AMD Instinct MI325X: what runs best?

RTX 4090 RTX 5090 RTX 3090 RTX 4060 Ti (16 GB)RTX 5070 Ti Radeon RX 7900 XTX M4 Max 128 GB M3 Ultra 512 GB DGX Spark (128 GB)All devices

AMD · GPU

AMD Instinct MI325X

Memory256 GB HBM3EBandwidth6,000 GB/s vendor specPlussystem RAM for offload (set below)

GPUs

Instinct MI325X256 GB · 6,000 GB/s1

Custom GPU

System RAM

InstalledGBTypeSpeedMT/sChannels× 64-bit

≈ 89.6 GB/s theoretical ceiling (MT/s × 8 bytes × channels). DDR5’s two 32-bit subchannels count as one channel.

CPU

PlatformThreads for inference

Used for CPU layers/experts and the -t flag. A reference platform fills its maximum channels and speed; set what you actually populated.

Advanced: reserves, bandwidth, workload, models

Available VRAM per GPUGiB Runtime allowance per deviceGiB Spare margin per deviceGiB Unified memory for models% RAM kept for OS and appsGiB Measured RAM bandwidthGB/s GPU linkLanes Prompt length (first token)tokens Custom contexttokens Parallel sequences

Allow disk-streamed experts (documented cases)Include base checkpointsInclude specialistsInclude community & quant-trained variantsInclude models with unverified release dates

Defaults are editorial planning choices, not measurements: 1 GiB runtime + 0.5 GiB margin per device, 75% of unified memory, 8 GiB RAM for the OS, a vendor “24 GB” treated as 24 GiB (enter your driver’s available VRAM to override), 5 t/s minimum speed (research editorial default). Memory ledger

Instinct MI325X64 GB DDR5-5600 ×2 ch · 56.0 GiB usable · 89.6 GB/s theoretical8K contextF16 KVany backend≥ 5 t/s

Best plans Instinct MI325X

Best quality at ≥ 5 t/s

Qwen3.8-Flash-Next

39.8 AA index single mode

Q8_0llama.cpp ROCmFull GPU

~205–593 t/s Estimated extrapolated

Speed extrapolated, not measured on this setup.

Faster tradeoff

K2-Horizon-MoVA-36B-A4B

25.3 AA index single mode

Q8_0llama.cpp ROCmFull GPU

~217–626 t/s Estimated extrapolated

Next point on the quality/speed front; then Qwen3.6-35B-A3B.

Possibly better · memory not established

GLM-5.3-Flash

41.8 AA index single mode

native fp8vLLMGPU streams weights from RAM · 52 GiB in RAM

≤ 9.9 t/s Ceiling

Only a bandwidth ceiling (≤ 9.9 t/s) is known: this backend or placement has no calibration. KV memory not established.

WeightsKV cacheReserveFreeFits* = a state term unsizedMeasuredlinks to its sourceEstimatedcalibrated or extrapolated range, ~Ceilingupper bound, ≤Unknownnever zero

Meets your 5 t/s floor 47

Best plan per model that reaches your floor by a matching measurement, a calibrated range, or an extrapolated estimate (labelled). Sorted by the selected quality metric, then speed.

