Intel Data Center GPU Flex 140 (two-GPU board): 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

Intel · Multi-GPU module

Intel Data Center GPU Flex 140 (two-GPU board)

Memory12 GB GDDR6Bandwidth336 GB/s vendor specLogical GPUs2 × 6 GB board totals are not one GPUPlussystem RAM for offload (set below)

GPUs

Data Center GPU Flex 140 (two-GPU board)12 GB · 336 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

2× Data Center GPU Flex 140 (two-GPU board)64 GB DDR5-5600 ×2 ch · 56.0 GiB usable · 89.6 GB/s theoretical8K contextF16 KVany backend≥ 5 t/s

Best plans 2× Data Center GPU Flex 140 (two-GPU board)

Best quality at ≥ 5 t/s

K2-Horizon-MoVA-36B-A4B

25.3 AA index single mode

Q4_K_Mllama.cpp CPUCPU only

~6.2–7.3 t/s Estimated extrapolated

Speed extrapolated, not measured on this setup.

Faster tradeoff

Qwen3.6-35B-A3B

18.2 AA index Reasoning

Q4_K_Mllama.cpp CPUCPU only

~11–13 t/s Estimated extrapolated

Next point on the quality/speed front; then LFM2.5-8B-A1B.

Possibly better · speed not established

Qwen3.8-27B

33.7 AA index Xhigh

Q8_0llama.cpp SYCLGPU + CPU layers · 21/64 layers on GPU

speed Unknown

No report, calibration or bandwidth figure for this setup.

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

Meets your 5 t/s floor 7

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
K2-Horizon-MoVA-36B-A4B
IFM/K2-Horizon-MoVA-36B-A4B
2026-09-0336B MoE · 4B active
AA index25.3single mode
Plan & memory
Q4_K_Mllama.cpp CPUCPU onlyFits
RAM
22.3 / 56.0 GiB
Generate~6.2–7.3 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
2
Qwen3.6-35B-A3B
Qwen/Qwen3.6-35B-A3B
2026-04-1635B MoE · 3B active
AA index18.2Reasoning
Plan & memory
Q4_K_Mllama.cpp CPUCPU onlyFits*
RAM
19.2 / 56.0 GiB
Generate~11–13 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
3
K2-Horizon-3.7B
IFM/K2-Horizon-3.7B
2026-09-035.1B tensors · dense
AA index15.6single mode
Plan & memory
Q5_K_Mllama.cpp CPUCPU onlyFits
RAM
4.5 / 56.0 GiB
Generate~4.8–5.7 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
4
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
Q4_K_Mllama.cpp CPUCPU onlyFits*
RAM
23.6 / 56.0 GiB
Generate~8.0–9.4 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
5
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
Q4_K_Mllama.cpp CPUCPU onlyFits
RAM
38.1 / 56.0 GiB
Generate~6.0–7.1 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
6
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
Q4_K_Mllama.cpp CPUCPU onlyFits*
RAM
23.1 / 56.0 GiB
Generate~8.1–9.6 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
7
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
Q4_K_Mllama.cpp CPUCPU onlyFits*
RAM
4.9 / 56.0 GiB
Generate~21–25 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown

+ 9 models without a AA index score

granite-4.0-h-tinyQ5_K_M · cpu only · ~21 t/s
LFM2.5-24B-A2BQ4_K_M · cpu only · ~17 t/s
Mellum2.1-12B-A2.5B-ThinkingQ4_K_M · cpu only · ~13 t/s
MiniCPM-V-4.7-35B-A3BQ4_K_M · cpu only · ~12 t/s
Qwen3.5-35B-A3BQ4_K_M · cpu only · ~11 t/s
Kolibri-1Q4_K_M · cpu only · ~10 t/s
Qwen3-30B-A3BQ4_K_M · cpu only · ~9.2 t/s
Qwen3-Coder-30B-A3B-InstructQ4_K_M · cpu only · ~9.1 t/s
Jais-3-3.7BQ5_K_M · cpu only · ~5.3 t/s

