Skip to content
RigCheck

Models / Qwen

Can an RTX 4090 run Qwen3.8 27B?

Dense 27B vision-language model. The most-measured model in the current RigBase import.

Stored parameters
27B
Active parameters
Dense
Q4 total (estimate)
17.6 GB
License
See model card

Model card. Qwen's card lists 27B parameters and a 262,144-token native context.

Compatibility

HardwareQ4Q8FP16
Ryzen AI Max+ 395 128GB
AMD · 128 GB unified
55.3 GB
Fits
estimate · KV at 8k
No measured run
M5 Max 128GB
Apple · 128 GB unified
17.6 GB
Fits
estimate · KV at 8k
Measured · 31.0 tok/s median
29.9 GB
Fits
estimate · KV at 8k
No measured run
55.3 GB
Fits
estimate · KV at 8k
No measured run
Ryzen AI Max+ 395 64GB
AMD · 64 GB unified
17.6 GB
Fits
estimate · KV at 8k
Measured · 12.1 tok/s median
29.9 GB
Fits
estimate · KV at 8k
No measured run
55.3 GB
Tight
estimate · KV at 8k
No measured run
2× RTX 3090
NVIDIA · 48 GB VRAM
29.9 GB
Fits
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
M5 Max 36GB
Apple · 36 GB unified
29.9 GB
Tight
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
RTX 5090
NVIDIA · 32 GB VRAM
29.9 GB
Tight
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
Radeon AI Pro R9700
AMD · 32 GB VRAM
29.9 GB
Tight
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
Intel Arc Pro B70
Intel · 32 GB VRAM
17.6 GB
Fits
estimate · KV at 8k
Measured · 27.8 tok/s median
29.9 GB
Tight
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
M6 32GB
Apple · 32 GB unified
17.6 GB
Fits
estimate · KV at 8k
Measured · 11.2 tok/s median
29.9 GB
Tight
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
RTX 3090
NVIDIA · 24 GB VRAM
29.9 GB
Over
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
RTX 4090
NVIDIA · 24 GB VRAM
29.9 GB
Over
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
TITAN RTX
NVIDIA · 24 GB VRAM
17.6 GB
Fits
estimate · KV at 8k
Measured · 30.2 tok/s median
29.9 GB
Over
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
RTX 5060 Ti 16GB
NVIDIA · 16 GB VRAM
29.9 GB
Over
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
M3 Ultra 512GB
Apple · 512 GB unified
17.6 GB
Fits
estimate · KV at 8k
No measured run
29.9 GB
Fits
estimate · KV at 8k
No measured run
55.3 GB
Fits
estimate · KV at 8k
Measured · 12.5 tok/s median
2× NVIDIA GB10
NVIDIA · 256 GB unified
17.6 GB
Fits
estimate · KV at 8k
No measured run
29.9 GB
Fits
estimate · KV at 8k
No measured run
55.3 GB
Fits
estimate · KV at 8k
No measured run
M3 Ultra 256GB
Apple · 256 GB unified
17.6 GB
Fits
estimate · KV at 8k
No measured run
29.9 GB
Fits
estimate · KV at 8k
No measured run
55.3 GB
Fits
estimate · KV at 8k
No measured run
M5 Ultra 256GB
Apple · 256 GB unified
17.6 GB
Fits
estimate · KV at 8k
No measured run
29.9 GB
Fits
estimate · KV at 8k
No measured run
55.3 GB
Fits
estimate · KV at 8k
No measured run
M2 Ultra 192GB
Apple · 192 GB unified
17.6 GB
Fits
estimate · KV at 8k
No measured run
29.9 GB
Fits
estimate · KV at 8k
No measured run