1
Qwen3.8-Flash-Next
Qwen/Qwen3.8-Flash-Next
2026-08-26180B MoE · 6B active
AA index39.8single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits*
GPU
153 / 256 GiB
Generate~205–593 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
2
Qwen3.8-27B
Qwen/Qwen3.8-27B
2026-08-1427B dense
AA index33.7Xhigh
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits*
GPU
28.6 / 256 GiB
Generate~100–145 t/sEstimatedmeets 5 t/s
First token · 4K prompt~411 ms–927 msEstimated
3
MiniMax-M3
MiniMaxAI/MiniMax-M3
2026-06 (day unknown)428B MoE · 23B active
AA index29.2single mode
Plan & memory
Q4_K_Mllama.cpp ROCmFull GPUFits
GPU
249 / 256 GiB
Generate~77–222 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
4
Nex-N2-Pro
nex-agi/Nex-N2-Pro
2026-06 (day unknown)397B MoE · 17B active
AA index28.2 est.Based on Qwen3.5-3
Plan & memory
Q4_K_Mllama.cpp ROCmFull GPUFits*
GPU
227 / 256 GiB
Generate~108–313 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
5
NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16Specialist
nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16
2026-10-15120B MoE · 12B active
AA index26.5single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits*
GPU
121 / 256 GiB
Generate~88–254 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
6
Inkling-Small
thinkingmachines/Inkling-Small
2026-07-30276B MoE · 12B active
AA index25.7single mode
Plan & memory
Q5_K_Mllama.cpp ROCmFull GPUFits
GPU
185 / 256 GiB
Generate~123–355 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
7
K2-Horizon-MoVA-36B-A4B
IFM/K2-Horizon-MoVA-36B-A4B
2026-09-0336B MoE · 4B active
AA index25.3single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
40.1 / 256 GiB
Generate~217–626 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
8
Hy3
tencent/Hy3
2026-07-06295B MoE · 21B active
AA index25.3single mode
Plan & memory
Q4_K_Mllama.cpp ROCmFull GPUFits
GPU
174 / 256 GiB
Generate~79–229 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
9
NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16
nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16
2026-10-1574B MoE · 9B active
AA index21.4single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits*
GPU
75.0 / 256 GiB
Generate~118–340 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
10
K2-Horizon-7B
IFM/K2-Horizon-7B
2026-09-039B tensors · dense
AA index20.6single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
11.5 / 256 GiB
Generate~284–411 t/sEstimatedmeets 5 t/s
First token · 4K prompt~137 ms–309 msEstimated
11
Qwen3.5-397B-A17B
Qwen/Qwen3.5-397B-A17B
2026-02-16397B MoE · 17B active
AA index18.4Reasoning
Plan & memory
Q4_K_Mllama.cpp ROCmFull GPUFits*
GPU
229 / 256 GiB
Generate~107–310 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
12
Qwen3.6-35B-A3B
Qwen/Qwen3.6-35B-A3B
2026-04-1635B MoE · 3B active
AA index18.2Reasoning
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits*
GPU
36.0 / 256 GiB
Generate~349–1,009 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
13
Muse-Glimmer-30B
meta-models/Muse-Glimmer-30B
2026-08-1030B dense
AA index17.5High
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
29.3 / 256 GiB
Generate~97–141 t/sEstimatedmeets 5 t/s
First token · 4K prompt~450 ms–1.0 sEstimated
14
K2-Horizon-3.7B
IFM/K2-Horizon-3.7B
2026-09-035.1B tensors · dense
AA index15.6single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
7.6 / 256 GiB
Generate~483–699 t/sEstimatedmeets 5 t/s
First token · 4K prompt~77 ms–174 msEstimated
15
Qwen3.5-122B-A10B
Qwen/Qwen3.5-122B-A10B
2026-02-24122B MoE · 10B active
AA index15.6Reasoning
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits*
GPU
125 / 256 GiB
Generate~103–299 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
16
Mistral-Medium-3.5-128B
mistralai/Mistral-Medium-3.5-128B
2026-04-28128B dense
AA index14.2single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
128 / 256 GiB
Generate~22–31 t/sEstimatedmeets 5 t/s
First token · 4K prompt~1.9 s–4.4 sEstimated
17
EXAONE-4.5-33B
LGAI-EXAONE/EXAONE-4.5-33B
2026-04-0933B dense
AA index13.2 est.Reasoning
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
35.5 / 256 GiB
Generate~80–116 t/sEstimatedmeets 5 t/s
First token · 4K prompt~502 ms–1.1 sEstimated
18
Qwen3.5-4B
Qwen/Qwen3.5-4B
2026-03-024B dense
AA index13.1 est.Reasoning
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits*
GPU
6.1 / 256 GiB
Generate~609–880 t/sEstimatedmeets 5 t/s
First token · 4K prompt~61 ms–137 msEstimated
19
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16
nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16
2026-08-1130B MoE · 3B active
AA index12.9single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits*
GPU
32.8 / 256 GiB
Generate~335–968 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
20
granite-4.2-30b
ibm-granite/granite-4.2-30b
2026-08-2530B dense
AA index12.7single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
32.5 / 256 GiB
Generate~90–130 t/sEstimatedmeets 5 t/s
First token · 4K prompt~456 ms–1.0 sEstimated
21
MiniCPM5-2B
openbmb/MiniCPM5-2B
2026-09-072.5B dense
AA index12.5single mode
Plan & memory
FP16llama.cpp ROCmFull GPUFits
GPU
6.5 / 256 GiB
Generate~555–802 t/sEstimatedmeets 5 t/s
First token · 4K prompt~38 ms–86 msEstimated
22
Hypernova-60B-2605
MultiverseComputingCAI/Hypernova-60B-2605
2026-05-0660B MoE · 4.8B active
Also: HyperNova-60B (listed with its dated update); Hypernova-60B-2602 (dated update, not separately evaluated)
AA index11.7 est.High
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
59.9 / 256 GiB
Generate~221–638 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
23
Nemotron-Cascade-2-30B-A3B
nvidia/Nemotron-Cascade-2-30B-A3B
2026-03-1930B MoE · 3B active
AA index11.7 est.single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits*
GPU
32.8 / 256 GiB
Generate~335–968 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
24
Qwen3.5-9B
Qwen/Qwen3.5-9B
2026-03-029B dense
AA index11.2Reasoning
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits*