Memory not fully resolved 13

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
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
Q4_K_Mllama.cpp CPUCPU onlyKV unknown
RAM
15.9 / 56.0 GiB
Generate~7.9–9.4 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
2
gemma-4-31B-it
google/gemma-4-31B-it
2026-04-0231B dense
AA index14.7Reasoning
Plan & memory
Q8_0llama.cpp SYCLGPU + CPU layers · 17/60 layers on GPUKV unknown
2 GPUs
11.6+ / 12.0 GiB
RAM
21.8 / 56.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
3
gemma-4-12B-it
google/gemma-4-12B-it
2026-06-0312B dense
AA index14.2 est.Reasoning
Plan & memory
Q5_K_Mllama.cpp SYCLLayer split across GPUsKV unknown
2 GPUs
11.2+ / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
4
LongCat-Flash-Lite
meituan-longcat/LongCat-Flash-Lite
2026-01 (day unknown)69B MoE · 3B active
AA index11.5 est.single mode
Plan & memory
Q4_K_Mllama.cpp CPUCPU onlyKV unknown
RAM
37.5 / 56.0 GiB
Generate~12–14 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
5
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
Q4_K_Mllama.cpp CPUCPU onlyKV unknown
RAM
4.6 / 56.0 GiB
Generate~25–30 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
6
Qwen3-Next-80B-A3B-InstructOlder
Qwen/Qwen3-Next-80B-A3B-Instruct
2025-09-1181B tensors · MoE
AA index9.6 est.single mode
Plan & memory
Q4_K_Mllama.cpp CPUCPU onlyKV unknown
RAM
45.2 / 56.0 GiB
Generate~6.2–7.3 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
7
diffusiongemma-26B-A4B-it
google/diffusiongemma-26B-A4B-it
2026-06-1025B MoE · 3.8B active
AA index9.5 est.single mode
Plan & memory
Q4_K_Mllama.cpp CPUCPU onlyKV unknown
RAM
15.7 / 56.0 GiB
Generate~8.0–9.5 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
8
Qwen3-Coder-NextSpecialist
Qwen/Qwen3-Coder-Next
2026-02-0280B MoE · 3B active
AA index9.2single mode
Plan & memory
Q4_K_Mllama.cpp CPUCPU onlyKV unknown
RAM
45.4 / 56.0 GiB
Generate~11–13 t/sEstimatedextrapolated≈ meets 5 t/s
First token · 4K prompt—Unknown
9
gemma-4-E4B-it
google/gemma-4-E4B-it
2026-04-028B dense
AA index8.9Reasoning
Plan & memory
Q8_0llama.cpp SYCLLayer split across GPUsKV unknown
2 GPUs
10.5+ / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
10
gemma-4-E2B-it
google/gemma-4-E2B-it
2026-04-025.1B dense
AA index7.8 est.Reasoning
Plan & memory
Q8_0llama.cpp SYCLLayer split across GPUsKV unknown
2 GPUs
7.6+ / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
11
Llama 3.3 70B InstructOlderGated
meta-llama/Llama-3.3-70B-Instruct
2024-12-0671B tensors
AA index7.7 est.single mode
Plan & memory
Q5_K_Mllama.cpp CPUCPU onlyKV unknown
RAM
46.5 / 56.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
12
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 SYCLLayer split across GPUsKV unknown
2 GPUs
10.2+ / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
13
tiny-aya-globalGated
CohereLabs/tiny-aya-global
2026-02-173.3B tensors
AA index4.8 est.single mode
Plan & memory
FP16llama.cpp SYCLLayer split across GPUsKV unknown
2 GPUs
9.2+ / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown

+ 24 models without a AA index score

GLM-4.7-FlashQ4_K_M · cpu only · ~12 t/s
DeepSeek-R1-Distill-Qwen-1.5BQ8_0 · full gpu
DeepSeek-R1-Distill-Qwen-14BQ4_K_M · layer split across gpus
DeepSeek-R1-Distill-Qwen-32BQ8_0 · gpu + cpu layers
DeepSeek-R1-Distill-Qwen-7BQ8_0 · layer split across gpus
Gemma 3 12B ITQ5_K_M · layer split across gpus
Gemma 3 1B ITQ8_0 · full gpu
Gemma 3 27B ITQ8_0 · cpu only
Gemma 3 4B ITFP16 · layer split across gpus
Llama 3.1 70B InstructQ5_K_M · cpu only
Llama 3.1 8B InstructQ8_0 · layer split across gpus
Llama 3.2 1B InstructFP16 · full gpu
Llama 3.2 3B InstructFP16 · layer split across gpus
Nemotron-Elastic-12BQ5_K_M · layer split across gpus
NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16Q5_K_M · full gpu
Qwen2.5-14B-InstructQ4_K_M · layer split across gpus
Qwen2.5-32B-InstructFP16 · gpu + cpu layers
Qwen2.5-72B-InstructQ5_K_M · gpu + cpu layers
Qwen2.5-7B-InstructQ8_0 · layer split across gpus
Qwen2.5-Coder-32B-InstructFP16 · gpu + cpu layers
Qwen2.5-Coder-7B-InstructQ4_K_M · layer split across gpus
tiny-aya-earthFP16 · layer split across gpus
tiny-aya-fireFP16 · layer split across gpus
tiny-aya-waterFP16 · layer split across gpus

Speed unknown 35

No measurement, calibration or bandwidth figure for this plan.