55.3 GB
Fits
estimate · KV at 8k
No measured run
NVIDIA GB10
NVIDIA · 128 GB unified
17.6 GB
Fits
estimate · KV at 8k
No measured run
29.9 GB
Fits
estimate · KV at 8k
No measured run
55.3 GB
Fits
estimate · KV at 8k
No measured run
M1 Ultra 128GB
Apple · 128 GB unified
17.6 GB
Fits
estimate · KV at 8k
No measured run
29.9 GB
Fits
estimate · KV at 8k
No measured run
55.3 GB
Fits
estimate · KV at 8k
No measured run
4× RTX 3090
NVIDIA · 96 GB VRAM
17.6 GB
Fits
estimate · KV at 8k
No measured run
29.9 GB
Fits
estimate · KV at 8k
No measured run
55.3 GB
Fits
estimate · KV at 8k
No measured run
RTX Pro 6000 Blackwell
NVIDIA · 96 GB VRAM
17.6 GB
Fits
estimate · KV at 8k
No measured run
29.9 GB
Fits
estimate · KV at 8k
No measured run
55.3 GB
Fits
estimate · KV at 8k
No measured run
M5 Ultra 96GB
Apple · 96 GB unified
17.6 GB
Fits
estimate · KV at 8k
No measured run
29.9 GB
Fits
estimate · KV at 8k
No measured run
55.3 GB
Fits
estimate · KV at 8k
No measured run
M4 Pro 48GB
Apple · 48 GB unified
17.6 GB
Fits
estimate · KV at 8k
No measured run
29.9 GB
Fits
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
M4 Max 36GB
Apple · 36 GB unified
17.6 GB
Fits
estimate · KV at 8k
No measured run
29.9 GB
Tight
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
M4 32GB
Apple · 32 GB unified
17.6 GB
Fits
estimate · KV at 8k
No measured run
29.9 GB
Tight
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
Radeon RX 7900 XTX
AMD · 24 GB VRAM
17.6 GB
Fits
estimate · KV at 8k
No measured run
29.9 GB
Over
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
M4 Pro 24GB
Apple · 24 GB unified
17.6 GB
Tight
estimate · KV at 8k
No measured run
29.9 GB
Over
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
M3 Pro 18GB
Apple · 18 GB unified
17.6 GB
Tight
estimate · KV at 8k
No measured run
29.9 GB
Over
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
RTX 4060 Ti 16GB
NVIDIA · 16 GB VRAM
17.6 GB
Over
estimate · KV at 8k
No measured run
29.9 GB
Over
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
RTX 5070 Ti
NVIDIA · 16 GB VRAM
17.6 GB
Over
estimate · KV at 8k
No measured run
29.9 GB
Over
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
M4 16GB
Apple · 16 GB unified
17.6 GB
Over
estimate · KV at 8k
No measured run
29.9 GB
Over
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
RTX 3060 12GB
NVIDIA · 12 GB VRAM
17.6 GB
Over
estimate · KV at 8k
No measured run
29.9 GB
Over
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
RTX 3080 Ti
NVIDIA · 12 GB VRAM
17.6 GB
Over
estimate · KV at 8k
No measured run
29.9 GB
Over
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
RTX 4070 Ti
NVIDIA · 12 GB VRAM
17.6 GB
Over
estimate · KV at 8k
No measured run
29.9 GB
Over
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run
RTX 2060 Super
NVIDIA · 8 GB VRAM
17.6 GB
Over
estimate · KV at 8k
No measured run
29.9 GB
Over
estimate · KV at 8k
No measured run
55.3 GB
Over
estimate · KV at 8k
No measured run