GPU
10.9 / 256 GiB
Generate~291–421 t/sEstimatedmeets 5 t/s
First token · 4K prompt~137 ms–309 msEstimated
25
granite-4.2-8b
ibm-granite/granite-4.2-8b
2026-08-258B dense
AA index11.1single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
11.5 / 256 GiB
Generate~289–418 t/sEstimatedmeets 5 t/s
First token · 4K prompt~122 ms–275 msEstimated
26
G9v3-3B
ai9stars/G9v3-3B
2026-07 (day unknown)3B dense
AA index10.8 est.single mode
Plan & memory
FP16llama.cpp ROCmFull GPUFits
GPU
7.5 / 256 GiB
Generate~467–675 t/sEstimatedmeets 5 t/s
First token · 4K prompt~46 ms–103 msEstimated
27
Trinity-Large-Thinking
arcee-ai/Trinity-Large-Thinking
2026-04-01398B MoE · 13B active
AA index10.8single mode
Plan & memory
Q4_K_Mllama.cpp ROCmFull GPUFits
GPU
228 / 256 GiB
Generate~127–368 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
28
granite-4.2-3b
ibm-granite/granite-4.2-3b
2026-08-253B dense
AA index9.1single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
5.8 / 256 GiB
Generate~685–990 t/sEstimatedmeets 5 t/s
First token · 4K prompt~46 ms–103 msEstimated
29
Tri-21B-Think
trillionlabs/Tri-21B-Think
2026-02 (day unknown)21B dense
AA index9.0 est.single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
23.6 / 256 GiB
Generate~127–183 t/sEstimatedmeets 5 t/s
First token · 4K prompt~315 ms–711 msEstimated
30
LFM2.5-2.6B
LiquidAI/LFM2.5-2.6B
2026-08-042.7B dense
AA index8.4 est.single mode
Plan & memory
FP16llama.cpp ROCmFull GPUFits*
GPU
6.7 / 256 GiB
Generate~529–765 t/sEstimatedmeets 5 t/s
First token · 4K prompt~41 ms–92 msEstimated
31
LFM2.5-8B-A1B
LiquidAI/LFM2.5-8B-A1B
2026-05-288.3B MoE · 1.5B active
AA index7.2 est.single mode
Plan & memory
FP16llama.cpp ROCmFull GPUFits*
GPU
17.4 / 256 GiB
Generate~364–1,052 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
32
Olmo-3.1-32B-ThinkSpecialist
allenai/Olmo-3.1-32B-Think
2025-12-1232B tensors · dense
AA index7.1 est.single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
34.7 / 256 GiB
Generate~82–118 t/sEstimatedmeets 5 t/s
First token · 4K prompt~490 ms–1.1 sEstimated
33
Qwen3.5-2B
Qwen/Qwen3.5-2B
2026-03-022B dense
AA index6.9 est.Reasoning
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits*
GPU
3.5 / 256 GiB
Generate~1,359–1,964 t/sEstimatedmeets 5 t/s
First token · 4K prompt~30 ms–69 msEstimated
34
Olmo-3.1-32B-Instruct
allenai/Olmo-3.1-32B-Instruct
2025-12-1232B tensors · dense
AA index6.5 est.single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
34.7 / 256 GiB
Generate~82–118 t/sEstimatedmeets 5 t/s
First token · 4K prompt~490 ms–1.1 sEstimated
35
Qwen3.5-0.8B
Qwen/Qwen3.5-0.8B
2026-03-020.8B dense
AA index6.1 est.Reasoning
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits*
GPU
2.4 / 256 GiB
Generate~3,274–4,734 t/sEstimatedmeets 5 t/s
First token · 4K prompt~12 ms–27 msEstimated
36
granite-4.0-h-small
ibm-granite/granite-4.0-h-small
2025-10-0232B tensors · MoE
AA index6.0 est.single mode
Plan & memory
FP16llama.cpp ROCmFull GPUFits*
GPU
61.6 / 256 GiB
Generate~64–185 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
37
Phi-4Older
microsoft/phi-4
2024-12-1215B tensors · dense
AA index5.9 est.single mode
Plan & memory
FP16llama.cpp ROCmFull GPUFits
GPU
30.4 / 256 GiB
Generate~96–139 t/sEstimatedmeets 5 t/s
First token · 4K prompt~223 ms–503 msEstimated
38
Olmo-3-7B-ThinkSpecialist
allenai/Olmo-3-7B-Think
2025-11-207.3B tensors · dense
AA index5.6 est.single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
11.2 / 256 GiB
Generate~292–422 t/sEstimatedmeets 5 t/s
First token · 4K prompt~111 ms–250 msEstimated
39
Olmo-3-7B-Instruct
allenai/Olmo-3-7B-Instruct
2025-11-207.3B tensors · dense
AA index5.2 est.single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
11.2 / 256 GiB
Generate~292–422 t/sEstimatedmeets 5 t/s
First token · 4K prompt~111 ms–250 msEstimated
40
LFM2.5-1.2B-ThinkingSpecialist
LiquidAI/LFM2.5-1.2B-Thinking
2026-01-201.2B tensors · dense
AA index5.2 est.single mode
Plan & memory
FP16llama.cpp ROCmFull GPUFits*
GPU
3.8 / 256 GiB
Generate~1,209–1,748 t/sEstimatedmeets 5 t/s
First token · 4K prompt~18 ms–40 msEstimated
41
LFM2.5-1.2B-Instruct
LiquidAI/LFM2.5-1.2B-Instruct
2026-01-051.2B tensors · dense
AA index5.2 est.single mode
Plan & memory
FP16llama.cpp ROCmFull GPUFits*
GPU
3.8 / 256 GiB
Generate~1,209–1,748 t/sEstimatedmeets 5 t/s
First token · 4K prompt~18 ms–40 msEstimated
42
Apertus-70B-Instruct-2509
swiss-ai/Apertus-70B-Instruct-2509
2025-09-0271B tensors · dense
AA index5.1 est.single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
73.9 / 256 GiB
Generate~38–55 t/sEstimatedmeets 5 t/s
First token · 4K prompt~1.1 s–2.4 sEstimated
43
Molmo2-8BSpecialist
allenai/Molmo2-8B
2025-12-118.7B tensors · dense
AA index5.0 est.single mode
Plan & memory
FP16llama.cpp ROCmFull GPUFits
GPU
17.9 / 256 GiB
Generate~170–246 t/sEstimatedmeets 5 t/s
First token · 4K prompt~132 ms–297 msEstimated
44
LFM2.5-VL-1.6BSpecialist
LiquidAI/LFM2.5-VL-1.6B
2026-01-051.6B tensors · dense
AA index4.8 est.single mode
Plan & memory
FP16llama.cpp ROCmFull GPUFits*
GPU
3.8 / 256 GiB
Generate~1,209–1,748 t/sEstimatedmeets 5 t/s
First token · 4K prompt~24 ms–55 msEstimated
45
Apertus-8B-Instruct-2509
swiss-ai/Apertus-8B-Instruct-2509
2025-09-028.1B tensors · dense
AA index4.8 est.single mode
Plan & memory
FP16llama.cpp ROCmFull GPUFits
GPU
17.5 / 256 GiB
Generate~174–251 t/sEstimatedmeets 5 t/s
First token · 4K prompt~122 ms–276 msEstimated
46
NVIDIA-Nemotron-3-Nano-4B-BF16
nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16
2026-03-164B dense
AA index4.8single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits*
GPU
5.6 / 256 GiB
Generate~673–973 t/sEstimatedmeets 5 t/s
First token · 4K prompt~60 ms–136 msEstimated
47
K2-Horizon-0.9B
IFM/K2-Horizon-0.9B
2026-09-031.1B tensors · dense
AA index3.0single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
3.0 / 256 GiB
Generate~2,092–3,025 t/sEstimatedmeets 5 t/s
First token · 4K prompt~16 ms–37 msEstimated