1
Qwen3.8-27B
Qwen/Qwen3.8-27B
2026-08-1427B dense
AA index33.7Xhigh
Plan & memory
Q8_0llama.cpp SYCLGPU + CPU layers · 21/64 layers on GPUFits*
2 GPUs
11.9 / 12.0 GiB
RAM
18.2 / 56.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
2
K2-Horizon-7B
IFM/K2-Horizon-7B
2026-09-039B tensors · dense
AA index20.6single mode
Plan & memory
Q5_K_Mllama.cpp SYCLLayer split across GPUsFits
2 GPUs
10.1 / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
3
Muse-Glimmer-30B
meta-models/Muse-Glimmer-30B
2026-08-1030B dense
AA index17.5High
Plan & memory
Q8_0llama.cpp SYCLGPU + CPU layers · 16/52 layers on GPUFits
2 GPUs
11.5 / 12.0 GiB
RAM
19.2 / 56.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
4
EXAONE-4.5-33B
LGAI-EXAONE/EXAONE-4.5-33B
2026-04-0933B dense
AA index13.2 est.Reasoning
Plan & memory
Q8_0llama.cpp SYCLGPU + CPU layers · 16/64 layers on GPUFits
2 GPUs
11.5 / 12.0 GiB
RAM
25.5 / 56.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
5
Qwen3.5-4B
Qwen/Qwen3.5-4B
2026-03-024B dense
AA index13.1 est.Reasoning
Plan & memory
Q8_0llama.cpp SYCLLayer split across GPUsFits*
2 GPUs
7.6 / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
6
granite-4.2-30b
ibm-granite/granite-4.2-30b
2026-08-2530B dense
AA index12.7single mode
Plan & memory
Q8_0llama.cpp SYCLGPU + CPU layers · 18/64 layers on GPUFits
2 GPUs
11.7 / 12.0 GiB
RAM
22.3 / 56.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
7
MiniCPM5-2B
openbmb/MiniCPM5-2B
2026-09-072.5B dense
AA index12.5single mode
Plan & memory
FP16llama.cpp SYCLLayer split across GPUsFits
2 GPUs
8.0 / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
8
gpt-oss-120bOlder
openai/gpt-oss-120b
2025-08-05117B tensors · MoE
AA index11.6High
Plan & memory
Q8_0llama.cpp SYCLGPU + CPU layers · 5/36 layers on GPUFits
2 GPUs
11.2 / 12.0 GiB
RAM
51.1 / 56.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
9
Qwen3.5-9B
Qwen/Qwen3.5-9B
2026-03-029B dense
AA index11.2Reasoning
Plan & memory
Q5_K_Mllama.cpp SYCLLayer split across GPUsFits*
2 GPUs
9.9 / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
10
granite-4.2-8b
ibm-granite/granite-4.2-8b
2026-08-258B dense
AA index11.1single mode
Plan & memory
Q5_K_Mllama.cpp SYCLLayer split across GPUsFits
2 GPUs
10.1 / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
11
G9v3-3B
ai9stars/G9v3-3B
2026-07 (day unknown)3B dense
AA index10.8 est.single mode
Plan & memory
FP16llama.cpp SYCLLayer split across GPUsFits
2 GPUs
9.0 / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
12
granite-4.2-3b
ibm-granite/granite-4.2-3b
2026-08-253B dense
AA index9.1single mode
Plan & memory
Q8_0llama.cpp SYCLFull GPUFits
GPU
5.8 / 6.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
13
Tri-21B-Think
trillionlabs/Tri-21B-Think
2026-02 (day unknown)21B dense
AA index9.0 est.single mode
Plan & memory
Q8_0llama.cpp SYCLGPU + CPU layers · 16/40 layers on GPUFits
2 GPUs
11.8 / 12.0 GiB
RAM
13.2 / 56.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
14
gpt-oss-20bOlder
openai/gpt-oss-20b
2025-08-0521B tensors · MoE
AA index9.0High
Plan & memory
Q8_0llama.cpp SYCLGPU + CPU layers · 18/24 layers on GPUFits
2 GPUs
11.6 / 12.0 GiB
RAM
2.9 / 56.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
15
LFM2.5-2.6B
LiquidAI/LFM2.5-2.6B
2026-08-042.7B dense
AA index8.4 est.single mode
Plan & memory
FP16llama.cpp SYCLLayer split across GPUsFits*
2 GPUs
8.2 / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
16
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 SYCLGPU + CPU layers · 17/64 layers on GPUFits
2 GPUs
11.8 / 12.0 GiB
RAM
24.3 / 56.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
17
Qwen3.5-2B
Qwen/Qwen3.5-2B
2026-03-022B dense
AA index6.9 est.Reasoning
Plan & memory