Measured runs

HardwareQuanttok/sRuntimeSource
RTX 5090all-NVFP4 NInfer v3 pack of the stock model (kaushikvira/Qwen3.8-27B-nvfp4full-dflash2-NInfer-v3)400 · speculativeNInfer (ninfer-all fork) + DFlash2 drafter, 7 draft tokens, INT8 KVSource · checked · prefill 11181
RTX 5090all-NVFP4 NInfer v3 pack of the stock model (kaushikvira/Qwen3.8-27B-nvfp4full-dflash2-NInfer-v3)326 · speculativeNInfer (ninfer-all fork) + DFlash2 drafter, 7 draft tokens, INT8 KVSource · checked · prefill 7856
RTX 5090all-NVFP4 NInfer v3 pack of the stock model (kaushikvira/Qwen3.8-27B-nvfp4full-dflash2-NInfer-v3)165 · speculativeNInfer (ninfer-all fork) + DFlash2 drafter, 7 draft tokens, INT8 KVSource · checked · prefill 12908
RTX Pro 6000 BlackwellNVFP4 (W4A16) + MTP (5 draft tokens), 256K context150 · speculativevLLM 0.27.1Source · checked · prefill 3800
RTX 3090Q4_K_M (unsloth), greedy code prompt139.7 · speculativeOpenJet megakernel (MTP, 4 drafts)Source · checked
RTX 5090Q4_K_M, q4_0 KV136.7 · speculativellama.cpp b10448, MTP n=2 (--spec-type draft-mtp)Source · checked
RTX 3090AutoRound W4A16129.8 · speculativevLLM (patched)Source · checked
2× RTX 3090INT4 (RedHat INT4 checkpoint, QEngine layout), fp8 KV127.5 · speculativeQEngine 1.3.2 (own collectives, P2P over NVLink), MTP headSource · checked · prefill 2363
RTX 5090Ollama qwen3.8:27b q4 (17 GB on disk)124Ollama 0.32.13Source · checked
Intel Arc Pro B70GPTQ Int4 (asym)96 · speculativevLLM XPU b70 image + MTP, reasoning promptSource · checked · prefill 2042
2× RTX 3090W4A16 AutoRound + DFlash291.2 · speculativevLLM TP2, 1 streamSource · checked
RTX Pro 6000 BlackwellBF16 + MTP (5 draft tokens), 256K context85 · speculativevLLM 0.20.1Source · checked · prefill 2400
RTX 3090Q4_K_M (unsloth), reasoning mode78.0 · speculativeOpenJet megakernel (MTP)Source · checked
RTX 5090Q4_K_M (EPYC 9534 host)75.2llama.cpp 0.7.0-dev (b11669)Source · checked · prefill 2500
RTX 5090Q4_K_M75.19llama.cppSource · checked · prefill 2500
RTX 5090Q4_K_M, f16 KV73.6llama.cpp b10448, MTP offSource · checked · prefill 3036
RTX 3090Q4_K_M (unsloth), greedy code prompt72.5 · speculativellama.cpp (MTP, 2 drafts)Source · checked
2× RTX 3090AWQ + DSpark69.9 · speculativeSGLang TP2, 1 streamSource · checked
Intel Arc Pro B70GPTQ Int4 (asym)66 · speculativevLLM XPU b70 image + MTP, prose promptSource · checked · prefill 2042
RTX 4090Q4_K_M (16.8 GB, LM Studio catalogue) + the model's own MTP (LM Studio default)62.4 · speculativeLM Studio 2.55.0 (llama.cpp CUDA)Source · checked · prefill 1994
NVIDIA GB10MLX 4-bit + DFlash258.3 · speculativeTensorFold, 1 streamSource · checked · prefill 1633
RTX 3090Unsloth UD-Q4_K_M (nextn/MTP head embedded), f16 KV56.37 · speculativellama.cpp b11382 (Windows, CUDA 13), MTP n=2Source · checked · prefill 1073
2× RTX 3090Q6_K_L (bartowski GGUF)55.4 · speculativellama.cpp b10666, -sm tensor (NCCL)Source · checked · prefill 743
RTX 3090Q4_K_M (unsloth), reasoning mode50.9 · speculativellama.cpp (MTP, 2 drafts)Source · checked
RTX 5060 Ti 16GBUnsloth UD-IQ4_XS + mtp-Qwen3.8-27B-Q4_0 head, q8_0 KV50.2 · speculativellama.cpp 23b0202a, MTP-2Source · checked
TITAN RTXUnsloth UD-Q4_K_XL, q8_0 KV, 65k context (19.5 GiB)49.0 · speculativellama.cpp b11222 (CUDA), llama-server, MTP draft 2Source · checked
