aliteq.

Type to search every guide, tool and GPU page.

How much VRAM Inkling 2 Mini (MoE) 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 Inkling 2 Mini (MoE) ↓ · Got a card already? See what your PC can run →

26.4B paramsMoE · 128 experts, 8 active1024k max contextapache-2.0

Cheapest way to rent it right now

$0.04/hr

NVIDIA GeForce RTX 3060 12GB · Q2_K · spot · ≈ $9/mo at 8h/day

Live price captured Thu, 19 Nov 2026 17:20:08 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
F1649.2 GB0.5 GB50.4 GBFull precision. Reference quality, twice the size of Q8 for no practical gain in most chat use.
Q8_026.1 GB0.5 GB27.4 GBEffectively lossless. Use when VRAM is not the constraint.
Q6_K20.2 GB0.5 GB21.4 GBVery close to Q8 at meaningfully less memory. A safe high-quality pick.
Q5_K_M17.4 GB0.5 GB18.7 GBSmall, hard-to-notice quality loss. Good balance.
Q5_017.0 GB0.5 GB18.3 GBOlder-style 5-bit. Q5_K_M is usually the better pick at the same size.
Q4_K_M14.9 GB0.5 GB16.2 GBThe community default. Best quality-per-gigabyte for most people.
Q4_014.0 GB0.5 GB15.3 GBOlder-style 4-bit, measurably worse than Q4_K_M at a similar size. Avoid unless required.
IQ4_XS13.1 GB0.5 GB14.3 GBNewer 4-bit, smaller than Q4_K_M with comparable quality. Needs a recent llama.cpp.
Q3_K_M12.0 GB0.5 GB13.3 GBNoticeable degradation. Use to fit a larger model that would otherwise not run.
Q2_K10.3 GB0.5 GB11.6 GBHeavy degradation. Almost always better to run a smaller model at Q4_K_M instead.
IQ3_XS10.1 GB0.5 GB11.4 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 GBF1650 GB——
Apple M5 Ultra512 GBF1650 GB——
Apple M2 Ultra192 GBF1650 GB——
Apple M5 Max (18-core CPU / 40-core GPU)128 GBF1650 GB——
NVIDIA DGX Spark (GB10 Grace Blackwell)128 GBF1650 GB—$0.240/hrVast.ai
Apple M4 Max (16-core CPU / 40-core GPU)128 GBF1650 GB——
AMD Ryzen AI Max+ 395 (Strix Halo, 128 GB)128 GBF1650 GB——
NVIDIA RTX PRO 6000 Blackwell Workstation Edition96 GBF1650 GB—$0.467/hrVast.ai
Apple M2 Max96 GBF1650 GB——
NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition96 GBF1650 GB—$0.333/hrVast.ai
Apple M4 Pro64 GBF1650 GB——
NVIDIA RTX 6000 Ada Generation48 GBQ8_027 GB—$0.223/hrVast.ai
NVIDIA RTX PRO 5000 Blackwell48 GBQ8_027 GB—$0.202/hrVast.ai
NVIDIA RTX A600048 GBQ8_027 GB—$0.200/hrVast.ai
AMD Radeon PRO W790048 GBQ8_027 GB——
Apple M4 Max (14-core CPU / 32-core GPU)36 GBQ8_027 GB——
Apple M5 Max (18-core CPU / 32-core GPU)36 GBQ8_027 GB——
NVIDIA RTX PRO 4500 Blackwell32 GBQ8_027 GB—$0.200/hrVast.ai
NVIDIA GeForce RTX 509032 GBQ8_027 GB—$0.550/hrLium
NVIDIA RTX 5000 Ada Generation32 GBQ8_027 GB—$0.265/hrVast.ai
NVIDIA GeForce RTX 3090 Ti24 GBQ6_K21 GB—$0.144/hrVast.ai
AMD Radeon RX 7900 XTX24 GBQ6_K21 GB——
NVIDIA RTX PRO 4000 Blackwell24 GBQ6_K21 GB—$0.100/hrVast.ai
NVIDIA RTX A500024 GBQ6_K21 GB—$0.070/hrVast.ai
NVIDIA GeForce RTX 309024 GBQ6_K21 GB—$0.067/hrVast.ai
NVIDIA GeForce RTX 409024 GBQ6_K21 GB—$0.300/hrLium
NVIDIA RTX 4000 Ada Generation20 GBQ5_K_Mtight19 GB—$0.093/hrVast.ai
AMD Radeon RX 7900 XT20 GBQ5_K_Mtight19 GB——
AMD Radeon RX 9070 XT16 GBQ4_0tight15 GB——
NVIDIA GeForce RTX 4060 Ti 16GB16 GBQ4_0tight15 GB—$0.053/hrVast.ai
NVIDIA GeForce RTX 5070 Ti16 GBQ4_0tight15 GB—$0.130/hrVast.ai
NVIDIA GeForce RTX 5060 Ti 16GB16 GBQ4_0tight15 GB—$0.052/hrVast.ai
NVIDIA GeForce RTX 4070 Ti SUPER16 GBQ4_0tight15 GB—$0.093/hrVast.ai
NVIDIA GeForce RTX 408016 GBQ4_0tight15 GB—$0.120/hrVast.ai
NVIDIA GeForce RTX 4080 SUPER16 GBQ4_0tight15 GB—$0.160/hrVast.ai
NVIDIA RTX A400016 GBQ4_0tight15 GB—$0.053/hrVast.ai
AMD Radeon RX 907016 GBQ4_0tight15 GB——
AMD Radeon RX 9060 XT 16GB16 GBQ4_0tight15 GB——
NVIDIA GeForce RTX 508016 GBQ4_0tight15 GB—$0.133/hrVast.ai
NVIDIA GeForce RTX 4070 SUPER12 GBQ2_Ktight12 GB—$0.087/hrVast.ai
NVIDIA GeForce RTX 4070 Ti12 GBQ2_Ktight12 GB—$0.133/hrVast.ai
NVIDIA GeForce RTX 507012 GBQ2_Ktight12 GB—$0.097/hrVast.ai
NVIDIA GeForce RTX 3060 12GB12 GBQ2_Ktight12 GB—$0.036/hrVast.ai
NVIDIA GeForce RTX 3080 Ti12 GBQ2_Ktight12 GB—$0.078/hrVast.ai
NVIDIA GeForce RTX 407012 GBQ2_Ktight12 GB—$0.075/hrVast.ai
NVIDIA GeForce RTX 3080 12GB12 GBQ2_Ktight12 GB——