Q8_0llama.cpp SYCLFull GPUFits*
GPU
3.5 / 6.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
18
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 SYCLGPU + CPU layers · 17/64 layers on GPUFits
2 GPUs
11.8 / 12.0 GiB
RAM
24.3 / 56.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
19
Qwen3.5-0.8B
Qwen/Qwen3.5-0.8B
2026-03-020.8B dense
AA index6.1 est.Reasoning
Plan & memory
Q8_0llama.cpp SYCLFull GPUFits*
GPU
2.4 / 6.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
20
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
Q8_0llama.cpp SYCLGPU + CPU experts · 35 expert layers on CPUFits*
2 GPUs
11.5 / 12.0 GiB
RAM
23.6 / 56.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
21
Phi-4Older
microsoft/phi-4
2024-12-1215B tensors · dense
AA index5.9 est.single mode
Plan & memory
FP16llama.cpp SYCLGPU + CPU layers · 12/40 layers on GPUFits
2 GPUs
11.7 / 12.0 GiB
RAM
20.2 / 56.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
22
Olmo-3-7B-ThinkSpecialist
allenai/Olmo-3-7B-Think
2025-11-207.3B tensors · dense
AA index5.6 est.single mode
Plan & memory
Q5_K_Mllama.cpp SYCLLayer split across GPUsFits
2 GPUs
10.4 / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
23
Olmo-3-7B-Instruct
allenai/Olmo-3-7B-Instruct
2025-11-207.3B tensors · dense
AA index5.2 est.single mode
Plan & memory
Q5_K_Mllama.cpp SYCLLayer split across GPUsFits
2 GPUs
10.4 / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
24
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 SYCLFull GPUFits*
GPU
3.8 / 6.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
25
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 SYCLFull GPUFits*
GPU
3.8 / 6.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
26
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
Q8_0llama.cpp SYCLFull GPUFits*
GPU
3.0 / 6.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
27
Apertus-70B-Instruct-2509
swiss-ai/Apertus-70B-Instruct-2509
2025-09-0271B tensors · dense
AA index5.1 est.single mode
Plan & memory
Q5_K_Mllama.cpp SYCLGPU + CPU layers · 14/80 layers on GPUFits
2 GPUs
11.7 / 12.0 GiB
RAM
40.9 / 56.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
28
granite-4.0-micro
ibm-granite/granite-4.0-micro
2025-10-023.4B tensors · MoE
AA index5.1 est.single mode
Plan & memory
Q8_0llama.cpp SYCLFull GPUFits
GPU
5.5 / 6.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
29
Molmo2-8BSpecialist
allenai/Molmo2-8B
2025-12-118.7B tensors · dense
AA index5.0 est.single mode
Plan & memory
Q4_K_Mllama.cpp SYCLLayer split across GPUsFits
2 GPUs
8.8 / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
30
Apertus-8B-Instruct-2509
swiss-ai/Apertus-8B-Instruct-2509
2025-09-028.1B tensors · dense
AA index4.8 est.single mode
Plan & memory
Q8_0llama.cpp SYCLLayer split across GPUsFits
2 GPUs
12.0 / 12.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
31
granite-4.0-350m
ibm-granite/granite-4.0-350m
2025-10-280.4B tensors · MoE
AA index4.8 est.single mode
Plan & memory
Q8_0llama.cpp SYCLFull GPUFits
GPU
2.1 / 6.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
32
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
Q8_0llama.cpp SYCLFull GPUFits*
GPU
1.9 / 6.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
33
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 SYCLFull GPUFits*
GPU
3.8 / 6.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
34
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 SYCLFull GPUFits*
GPU
5.6 / 6.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown
35
K2-Horizon-0.9B
IFM/K2-Horizon-0.9B
2026-09-031.1B tensors · dense
AA index3.0single mode
Plan & memory
Q8_0llama.cpp SYCLFull GPUFits
GPU
3.0 / 6.0 GiB
Generate—Unknownspeed unknown
First token · 4K prompt—Unknown