2× RTX 3090Q6_K_L (bartowski GGUF)48.2llama.cpp b10666, -sm tensor (NCCL)Source · checked
M5 Ultra 96GBMXFP8 (Ollama qwen3.8:27b-mxfp8, 31 GB)47.2Ollama 0.40.13 (MLX runner)Source · checked · prefill 1376
RTX 5060 Ti 16GBUnsloth UD-IQ4_XS + mtp-Qwen3.8-27B-Q4_0 head, q8_0 KV45.8 · speculativellama.cpp 23b0202a, MTP-1Source · checked
M5 Max 128GB4-bit MLX + Qwen3.8-27B-MTP-4bit drafter45.3 · speculativeoMLX v0.6.0, MTP enabledSource · checked · prefill 782
2× NVIDIA GB10NVFP4 (Unsloth, 4-bit, 23.4 GB on disk)44.7 · speculativevLLM 0.1.dev20003 TP2, MTP n=3, FP8 KV, 351K contextSource · checked · prefill 2034
RTX 4090Q4_K_M43.1llama.cpp build 11803 (a37cc4610)Source · checked · prefill 515
RTX 4090Q4_K_M43.06llama.cpp (df18bee)Source · checked · prefill 429
RTX 4090Q4_K_M43.0llama.cpp build 11803 (a37cc4610)Source · checked · prefill 2015
Radeon AI Pro R9700IQ4_NL + MTP draft 243.0 · speculativellama-server HIP, 4 slots, 131k unified contextSource · checked · prefill 979
RTX 3090Q4_K_M (unsloth), no speculation42.9OpenJet megakernelSource · checked
RTX 4090Q4_K_M42.63llama.cpp (df18bee)Source · checked · prefill 1915
RTX 3090UD-Q4_K_XL (GGUF)40.6 · speculativellama.cpp, tunedSource · checked
2× RTX 3090UD-Q4_K_M (GGUF)40.556llama.cpp b10775, 2 GPU -sm layerSource · checked · prefill 1616
RTX 3090Q4_K_M (unsloth), no speculation40.4llama.cppSource · checked
2× RTX 3090UD-Q4_K_M (GGUF)39.157llama.cpp b10775, 1 GPUSource · checked · prefill 1235
RTX 3090Q4_K_M39llama.cpp serverSource · checked · prefill 1226
RTX 3090UD-Q4_K_XL (GGUF)38.8llama.cpp llama-bench, 1 GPU at 350 WSource · checked · prefill 1027
Ryzen AI Max+ 395 128GBROCmFP4-FAST (julianmb/Qwen-3.8-27B-ROCmFP4-FAST-GGUF), q8_0 KV38.6 · speculativeROCmFPX llama.cpp fork main 120141f1f (HIP), --spec-type draft-mtp, 4 draft tokensSource · checked · prefill 353
RTX 3090Q4_K_M38.6llama.cpp build 11803 (a37cc4610)Source · checked · prefill 483
RTX 3090Unsloth UD-Q4_K_M, f16 KV38.3llama.cpp b11382 (Windows, CUDA 13), MTP off, -np 4Source · checked
RTX 3090llama-bench tg128 (quant per result)38.2–71.6llama.cppSource · checked
RTX 3090Q4_K_M38.1llama.cpp build 11803 (a37cc4610)Source · checked · prefill 1209
M5 Max 36GB4-bit MLX37.9 · speculativeRapid-MLX 0.16.1 (MTP)Source · checked
RTX 3090AutoRound W4A1632.6vLLM (stock)Source · checked
2× RTX 3090Q6_K_L (bartowski GGUF)32.4llama.cpp b10666, -sm layer, out of the boxSource · checked
2× RTX 3090UD-Q4_K_M (GGUF)31.252llama.cpp b10775, 2 GPU -sm tensor + NCCL 2.27.5Source · checked · prefill 169
M5 Max 128GB4-bit MLX31same server, no speculationSource · checked
TITAN RTXQ4_K_M30.2llama.cpp b11222 (CUDA)Source · checked · prefill 768
Radeon AI Pro R9700IQ4_NL29.8llama.cpp HIP b10687 (llama-bench)Source · checked · prefill 1287
Ryzen AI Max+ 395 128GBGSQ-RCO IQ3_S + MTP head (ISTA-DASLab GGUF, 12.12 GB)28.8 · speculativellama.cpp 79e2e74eb (Vulkan)Source · checked
Intel Arc Pro B70Q4_K_M27.8llama.cpp SYCL + neural.download TP1 stack, no speculationSource · checked · prefill 825
RTX 5060 Ti 16GBUnsloth UD-IQ4_XS, q8_0 KV27.6llama.cpp 23b0202a, no speculative decodingSource · checked
Radeon AI Pro R9700IQ4_NL, no speculation27.1llama-server HIP, 4 slots, 131k unified contextSource · checked · prefill 1031
NVIDIA GB10NVFP4 (Unsloth, 4-bit, 23.4 GB on disk)26.7 · speculativevLLM 0.1.dev20003, MTP n=3, FP8 KV, 351K contextSource · checked · prefill 1188