+ 65 models without a AA index score

LFM2.5-230MFP16 · full gpu · ~6,569 t/s
LFM2.5-350MFP16 · full gpu · ~4,417 t/s
LittleLambFP16 · full gpu · ~3,167 t/s
LittleLamb-ToolCallingFP16 · full gpu · ~3,167 t/s
Jais-3-0.9BQ8_0 · full gpu · ~2,432 t/s
Qwen3-0.6BFP16 · full gpu · ~1,700 t/s
granite-4.0-h-tinyQ8_0 · full gpu · ~1,448 t/s
LFM2.5-1.2B-JPFP16 · full gpu · ~1,406 t/s
granite-4.1-3bQ8_0 · full gpu · ~851 t/s
Mellum2.1-12B-A2.5B-ThinkingQ8_0 · full gpu · ~826 t/s
Qwen3-1.7BFP16 · full gpu · ~741 t/s
Agents-A1-4BQ8_0 · full gpu · ~729 t/s
FrogNano-4B-2609Q8_0 · full gpu · ~707 t/s
Qwen3.5-35B-A3BQ8_0 · full gpu · ~694 t/s
Qwen3-4BQ8_0 · full gpu · ~689 t/s
AgentCPM-ReportQ4_K_M · full gpu · ~660 t/s
Kolibri-1Q8_0 · full gpu · ~605 t/s
Qwen3-30B-A3BQ8_0 · full gpu · ~581 t/s
Qwen3-Coder-30B-A3B-InstructQ8_0 · full gpu · ~581 t/s
LFM2.5-24B-A2BFP16 · full gpu · ~566 t/s
Jais-3-3.7BQ8_0 · full gpu · ~562 t/s
MiniCPM-V-4.7-35B-A3BFP16 · full gpu · ~383 t/s
PaCoRe-8BQ8_0 · full gpu · ~361 t/s
Qwen3-8BQ8_0 · full gpu · ~361 t/s
AgentCPM-ExploreFP16 · full gpu · ~357 t/s
MiMo-V2.6-Distill-Qwen-9BQ8_0 · full gpu · ~348 t/s
Ornith-1.5-9BQ8_0 · full gpu · ~348 t/s
granite-4.1-8bQ8_0 · full gpu · ~336 t/s
Jais-3-7BQ8_0 · full gpu · ~331 t/s
Mistral 7B Instruct v0.3FP16 · full gpu · ~224 t/s
IQuest-Coder-V1-7B-Base-Stage1FP16 · full gpu · ~218 t/s
IQuest-Coder-V1-7B-BaseFP16 · full gpu · ~218 t/s
IQuest-Coder-V1-7B-InstructFP16 · full gpu · ~218 t/s
IQuest-Coder-V1-7B-ThinkingFP16 · full gpu · ~218 t/s
MagenticBrainQ8_0 · full gpu · ~206 t/s
DeepSeek-R1-Distill-Llama-8BFP16 · full gpu · ~203 t/s
RLVR-8B-0926FP16 · full gpu · ~198 t/s
Nemotron-Cascade-8BFP16 · full gpu · ~198 t/s
NVIDIA-Nemotron-Nano-9B-v2-JapaneseFP16 · full gpu · ~188 t/s
MiniMax-M2.7Q8_0 · full gpu · ~181 t/s
Mixtral 8x7B Instruct v0.1Q8_0 · full gpu · ~162 t/s
Laguna-S-2.1FP16 · full gpu · ~142 t/s
Mistral NeMo 12B Instruct (2407)FP16 · full gpu · ~134 t/s
Mistral Small 3.2 24B Instruct (2506)Q8_0 · full gpu · ~131 t/s
GLM-4.5-AirQ8_0 · full gpu · ~122 t/s
Qwen3.6-27BQ8_0 · full gpu · ~117 t/s
IQuest-Coder-V1-14B-Base-Stage1FP16 · full gpu · ~115 t/s
IQuest-Coder-V1-14B-BaseFP16 · full gpu · ~115 t/s
IQuest-Coder-V1-14B-InstructFP16 · full gpu · ~115 t/s
IQuest-Coder-V1-14B-ThinkingFP16 · full gpu · ~115 t/s
Qwen3.5-27BQ8_0 · full gpu · ~115 t/s
Qwen3-14BFP16 · full gpu · ~111 t/s
FrogMini-14B-2510FP16 · full gpu · ~111 t/s
granite-4.1-30bQ8_0 · full gpu · ~106 t/s
Olmo-3-32B-ThinkQ8_0 · full gpu · ~95 t/s
Qwen3-235B-A22BQ8_0 · full gpu · ~95 t/s
Qwen3-32BQ8_0 · full gpu · ~94 t/s
Qwen3-Nemotron-32B-RLBFFQ8_0 · full gpu · ~94 t/s
Qwen3-VL-235B-A22B-ThinkingQ8_0 · full gpu · ~93 t/s
Qwen3-VL-235B-A22B-InstructQ8_0 · full gpu · ~93 t/s
Jais-3-32BQ8_0 · full gpu · ~88 t/s
IQuest-Coder-V1-40B-ThinkingQ8_0 · full gpu · ~77 t/s
FrogBoss-32B-2510FP16 · full gpu · ~51 t/s
DeepSeek-R1-Distill-Llama-70BFP16 · full gpu · ~24 t/s
Qwen3-Coder-480B-A35B-InstructQ4_K_M · gpu + cpu experts · ~8.0 t/s