+ 53 models without a AA index score

AgentCPM-ExploreQ8_0 · layer split across gpus
AgentCPM-ReportQ4_K_M · layer split across gpus
Agents-A1-4BQ8_0 · full gpu
DeepSeek-R1-Distill-Llama-70BQ5_K_M · gpu + cpu layers
DeepSeek-R1-Distill-Llama-8BQ8_0 · layer split across gpus
FrogBoss-32B-2510FP16 · gpu + cpu layers
FrogMini-14B-2510FP16 · gpu + cpu layers
FrogNano-4B-2609Q8_0 · layer split across gpus
granite-4.0-1bQ8_0 · full gpu
granite-4.0-h-microQ8_0 · full gpu
granite-4.1-30bQ8_0 · gpu + cpu layers
granite-4.1-3bQ8_0 · full gpu
granite-4.1-8bQ5_K_M · layer split across gpus
IQuest-Coder-V1-14B-BaseFP16 · gpu + cpu layers
IQuest-Coder-V1-14B-Base-Stage1FP16 · gpu + cpu layers
IQuest-Coder-V1-14B-InstructFP16 · gpu + cpu layers
IQuest-Coder-V1-14B-ThinkingFP16 · gpu + cpu layers
IQuest-Coder-V1-40B-ThinkingQ8_0 · gpu + cpu layers
IQuest-Coder-V1-7B-BaseQ8_0 · layer split across gpus
IQuest-Coder-V1-7B-Base-Stage1Q8_0 · layer split across gpus
IQuest-Coder-V1-7B-InstructQ8_0 · layer split across gpus
IQuest-Coder-V1-7B-ThinkingQ8_0 · layer split across gpus
Jais-3-0.9BQ8_0 · full gpu
Jais-3-32BQ8_0 · gpu + cpu layers
Jais-3-7BQ5_K_M · layer split across gpus
Jais-3-MoVA-36B-A4BQ8_0 · gpu + cpu layers
LFM2.5-1.2B-JPFP16 · full gpu
LFM2.5-230MFP16 · full gpu
LFM2.5-350MFP16 · full gpu
LittleLambFP16 · full gpu
LittleLamb-ToolCallingFP16 · full gpu
MagenticBrainQ8_0 · gpu + cpu layers
MiMo-V2.6-Distill-Qwen-9BQ5_K_M · layer split across gpus
Mistral 7B Instruct v0.3Q8_0 · layer split across gpus
Mistral NeMo 12B Instruct (2407)Q4_K_M · layer split across gpus
Mistral Small 3.2 24B Instruct (2506)Q8_0 · gpu + cpu layers
Mixtral 8x7B Instruct v0.1Q8_0 · gpu + cpu experts
Nemotron-Cascade-8BQ5_K_M · layer split across gpus
NVIDIA-Nemotron-Labs-3-Elastic-30B-A3B-BF16Q8_0 · gpu + cpu layers
NVIDIA-Nemotron-Nano-9B-v2-JapaneseQ8_0 · layer split across gpus
Olmo-3-32B-ThinkQ8_0 · gpu + cpu layers
Ornith-1.5-9BQ5_K_M · layer split across gpus
PaCoRe-8BQ4_K_M · layer split across gpus
Qwen3-0.6BFP16 · full gpu
Qwen3-1.7BFP16 · layer split across gpus
Qwen3-14BFP16 · gpu + cpu layers
Qwen3-32BQ8_0 · gpu + cpu layers
Qwen3-4BQ8_0 · layer split across gpus
Qwen3-8BQ5_K_M · layer split across gpus
Qwen3-Nemotron-32B-RLBFFQ8_0 · gpu + cpu layers
Qwen3.5-27BQ8_0 · gpu + cpu layers
Qwen3.6-27BQ8_0 · gpu + cpu layers
RLVR-8B-0926Q5_K_M · layer split across gpus

