llmrun
silx-aiQuasarForCausalLM

Quasar 10B — Hardware Requirements & GPU Compatibility

Chat

Quasar 10B is a 8.6B-parameter open language model from silx-ai. It supports a context window of up to 2,097,152 tokens. At BF16 it needs about 17.77 GB of VRAM — see which GPUs and Macs can run it below.

8.6B params · dense2097K contextApache 2.01.6K downloads55 likes
A

Runs well on a 24 GB GPU

Runs on24 GB BF16 · ~37 tok/s · 32KMacs from24 GB BF16 · ~12 tok/s

Apache 2.0

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run Quasar 10B

One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.

Recommended

vLLM
one command · needs vLLM installed
$vllm serve silx-ai/Quasar-10B --max-model-len 16384

vLLM repo with its own quantisation

Download: Weights on Hugging Face

install, OS, other quants

Ways to runOther ways

each opens step-by-step

Run in cloud

Runs on RTX 3090 (24 GB) (4 GB headroom) · BF16

Generation speed
~34 tok/s
generation speed
Cost per 1M output tokens
$1.61
per 1M output tokens
Deployon Vast.aiBest valuefrom $0.20/hr (opens in new tab)
Compare GPUs →
or

How Much VRAM Does Quasar 10B Need?

Select a quantization level; the listed hardware adheres to it.

QuantizationVRAM+32KOn RTX 4090 (reference)
17.8 GB21.8 GB16-bit · bfloat — Brain floating point 16 — preferred for trainingFits: 32K context · ~37 tok/s
est.calculated, no published file yet

Selected BF16

Brain floating point 16

19.7GBat 16K context

Weights
17.2 GB
KV cache
2.2 GB
Overhead
0.3 GB

Fits RTX 4090 (reference) · 4.4 GB free · ~37 tok/s

≥50 tok/s needs ≥1.37 TB/s · RTX 5090 ~66

Which Hardware Can Run Quasar 10B?

BF16 · 17.8 GB

Quasar 10B (BF16) requires 17.8 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 24+ GB is recommended. Using the full 2097K context window can add up to 274.6 GB, bringing total usage to 292.4 GB. 8 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 122 devices with unified memory can run Quasar 10B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Mini M5 Pro (24 GB).

By VRAM tier · typical GPUs and Macs at BF16
8–16 GB · RTX 4060, RTX 3060 12GB, RTX 5060 Ti 16GBDoesn't fit — smallest quant needs 19.7 GB8 GB tier →12 GB tier →16 GB tier →
24 GB · RTX 4090referenceBF16 · 19.7 GB · ~37 tok/s · 32K24 GB tier →
Mac 32 GB · Mac Mini M4BF16 · 19.7 GB · ~5 tok/s · 64KMacs →
Mac 64 GB · Mac Studio M4 MaxBF16 · 19.7 GB · ~22 tok/s · 256KMacs →

Alternatives to Quasar 10B

Similar models

6
ModelScoreVRAMtok/sWhy it is similar
Granite Switch 4.1 8BIBM—8.4 GB~111same size
Apertus V1.5 8Bswiss-ai—8.5 GB~111same size
LFM2.5 8B A1BLiquid AI—5.6 GB~335same size
BitCPM CANN 8BOpenBMB—5.6 GB~127same size
Gemma 4 E4BGoogle—5.6 GB~123same size
Qwen3 VL 8B InstructAlibaba—8 GB~112same size

Specifications

Publisher
silx-ai
Parameters
8.6B
Architecture
QuasarForCausalLM
Context Length
2,097,152 tokens
Vocabulary Size
248,320
Release Date
2026-03-09
License
Apache 2.0

Resources

Frequently Asked Questions

How much VRAM does Quasar 10B need?

Quasar 10B requires 17.8 GB of VRAM at BF16. Full 2097K context adds up to 274.6 GB (292.4 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 8.6B × 16 bits ÷ 8 = 17.2 GB

KV Cache + Overhead ≈ 0.6 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 275.2 GB (at full 2097K context)

VRAM usage by quantization

17.8 GB
292.4 GB

Learn more about VRAM estimation →

Can I run Quasar 10B on a Mac?

Yes — Mac Mini M5 Pro (24 GB) and 106 other Macs can run Quasar 10B. Apple Silicon uses unified memory, so the model shares RAM with the system. At BF16 you need at least 17.8 GB of usable unified memory (RAM minus macOS overhead).

Can I run Quasar 10B locally?

Yes — Quasar 10B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Quasar 10B?

At BF16, Quasar 10B can reach ~270 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~37 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = (bandwidth GB/s ÷ model GB) × efficiency

Example: NVIDIA B200 → 8000 ÷ 17.77 × 0.65 = ~293 tok/s

Estimated speed at BF16 (17.8 GB)

~293 tok/s
~37 tok/s
~293 tok/s
~270 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of Quasar 10B?

At BF16, the download is about 17.20 GB.

Which GPUs can run Quasar 10B?

8 consumer GPUs can run Quasar 10B at BF16. Top options include NVIDIA GeForce RTX 5090, AMD Radeon RX 7900 XT, AMD Radeon RX 7900 XTX. 1 GPU has plenty of headroom for comfortable inference.

Which devices can run Quasar 10B?

122 devices with unified memory can run Quasar 10B at BF16 (17.8 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.