Speed not established 7

Feasible plans with only a bandwidth ceiling (backends or placements without calibration). Check the evidence before relying on them.

1
G9v3-39A5B (preview)
ai9stars/G9v3-39A5B
2026-08 (day unknown)39B MoE · 5B active
AA index21.8 est.single mode
Plan & memory
native F32/BF16vLLMFull GPUFits
GPU
74.4 / 256 GiB
Generate≤591 t/sCeilingnot established
First token · 4K prompt—Unknown
2
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16Specialist
nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16
2026-04-2830B MoE · 3B active
AA index10.3 est.single mode
Plan & memory
native F32/BF16vLLMFull GPUFits*
GPU
63.0 / 256 GiB
Generate≤905 t/sCeilingnot established
First token · 4K prompt—Unknown
3
North-Mini-Code-1.0Specialist
CohereLabs/North-Mini-Code-1.0
2026-06-0930B MoE · 3B active
AA index9.9single mode
Plan & memory
native BF16vLLMFull GPUFits
GPU
58.8 / 256 GiB
Generate≤922 t/sCeilingnot established
First token · 4K prompt—Unknown
4
MiniCPM5-1B
openbmb/MiniCPM5-1B
2026-05-191.1B dense
AA index8.8 est.Reasoning
Plan & memory
native BF16vLLMFull GPUFits
GPU
3.7 / 256 GiB
Generate≤2,653 t/sCeilingnot established
First token · 4K prompt—Unknown
5
Nanbeige4.1-3B
Nanbeige/Nanbeige4.1-3B
2026-02 (day unknown)3B dense
AA index8.4 est.single mode
Plan & memory
native BF16vLLMFull GPUFits
GPU
9.3 / 256 GiB
Generate≤737 t/sCeilingnot established
First token · 4K prompt—Unknown
6
Step3-VL-10B
stepfun-ai/Step3-VL-10B
2026-01 (day unknown)10B dense
AA index7.7 est.single mode
Plan & memory
native BF16vLLMFull GPUFits
GPU
21.6 / 256 GiB
Generate≤286 t/sCeilingnot established
First token · 4K prompt—Unknown
7
NVIDIA-Nemotron-Nano-12B-v2-VL-BF16Specialist
nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16
2025-10-2813B tensors · dense
AA index7.5 est.Reasoning
Plan & memory
native F32/BF16vLLMFull GPUFits*
GPU
26.2 / 256 GiB
Generate≤227 t/sCeilingnot established
First token · 4K prompt—Unknown

+ 9 models without a AA index score

Agents-A1native BF16 · full gpu · ≤ 761 t/s
IQuest-Coder-V1-40B-Loop-Thinkingnative BF16 · full gpu · ≤ 74 t/s
K2-Horizon-32Bnative BF16 · full gpu · ≤ 85 t/s
Ming-flash-omni-2.0native BF16 · full gpu · ≤ 393 t/s
MiniCPM-SALAnative BF16 · full gpu · ≤ 314 t/s
Mixtral 8x22B Instruct v0.1AWQ · full gpu · ≤ 280 t/s
NVIDIA-Nemotron-3-Super-120B-A12B-BF16native F32/BF16 · full gpu · ≤ 242 t/s
Penguin-VL-2Bnative BF16 · full gpu · ≤ 1,249 t/s
Penguin-VL-8Bnative BF16 · full gpu · ≤ 335 t/s

Memory not fully resolved 22

The weights fit, but the KV cache for this attention type (or the chosen cache layout) is not established, so a fit cannot be guaranteed.