The best GPU for Inkling 2 Mini (MoE)

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 GeForce RTX 3090

Best throughput per dollar among cards that fit.

Cheapest that runs it

NVIDIA GeForce RTX 3090

Lowest hourly rental that fits it — $0.067/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 GB16 GB17%—$0.467—
NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition96 GB16 GB17%—$0.333—
NVIDIA RTX 6000 Ada Generation48 GB16 GB33%—$0.223—
NVIDIA RTX PRO 5000 Blackwell48 GB16 GB33%—$0.202—
NVIDIA RTX A600048 GB16 GB33%—$0.200—
AMD Radeon PRO W790048 GB16 GB33%———
NVIDIA RTX PRO 4500 Blackwell32 GB16 GB50%—$0.200—
NVIDIA GeForce RTX 509032 GB16 GB50%—$0.550—
NVIDIA RTX 5000 Ada Generation32 GB16 GB50%—$0.265—
NVIDIA GeForce RTX 3090 Ti24 GB16 GB66%—$0.144—
AMD Radeon RX 7900 XTX24 GB16 GB66%———
NVIDIA RTX PRO 4000 Blackwell24 GB16 GB66%—$0.100—

The value column ranks renting, because that's what we can price precisely (prices captured Thu, 19 Nov 2026 17:20:08 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.

inkling-2-mini

How these numbers are produced

  • Model shape comes from the model's own config.json — 8 layers, 24 attention heads, 8 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.
  • This is a mixture-of-experts model. Generation reads only the experts active for each token (8 of 128), so throughput is scored on ~3.4B active parameters, not the full 26B. Memory is the opposite: every expert must still be resident in VRAM, so the figures above use the full weight set.Looped hybrid transformer: 24 stored layers, and the middle 12 (layers 6 to 17) run three times with shared weights, so every token goes through 48 layer passes and each pass computes its own keys and values. Attention mixes local and global layers: 20 of the 24 layers use a 1,024-token sliding window, only layers 5, 11, 17 and 23 are global. Global passes = layer 5 once + layers 11 and 17 three times each + layer 23 once = 8, so the KV figures here count 8 full-context caches (8 KV heads x 128); the 40 sliding-window passes add a fixed ~0.16 GB at any context. A plain 24-layer model with the same pattern would hold 4, so the loop doubles the KV cache. Model card: 26.4B total, 3.4B active (5.4B effective with the three passes), 128 experts with 8 active, up to 1M context, text and image input. The official checkpoint is mixed NVFP4 + MXFP8 at about 20.2 GB; GGUF builds go below that (Unsloth UD-Q2_K_XL 9.1 GB, UD-Q4_K_M 16.6 GB), so this page keeps the full quant ladder. Params = the HF safetensors total. HF config.json and model card read Nov 21, 2026.
  • 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 17:20:08 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 Inkling 2 Mini (MoE): common questions

How much VRAM do you need to run Inkling 2 Mini (MoE)?
At Q4_K_M and at 8k context, Inkling 2 Mini (MoE) needs about 15.9 GB — 14.9 GB of weights plus 0.3 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 Inkling 2 Mini (MoE)?
Yes. At Q4_K_M it uses about 15.9 GB of the card's 24 GB (66%).
Can an RTX 3090 run Inkling 2 Mini (MoE)?
Yes. At Q4_K_M it uses about 15.9 GB of the card's 24 GB (66%).
Can an RTX 4060 Ti 16GB run Inkling 2 Mini (MoE)?
Yes. At Q4_K_M it uses about 15.9 GB of the card's 16 GB (100%), which is tight — leave headroom for your desktop.
What's the cheapest way to run Inkling 2 Mini (MoE)?
Among cards that fit it at Q4_K_M, the cheapest to rent right now is the NVIDIA GeForce RTX 3090 at about $0.067/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