96 models with no feasible plan on this setup

MiMo-V2.6-Pro-MOPDno backend/format combination
GLM-5.3Q8_0 746 GiB exceeds memory
Step-5-Previewno backend/format combination
Kimi-K3no backend/format combination
GLM-5.3-FlashQ8_0 318 GiB exceeds memory
Ling-3.1-flashQ4_K_M 333 GiB exceeds memory
Qwen3.8-2.4T-A95Bno backend/format combination
Qwen3.8-Flash-NextQ4_K_M 111 GiB exceeds memory
DeepSeek-V4.1-FlashQ4_K_M 414 GiB exceeds memory
Muse-Spark-Open-816B-A42BQ4_K_M 465 GiB exceeds memory
MiMo-V2.6-Flash-MOPDno backend/format combination
DeepSeek-V4-Pro-0813no backend/format combination
Motif-3no backend/format combination
K2-Horizon-375B-A23Bno backend/format combination
Beam-1.0-501B-A23BQ5_K_M 331 GiB exceeds memory
MiniMax-M3Q4_K_M 247 GiB exceeds memory
Nex-N2-ProQ4_K_M 225 GiB exceeds memory
NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16Q4_K_M 80.1 GiB exceeds memory
Kimi-K2.7-Codeno backend/format combination
Inkling-SmallQ4_K_M 151 GiB exceeds memory
Hy3Q4_K_M 170 GiB exceeds memory
InklingQ8_0 798 GiB exceeds memory
Solar-Open2-250Bno backend/format combination
Ling-3.0-flash-VLQ4_K_M 73.3 GiB exceeds memory
NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16no backend/format combination
Ling-3.0-flash-FinQ4_K_M 72.5 GiB exceeds memory
G9v3-39A5B (preview)no backend/format combination
A.X-K2no backend/format combination
NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16Q8_0 73.4 GiB exceeds memory
K-EXAONE-2.0-750B-A37Bno backend/format combination
LongCat-2.0no backend/format combination
Qwen3.5-397B-A17BQ4_K_M 227 GiB exceeds memory
Ring-2.6-1Tno backend/format combination
Qwen3.5-122B-A10BQ4_K_M 72.3 GiB exceeds memory
Mistral-Medium-3.5-128BQ4_K_M 69.8 GiB exceeds memory
command-a-plus-05-2026-bf16no backend/format combination
Mistral-Small-4-119B-2603Q4_K_M 67.6 GiB exceeds memory
Qwen3-Next-80B-A3B-Thinkingno backend/format combination
Trinity-Large-ThinkingQ4_K_M 225 GiB exceeds memory
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16no backend/format combination
North-Mini-Code-1.0no backend/format combination
Llama 4 Maverick 17B-128E InstructQ4_K_M 226 GiB exceeds memory
NVIDIA-Nemotron-3-Nano-30B-A3B-BF16no backend/format combination
MiniCPM5-1Bno backend/format combination
sarvam-105bno backend/format combination
Nanbeige4.1-3Bno backend/format combination
Falcon-H1R-7Bno backend/format combination
Step3-VL-10Bno backend/format combination
NVIDIA-Nemotron-Nano-12B-v2-VL-BF16no backend/format combination
Llama 4 Scout 17B-16E InstructQ4_K_M 62.9 GiB exceeds memory
Agents-A1no backend/format combination
Agnes-2.5-Pro-Alphano backend/format combination
Agnes-3.0-Qwenno backend/format combination
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 exceeds memory
DeepSeek-V3Q4_K_M 377 GiB exceeds memory
DeepSeek-V4-Flash-0731no backend/format combination
GLM-4.5-AirQ4_K_M 68.0 GiB exceeds memory
GLM-5.2Q4_K_M 434 GiB exceeds memory
gpt-oss-puzzle-88Bno backend/format combination
Hy3-previewno backend/format combination
Hy4-previewno backend/format combination
IQuest-Coder-V1-40B-Loop-Thinkingno backend/format combination
Jais-3-375B-A23Bno backend/format combination
K2-Horizon-32Bno backend/format combination
Kimi-K2.6no backend/format combination
Laguna-S-2.1Q5_K_M 81.8 GiB exceeds memory
Ling-3.0-flashQ4_K_M 72.9 GiB exceeds memory
Llama 2 7Bno backend/format combination
Ming-flash-omni-2.0no backend/format combination
Ming-flash-omni-Previewno backend/format combination
MiniCPM-SALAno backend/format combination
MiniMax-M2.7Q4_K_M 131 GiB exceeds memory
Mixtral 8x22B Instruct v0.1no backend/format combination
MonadContext 8K exceeds this model's configured maximum (2K)
Nemotron-3-Super-120B-A12B-BF16-MTPv2no backend/format combination
NVIDIA-Nemotron-3-Super-120B-A12B-BF16no backend/format combination
NVIDIA-Nemotron-Labs-3-Competitive-Coding-550B-A55B-NVFP4no backend/format combination
NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-BF16no backend/format combination
NVIDIA-Nemotron-Labs-Teacher-Chatno backend/format combination
NVIDIA-Nemotron-Labs-Teacher-Competition-Codingno backend/format combination
NVIDIA-Nemotron-Labs-Teacher-Instruction-Followingno backend/format combination
NVIDIA-Nemotron-Labs-Teacher-STEMno backend/format combination
Penguin-VL-2Bno backend/format combination
Penguin-VL-8Bno backend/format combination
Qwen3-235B-A22BQ4_K_M 132 GiB exceeds memory
Qwen3-Coder-480B-A35B-InstructQ4_K_M 270 GiB exceeds memory
Qwen3-VL-235B-A22B-InstructQ4_K_M 132 GiB exceeds memory
Qwen3-VL-235B-A22B-ThinkingQ4_K_M 132 GiB exceeds memory
Step-3.7-FlashQ4_K_M 114 GiB exceeds memory
Youtu-Parsing-Omnino backend/format combination
Apertus-v1.5-70Bno verified artifact for these backends

Specs & sources

Memory	12 GB GDDR6	source
Bandwidth	336 GB/s	source
FP16	unknown	No unambiguous comparable value verified; never substitute AI TOPS, BF16, or whole-system power.
Power	75 W	source

Intel ARK uses board aggregate memory and core counts. Do not fit a 12GB single-device allocation without supported distribution.

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.