1
GLM-5.3-Flash
zai-org/GLM-5.3-Flash
2026-08-26320B MoE · 18B active
AA index41.8single mode
Plan & memory
native fp8vLLMGPU streams weights from RAM · 52 GiB in RAMKV unknown
GPU
256+ / 256 GiB
RAM
51.3 / 56.0 GiB
Generate≤9.9 t/sCeilingnot established
First token · 4K prompt—Unknown
2
MiMo-V2.6-Flash-MOPD
XiaomiMiMo/MiMo-V2.6-Flash-MOPD
2026-09-27309B MoE · 15B active
Also: MiMo-V2.6-Flash-RL (shares this AA evaluation)
AA index37.9single mode
Plan & memory
native fp8vLLMFull GPUKV unknown
GPU
163+ / 256 GiB
Generate≤715 t/sCeilingnot established
First token · 4K prompt—Unknown
3
Ling-3.0-flash-VL
inclusionAI/Ling-3.0-flash-VL
2026-09-09124B MoE · 5.5B active
AA index24.6single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUKV unknown
GPU
125+ / 256 GiB
Generate~193–558 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
4
Ling-3.0-flash-Fin
inclusionAI/Ling-3.0-flash-Fin
2026 (month unknown)124B MoE · 5.1B active
AA index22.6single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUKV unknown
GPU
128+ / 256 GiB
Generate~203–587 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
5
gemma-4-26B-A4B-it
google/gemma-4-26B-A4B-it
2026-04-0225B MoE · 3.8B active
AA index16.7 est.Reasoning
Plan & memory
Q8_0llama.cpp ROCmFull GPUKV unknown
GPU
26.5+ / 256 GiB
Generate~280–809 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
6
gemma-4-31B-it
google/gemma-4-31B-it
2026-04-0231B dense
AA index14.7Reasoning
Plan & memory
Q8_0llama.cpp ROCmFull GPUKV unknown
GPU
31.9+ / 256 GiB
Generate~89–128 t/sEstimatedmeets 5 t/s
First token · 4K prompt~467 ms–1.1 sEstimated
7
gemma-4-12B-it
google/gemma-4-12B-it
2026-06-0312B dense
AA index14.2 est.Reasoning
Plan & memory
Q8_0llama.cpp ROCmFull GPUKV unknown
GPU
13.3+ / 256 GiB
Generate~228–330 t/sEstimatedmeets 5 t/s
First token · 4K prompt~182 ms–410 msEstimated
8
LongCat-Flash-Lite
meituan-longcat/LongCat-Flash-Lite
2026-01 (day unknown)69B MoE · 3B active
AA index11.5 est.single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUKV unknown
GPU
69.4+ / 256 GiB
Generate~355–1,026 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
9
Mistral-Small-4-119B-2603
mistralai/Mistral-Small-4-119B-2603
2026-03-16119B MoE · 6B active
AA index11.3Reasoning
Plan & memory
Q8_0llama.cpp ROCmFull GPUKV unknown
GPU
119+ / 256 GiB
Generate~178–514 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
10
Qwen3-Next-80B-A3B-ThinkingSpecialist
Qwen/Qwen3-Next-80B-A3B-Thinking
2025-09-1181B tensors · MoE
AA index11.2 est.Reasoning
Plan & memory
native BF16vLLMFull GPUKV unknown
GPU
153+ / 256 GiB
Generate≤543 t/sCeilingnot established
First token · 4K prompt—Unknown
11
Ling-3.0-tiny
inclusionAI/Ling-3.0-tiny
2026-08 (day unknown)7.9B MoE · 1.3B active
AA index11.1single mode
Plan & memory
FP16llama.cpp ROCmFull GPUKV unknown
GPU
16.2+ / 256 GiB
Generate~436–1,260 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
12
Qwen3-Next-80B-A3B-InstructOlder
Qwen/Qwen3-Next-80B-A3B-Instruct
2025-09-1181B tensors · MoE
AA index9.6 est.single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUKV unknown
GPU
80.5+ / 256 GiB
Generate~197–569 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
13
diffusiongemma-26B-A4B-it
google/diffusiongemma-26B-A4B-it
2026-06-1025B MoE · 3.8B active
AA index9.5 est.single mode
Plan & memory
Q8_0llama.cpp ROCmFull GPUKV unknown
GPU
26.5+ / 256 GiB
Generate~280–808 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
14
Llama 4 Maverick 17B-128E InstructOlderGated
meta-llama/Llama-4-Maverick-17B-128E-Instruct
2025-04-05402B tensors
AA index9.3single mode
Plan & memory
Q4_K_Mllama.cpp ROCmFull GPUKV unknown
GPU
228+ / 256 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
15
Qwen3-Coder-NextSpecialist
Qwen/Qwen3-Coder-Next
2026-02-0280B MoE · 3B active
AA index9.2single mode
Plan & memory
FP16llama.cpp ROCmFull GPUKV unknown
GPU
150+ / 256 GiB
Generate~190–548 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
16
gemma-4-E4B-it
google/gemma-4-E4B-it
2026-04-028B dense
AA index8.9Reasoning
Plan & memory
Q8_0llama.cpp ROCmFull GPUKV unknown
GPU
9.0+ / 256 GiB
Generate~360–521 t/sEstimatedmeets 5 t/s
First token · 4K prompt~122 ms–275 msEstimated
17
Falcon-H1R-7B
tiiuae/Falcon-H1R-7B
2026-01-057B dense
AA index7.8 est.single mode
Plan & memory
native BF16vLLMFull GPUKV unknown
GPU
15.6+ / 256 GiB
Generate≤395 t/sCeilingnot established
First token · 4K prompt—Unknown
18
gemma-4-E2B-it
google/gemma-4-E2B-it
2026-04-025.1B dense
AA index7.8 est.Reasoning
Plan & memory
Q8_0llama.cpp ROCmFull GPUKV unknown
GPU
6.1+ / 256 GiB
Generate~583–842 t/sEstimatedmeets 5 t/s
First token · 4K prompt~78 ms–175 msEstimated
19
Llama 3.3 70B InstructOlderGated
meta-llama/Llama-3.3-70B-Instruct
2024-12-0671B tensors
AA index7.7 est.single mode
Plan & memory
FP16llama.cpp ROCmFull GPUKV unknown
GPU
133+ / 256 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
20
Llama 4 Scout 17B-16E InstructOlderGated
meta-llama/Llama-4-Scout-17B-16E-Instruct
2025-04-05109B tensors
AA index6.5single mode
Plan & memory
native BF16vLLMFull GPUKV unknown
GPU
204+ / 256 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
21
Phi-4-mini-instructOlder
microsoft/Phi-4-mini-instruct
2025-02-263.8B tensors · dense
AA index6.3 est.single mode
Plan & memory
FP16llama.cpp ROCmFull GPUKV unknown
GPU
8.7+ / 256 GiB
Generate~377–545 t/sEstimatedmeets 5 t/s
First token · 4K prompt~58 ms–132 msEstimated
22
tiny-aya-globalGated
CohereLabs/tiny-aya-global
2026-02-173.3B tensors
AA index4.8 est.single mode
Plan & memory
FP16llama.cpp ROCmFull GPUKV unknown
GPU
7.7+ / 256 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown

