apple M4 Max (16-core CPU / 40-core GPU)128 GBF1657 GB——
AMD Ryzen AI Max+ 395 (Strix Halo, 128 GB)128 GBF1657 GB——
NVIDIA RTX PRO 6000 Blackwell Workstation Edition96 GBF1657 GB—$0.400/hrVast.ai
Apple M2 Max96 GBF1657 GB——
NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition96 GBF1657 GB—$0.387/hrVast.ai
Apple M4 Pro64 GBF1657 GB——
NVIDIA RTX 6000 Ada Generation48 GBQ8_033 GB—$0.207/hrVast.ai
NVIDIA RTX PRO 5000 Blackwell48 GBQ8_033 GB—$0.189/hrVast.ai
NVIDIA RTX A600048 GBQ8_033 GB—$0.200/hrVast.ai
AMD Radeon PRO W790048 GBQ8_033 GB——
Apple M4 Max (14-core CPU / 32-core GPU)36 GBQ8_0tight33 GB——
Apple M5 Max (18-core CPU / 32-core GPU)36 GBQ8_0tight33 GB——
NVIDIA RTX PRO 4500 Blackwell32 GBQ6_K27 GB—$0.267/hrVast.ai
NVIDIA GeForce RTX 509032 GBQ6_K27 GB—$0.231/hrVast.ai
NVIDIA RTX 5000 Ada Generation32 GBQ6_K27 GB—$0.265/hrVast.ai
NVIDIA GeForce RTX 3090 Ti24 GBQ4_K_Mtight22 GB—$0.173/hrVast.ai
AMD Radeon RX 7900 XTX24 GBQ4_K_Mtight22 GB——
NVIDIA RTX PRO 4000 Blackwell24 GBQ4_K_Mtight22 GB—$0.107/hrVast.ai
NVIDIA RTX A500024 GBQ4_K_Mtight22 GB—$0.076/hrVast.ai
NVIDIA GeForce RTX 309024 GBQ4_K_Mtight22 GB—$0.080/hrVast.ai
NVIDIA GeForce RTX 409024 GBQ4_K_Mtight22 GB—$0.170/hrNovita AI
NVIDIA RTX 4000 Ada Generation20 GBQ3_K_Mtight19 GB—$0.107/hrVast.ai
AMD Radeon RX 7900 XT20 GBQ3_K_Mtight19 GB——

The best GPU for Gemma 2 27B

Ranked from live rental prices and computed VRAM fit, not opinion. Every card is judged at Q4_K_M with a 8K context, so the comparison is fair; a card that could only run a crushed version isn't listed.

Best value

NVIDIA RTX A5000

Best throughput per dollar among cards that fit.

Cheapest that runs it

NVIDIA RTX A5000

Lowest hourly rental that fits it — $0.076/hr, Q4_K_M.

No compromise

—

Most headroom and highest throughput.

Every card that runs it, ranked

Sorted by tokens/sec per rental dollar. Throughput is a bandwidth-derived estimate.

GPUVRAMVRAM used~tok/sCheapest rentaltok/s per $
NVIDIA RTX PRO 6000 Blackwell Workstation Edition96 GB19 GB20%—$0.400—
NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition96 GB19 GB20%—$0.387—
NVIDIA RTX 6000 Ada Generation48 GB19 GB40%—$0.207—
NVIDIA RTX PRO 5000 Blackwell48 GB19 GB40%—$0.189—
NVIDIA RTX A600048 GB19 GB40%—$0.200—
AMD Radeon PRO W790048 GB19 GB40%———
NVIDIA RTX PRO 4500 Blackwell32 GB19 GB59%—$0.267—
NVIDIA GeForce RTX 509032 GB19 GB59%—$0.231—
NVIDIA RTX 5000 Ada Generation32 GB19 GB59%—$0.265—
NVIDIA GeForce RTX 3090 Ti24 GB19 GB79%—$0.173—
AMD Radeon RX 7900 XTX24 GB19 GB79%———
NVIDIA RTX PRO 4000 Blackwell24 GB19 GB79%—$0.107—

The value column ranks renting, because that's what we can price precisely (prices captured Thu, 19 Nov 2026 21:20:10 UTC). Buying a card? The fit and throughput columns are what you need; we don't publish a purchase-price ranking because we don't hold verified street prices. Tokens/sec is an estimate, not a benchmark. Got a card already? See everything your PC can run.

gemma-2-27b

How these numbers are produced

  • Model shape comes from the model's own config.json — 46 layers, 32 attention heads, 16 KV heads.
  • Weight size is parameters × effective bits-per-weight. Those constants are checked against real published quantised file sizes — median error 0.7% across 40 measurements.
  • KV cache is 2 × layers × KV-heads × head-dim × context × 2 bytes. Using KV-heads rather than attention heads is what makes this correct for grouped-query attention; treating a GQA model as multi-head overstates the cache by up to 8×.
  • Tokens/sec is an estimate, not a benchmark. Generation is memory-bandwidth-bound, so this is bandwidth ÷ weight-bytes derated to 75%. Real throughput depends on your runtime, batch size and kernels. We don't run our own hardware tests — see our editorial policy.
  • Some memory-bandwidth figures are not yet independently verified.Where that's the case we show no tokens/sec at all rather than a number we can't stand behind.
  • Rental prices are pulled hourly from provider APIs (last updated Thu, 19 Nov 2026 21:20:10 GMT). Spot/interruptible pricing can change or vanish without notice.

Try it interactively

Drag the context slider and watch the KV cache fill the card

Running Gemma 2 27B: common questions

How much VRAM do you need to run Gemma 2 27B?
At Q4_K_M and at 8k context, Gemma 2 27B needs about 19.0 GB — 15.4 GB of weights plus 2.9 GB of KV cache and ~0.8 GB overhead. So you want a card with at least that much free memory; the KV cache grows if you use longer context.
Can an RTX 4090 run Gemma 2 27B?
Yes. At Q4_K_M it uses about 19.0 GB of the card's 24 GB (79%).
Can an RTX 3090 run Gemma 2 27B?
Yes. At Q4_K_M it uses about 19.0 GB of the card's 24 GB (79%).
Can an RTX 4060 Ti 16GB run Gemma 2 27B?
Not at Q4_K_M — Gemma 2 27B needs more than the card's 16 GB at that quality and 8k context. You'd have to drop to a lower quantisation (worse quality), shorten the context, or use a bigger card or two cards.
What's the cheapest way to run Gemma 2 27B?
Among cards that fit it at Q4_K_M, the cheapest to rent right now is the NVIDIA RTX A5000 at about $0.076/hour. Whether renting or buying is cheaper overall depends on how many hours a day you'll actually use it.

Worth reading before you buy

Other models