Radeon AI Pro R9700UD-Q4_K_XL26.7llama.cpp HIP b10687 (llama-bench)Source · checked · prefill 1151
M5 Max 36GB4-bit MLX26.0LM Studio (MLX)Source · checked
M5 Max 36GB4-bit MLX25.3oMLX 0.7.0Source · checked
Ryzen AI Max+ 395 64GBQ4_K_M (17.1 GB, Unsloth) + MTP (2 draft tokens)25.0 · speculativellama.cpp b11669 (Vulkan)Source · checked · prefill 215
Radeon AI Pro R9700UD-Q5_K_M24.5llama.cpp HIP b10687 (llama-bench)Source · checked · prefill 946
RTX 5060 Ti 16GBQ4_024llama.cpp 65840ed (CUDA)Source · checked · prefill 849
Radeon AI Pro R9700UD-Q6_K23.2llama.cpp HIP b10687 (llama-bench)Source · checked · prefill 821
NVIDIA GB10+ MTP (k=2)19.0 · speculativevLLM 0.31.0, 128K ctx, 1 userSource · checked
NVIDIA GB10NVFP4 + DSpark18.3 · speculativeSGLang, 1 streamSource · checked
Ryzen AI Max+ 395 128GBQ6_K + MTP (draft-n-max 2)18.01 · speculativellama.cpp Vulkan (RADV)Source · checked · prefill 23.9
Ryzen AI Max+ 395 64GBUD-Q4_K_XL (17.6 GB) + MTP draft N=4, KV q8_016.87 · speculativellama.cpp (ROCm)Source · checked · prefill 299
Ryzen AI Max+ 395 128GBGSQ-RCO IQ3_S + MTP head (ISTA-DASLab GGUF, 12.12 GB)16.1llama.cpp 79e2e74eb (Vulkan)Source · checked
Ryzen AI Max+ 395 128GBUD-Q3_K_XL (Unsloth)15.8llama.cpp (Vulkan)Source · checked
Ryzen AI Max+ 395 128GBQ8_0 + MTP (draft-n-max 3), coding workload, reasoning on15.67 · speculativellama.cpp ROCmSource · checked
M6 32GBSplash package (incoai/Qwen3.8-27B-Splash)15.3 · speculativeLM Studio Splash backendSource · checked
Ryzen AI Max+ 395 128GBQ6_K + MTP (draft-n-max 5)15.16 · speculativellama.cpp Vulkan (RADV)Source · checked · prefill 23.4
Ryzen AI Max+ 395 128GBQ8_0 + MTP (draft-n-max 3), coding workload, reasoning on13.19 · speculativellama.cpp VulkanSource · checked
Ryzen AI Max+ 395 128GBUD-Q4_K_M (15.33 GiB)12.82llama.cpp (Vulkan, commit f5222f25c)Source · checked · prefill 414
Ryzen AI Max+ 395 128GBQ4_K_M12.6llama.cpp/Ollama, ROCmSource · checked
M3 Ultra 512GBBF16, 8,192-token context12.46MLX (mlx_lm.benchmark)Source · checked
Ryzen AI Max+ 395 64GBQ4_K_M (17.1 GB, Unsloth)12.1llama.cpp b11669 (Vulkan)Source · checked · prefill 205
M4 Pro 24GBGSQ-RCO IQ3_S + MTP head (ISTA-DASLab GGUF, 12.12 GB)11.4llama.cpp 0.6.0 (Metal)Source · checked
M6 32GBGGUF Q4_K_M11.2LM Studio (llama.cpp)Source · checked
NVIDIA GB10no speculative decoding11.0vLLM 0.31.0, 128K ctx, 1 userSource · checked
M4 Pro 24GBGSQ-RCO IQ3_S + MTP head (ISTA-DASLab GGUF, 12.12 GB)10.9 · speculativellama.cpp 0.6.0 (Metal)Source · checked
RTX 4060 Ti 16GBllama-bench tg128 (quant per result)10.7–17.6llama.cppSource · checked
Ryzen AI Max+ 395 128GBUD-Q6_K (20.47 GiB)9.71llama.cpp (Vulkan, commit f5222f25c)Source · checked · prefill 401
Ryzen AI Max+ 395 128GBQ6_K, no MTP9.59llama.cpp Vulkan (RADV)Source · checked · prefill 26.0
Ryzen AI Max+ 395 128GBQ8_0, llama-bench tg1287.77llama.cpp VulkanSource · checked · prefill 254
Ryzen AI Max+ 395 128GBQ8_0, llama-bench tg1287.76llama.cpp ROCmSource · checked · prefill 304
Ryzen AI Max+ 395 128GBQ8_0, --reasoning off7.7llama.cpp/Ollama, ROCmSource · checked
Ryzen AI Max+ 395 128GBQ8_0 (27.05 GiB)7.5llama.cpp (Vulkan, commit f5222f25c)Source · checked · prefill 477
RTX 3060 12GBllama-bench tg128 (quant per result)5.2–6.7llama.cppSource · checked