+ 32 models without a AA index score

DeepSeek-R1-Distill-Qwen-1.5BQ8_0 · full gpu · ~1,776 t/s
NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16Q5_K_M · full gpu · ~916 t/s
GLM-4.7-FlashQ8_0 · full gpu · ~725 t/s
Ling-3.0-flashQ8_0 · full gpu · ~414 t/s
DeepSeek-R1-Distill-Qwen-7BFP16 · full gpu · ~221 t/s
Qwen2.5-7B-InstructFP16 · full gpu · ~221 t/s
DeepSeek-R1-Distill-Qwen-14BFP16 · full gpu · ~114 t/s
Qwen2.5-14B-InstructFP16 · full gpu · ~114 t/s
Qwen2.5-Coder-7B-InstructFP16 · full gpu · ~110 t/s
DeepSeek-R1-Distill-Qwen-32BQ8_0 · full gpu · ~97 t/s
Qwen2.5-Coder-32B-InstructFP16 · full gpu · ~51 t/s
Qwen2.5-32B-InstructFP16 · full gpu · ~51 t/s
Qwen2.5-72B-InstructFP16 · full gpu · ~23 t/s
Step-3.7-FlashQ8_0 · full gpu · ~16 t/s
DeepSeek-V3AWQ · full gpu · ≤ 84,714 t/s
DeepSeek-V4-Flash-0731native fp8 · full gpu
Gemma 3 12B ITFP16 · full gpu
Gemma 3 1B ITnative BF16 · full gpu
Gemma 3 27B ITnative BF16 · full gpu
Gemma 3 4B ITFP16 · full gpu
gpt-oss-puzzle-88Bnative mxfp4 · full gpu · ≤ 120 t/s
Llama 2 7Bnative F32/F16 · full gpu
Llama 3.1 70B Instructnative BF16 · full gpu
Llama 3.1 8B Instructnative BF16 · full gpu
Llama 3.2 1B InstructFP16 · full gpu
Llama 3.2 3B InstructFP16 · full gpu
Ming-flash-omni-Previewnative BF16/U8 · full gpu
Nemotron-Elastic-12Bnative BF16 · full gpu
tiny-aya-earthFP16 · full gpu
tiny-aya-fireFP16 · full gpu
tiny-aya-waterFP16 · full gpu
Youtu-Parsing-Omninative BF16 · full gpu · ≤ 562 t/s

Speed unknown 7

No measurement, calibration or bandwidth figure for this plan.

1
gpt-oss-120bOlder
openai/gpt-oss-120b
2025-08-05117B tensors · MoE
AA index11.6High
Plan & memory
Q8_0llama.cpp ROCmFull GPUFits
GPU
60.8 / 256 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
2
gpt-oss-20bOlder
openai/gpt-oss-20b
2025-08-0521B tensors · MoE
AA index9.0High
Plan & memory
native mxfp4vLLMFull GPUFits
GPU
14.5 / 256 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
3
NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
2025-12-1532B tensors · MoE
AA index8.9Reasoning
Plan & memory
native F32/BF16vLLMFull GPUFits*
GPU
60.4 / 256 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
4
granite-4.0-h-1b
ibm-granite/granite-4.0-h-1b
2025-10-281.5B tensors · MoE
AA index5.2 est.single mode
Plan & memory
native BF16vLLMFull GPUFits*
GPU
4.3 / 256 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
5
granite-4.0-micro
ibm-granite/granite-4.0-micro
2025-10-023.4B tensors · MoE
AA index5.1 est.single mode
Plan & memory
native BF16vLLMFull GPUFits
GPU
8.5 / 256 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
6
granite-4.0-350m
ibm-granite/granite-4.0-350m
2025-10-280.4B tensors · MoE
AA index4.8 est.single mode
Plan & memory
native BF16vLLMFull GPUFits
GPU
2.4 / 256 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
7
granite-4.0-h-350m
ibm-granite/granite-4.0-h-350m
2025-10-280.3B tensors · MoE
AA index4.8 est.single mode
Plan & memory
native BF16vLLMFull GPUFits*
GPU
2.2 / 256 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown

+ 6 models without a AA index score

granite-4.0-1bnative BF16 · full gpu
granite-4.0-h-micronative BF16 · full gpu
Jais-3-MoVA-36B-A4Bnative F32 · full gpu
Nemotron-3-Super-120B-A12B-BF16-MTPv2native F32/BF16 · full gpu
NVIDIA-Nemotron-Labs-3-Elastic-30B-A3B-BF16native F32/BF16 · full gpu
NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-BF16native BF16/F32 · full gpu

42 models with no feasible plan on this setup

MiMo-V2.6-Pro-MOPDnative fp8 527 GiB
GLM-5.3native fp8 704 GiB
Step-5-Previewnative fp8 600 GiB
Kimi-K3native F32/BF16 1,454 GiB
Ling-3.1-flashQ4_K_M 333 GiB
Qwen3.8-2.4T-A95Bnative BF16 4,556 GiB
DeepSeek-V4.1-FlashQ4_K_M 414 GiB
Muse-Spark-Open-816B-A42BQ4_K_M 465 GiB exceeds memory
DeepSeek-V4-Pro-0813native fp8 831 GiB
Motif-3native BF16 586 GiB
K2-Horizon-375B-A23Bnative BF16 706 GiB
Beam-1.0-501B-A23BQ5_K_M 331 GiB
Kimi-K2.7-Codenative BF16/F32 554 GiB
InklingQ8_0 798 GiB
Solar-Open2-250Bnative BF16/F32 466 GiB
NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16native BF16/F32 1,044 GiB
A.X-K2native fp8 647 GiB
K-EXAONE-2.0-750B-A37Bnative F32/BF16 1,396 GiB
LongCat-2.0native BF16/F32 3,308 GiB
Ring-2.6-1Tnative quantized 970 GiB
command-a-plus-05-2026-bf16native BF16 407 GiB
sarvam-105bnative F32 395 GiB
Agnes-2.5-Pro-Alphanative BF16 739 GiB
Agnes-3.0-Qwennative BF16/I64 335 GiB
Apertus-v1.1-0.5B-InstructContext 8K exceeds this model's configured maximum (4K)
Apertus-v1.1-1.5B-InstructContext 8K exceeds this model's configured maximum (4K)
Apertus-v1.1-4B-InstructContext 8K exceeds this model's configured maximum (4K)
Apertus-v1.5-8Bno backend/format combination
BaguettotronContext 8K exceeds this model's configured maximum (4K)
DeepSeek-R1Q4_K_M 377 GiB
GLM-5.2Q4_K_M 434 GiB
Hy3-previewnative BF16/F32 557 GiB
Hy4-previewnative BF16/F32 1,453 GiB
Jais-3-375B-A23Bnative BF16/F32 706 GiB
Kimi-K2.6native F32/I32 554 GiB
MonadContext 8K exceeds this model's configured maximum (2K)
NVIDIA-Nemotron-Labs-3-Competitive-Coding-550B-A55B-NVFP4native modelopt 328 GiB
NVIDIA-Nemotron-Labs-Teacher-Chatnative F32/BF16 1,044 GiB
NVIDIA-Nemotron-Labs-Teacher-Competition-Codingnative BF16/F32 1,044 GiB
NVIDIA-Nemotron-Labs-Teacher-Instruction-Followingnative BF16/F32 1,044 GiB
NVIDIA-Nemotron-Labs-Teacher-STEMnative BF16/F32 1,044 GiB
Apertus-v1.5-70Bno verified artifact for these backends

Specs & sources

Memory	256 GB HBM3E	source
Bandwidth	6,000 GB/s	source
FP16	1300 TFLOPS	source
Power	1000 W	source

FP16: dense_matrix_fp16. Power: vendor maximum board/chip TDP; see source.

Reported speeds

0 observations

No reported observation resolved to this exact device yet.

Open in speed database Submit yours

Doesn't fit? Rent a GPU

RunpodPlaceholderCloud GPU pods you can rent for running larger models.
Vast.aiPlaceholderGPU rental marketplace with many hosts and GPU types.
LambdaPlaceholderCloud GPU instances for AI training and inference.
TensorDockPlaceholderGPU cloud marketplace with consumer and datacenter GPUs.

Sponsored placeholder links: LocalFit may earn a commission if these become affiliate links. They never influence fit results, estimates or picks.