NVIDIA A10G, AI & Machine Learning Benchmarks & Specs

24GB · AI Score 6.4/100 · first-party measured on 12 AI workloads

6.4 AI Score ✓ Measured

Every number on this page is first-party: NVIDIA A10G was run on our pinned 12-workload AI suite on 2026-07-10, with under 0.5% run-to-run variance. On Llama 3.1 8B (Q4_K_M) NVIDIA A10G delivers about 86.6 tokens/sec. Llama 3.3 70B does not fit. It needs roughly 42GB and this card has 24GB. For image generation, SDXL runs at 3.19 it/s, while FLUX.1-dev won't fit at BF16 (needs ~26GB). 5 of the 12 workloads won't fit on 24GB at the tested precision, Qwen3 32B, Llama 3.3 70B, FLUX.1-dev, FLUX.1 Kontext and others. We publish those as hard gates rather than quietly dropping to a smaller quant.

AI & Machine Learning benchmark results

Text Generation tok/s 123

SmolLM2-135M680.37

gemma-3-270m597.49

Qwen1.5-0.5B542.19

Qwen2.5-Coder-0.5B513.36

Qwen2.5-0.5B508.78

Qwen3 0.6B451.38

Qwen3-0.6B446.74

LFM2.5-1.2B429.96

Llama 3.2 1B400.6

gemma-3-1b284.78

LFM2.5-8B-A1B268.04

Qwen3 1.7B267.08

Workload	Result	Telemetry	Data
SmolLM2-135M	680.37 tok/s	71 W56°CQ4_K_M	✓ Measured
gemma-3-270m	597.49 tok/s	68 W54°CQ4_K_M	✓ Measured
Qwen1.5-0.5B	542.19 tok/s	74 W54°CQ4_K_M	✓ Measured
Qwen2.5-Coder-0.5B	513.36 tok/s	79 W54°CQ4_K_M	✓ Measured
Qwen2.5-0.5B	508.78 tok/s	76 W51°CQ4_K_M	✓ Measured
Qwen3 0.6B	451.38 tok/s	75 W34°CQ4_K_M	✓ Measured
Qwen3-0.6B	446.74 tok/s	80 W55°CQ4_K_M	✓ Measured
LFM2.5-1.2B	429.96 tok/s	82 W58°CQ4_K_M	✓ Measured
Llama 3.2 1B	400.6 tok/s	95 W46°CQ4_K_M	✓ Measured
gemma-3-1b	284.78 tok/s	97 W55°CQ4_K_M	✓ Measured
LFM2.5-8B-A1B	268.04 tok/s	96 W58°CQ4_K_M	✓ Measured
Qwen3 1.7B	267.08 tok/s	100 W36°CQ4_K_M	✓ Measured
DeepSeek-R1 Distill 1.5B	265.75 tok/s	102 W53°CQ4_K_M	✓ Measured
Qwen2.5-1.5B	265.7 tok/s	81 W54°CQ4_K_M	✓ Measured
Qwen2-1.5B	265.45 tok/s	96 W56°CQ4_K_M	✓ Measured
Qwen2.5-Coder-1.5B	264.79 tok/s	94 W57°CQ4_K_M	✓ Measured
Qwen3-1.7B	262.49 tok/s	101 W59°CQ4_K_M	✓ Measured
gemma-2-2b	175.03 tok/s	114 W56°CQ4_K_M	✓ Measured
gemma-2-2b-it-abliterated	174.56 tok/s	110 W59°CQ4_K_M	✓ Measured
phi-2	172.45 tok/s	112 W58°CQ4_K_M	✓ Measured
SmolLM3 3B	172.15 tok/s	107 W56°CQ4_K_M	✓ Measured
SmolLM3-3B	171.66 tok/s	111 W59°CQ4_K_M	✓ Measured
Dolphin X1 Trinity Nano 6B	171.23 tok/s	93 W60°CQ4_K_M	✓ Measured
Llama 3.2 3B	170.47 tok/s	107 W47°CQ4_K_M	✓ Measured
Hermes-3-Llama-3.2-3B	169.23 tok/s	102 W55°CQ4_K_M	✓ Measured
Qwen2.5-3B	168.23 tok/s	111 W56°CQ4_K_M	✓ Measured
Qwen2.5-Coder-3B	168 tok/s	108 W57°CQ4_K_M	✓ Measured
Llama-3.2-3B-Instruct-uncensored	167.63 tok/s	104 W58°CQ4_K_M	✓ Measured
AI21-Jamba-Reasoning-3B	166.29 tok/s	107 W55°CQ4_K_M	✓ Measured
DeepSeek-Coder-V2-Lite	160.13 tok/s	104 W57°CQ4_K_M	✓ Measured
Phi-3.5-mini	145.97 tok/s	120 W53°CQ4_K_M	✓ Measured
gpt-oss-20b	144.61 tok/s	99 W60°CQ4_K_M	✓ Measured
Phi-4 Mini 3.8B	144.42 tok/s	113 W52°CQ4_K_M	✓ Measured
Phi-4-mini	144.27 tok/s	114 W60°CQ4_K_M	✓ Measured
Qwen3-Coder 30B A3B	142.77 tok/s	97 W59°CQ4_K_M	✓ Measured
Qwen3-30B-A3B	140.14 tok/s	97 W58°CQ4_K_M	✓ Measured
Qwen3 30B A3B	140.12 tok/s	101 W59°CQ4_K_M	✓ Measured
Qwen3-4B-Instruct-2507	130.83 tok/s	111 W56°CQ4_K_M	✓ Measured
Qwen3-4B-Thinking-2507	129.99 tok/s	117 W58°CQ4_K_M	✓ Measured
Qwen3-4B-Instruct-2507	129.9 tok/s	118 W59°CQ4_K_M	✓ Measured
Qwen3-4B	129.87 tok/s	112 W57°CQ4_K_M	✓ Measured
Gemma 3 4B	124.85 tok/s	116 W49°CQ4_K_M	✓ Measured
Qwen3 30B A3B (Q3_K_M)	120.38 tok/s	102 W60°CQ3_K_M	✓ Measured
KAT-Coder-V2.5-Dev	113.6 tok/s	87 W55°CQ4_K_M	✓ Measured
Ornith-1.0-35B	104.9 tok/s	97 W56°CQ4_K_M	✓ Measured
Qwen-AgentWorld-35B-A3B	104.36 tok/s	96 W56°CQ4_K_M	✓ Measured
GLM-4.7-Flash	103.24 tok/s	101 W53°CQ4_K_M	✓ Measured
GLM-4.7-Flash-REAP-23B-A3B	97.48 tok/s	110 W58°CQ4_K_M	✓ Measured
Llama-2-7B	96.69 tok/s	121 W58°CQ4_K_M	✓ Measured
Mistral 7B v0.3	92.82 tok/s	125 W56°CQ4_K_M	✓ Measured
Mistral-7B-Instruct-v0.2	92.74 tok/s	124 W57°CQ4_K_M	✓ Measured
Mistral-7B-Instruct-v0.1	92.45 tok/s	128 W60°CQ4_K_M	✓ Measured
Mistral-7B-Instruct-v0.3	92.45 tok/s	119 W59°CQ4_K_M	✓ Measured
Qwen2.5-Coder-7B-Instruct-abliterated	91.41 tok/s	110 W55°CQ4_K_M	✓ Measured
Qwen2.5-Coder 7B	90.05 tok/s	121 W56°CQ4_K_M	✓ Measured
DeepSeek-R1 Distill 7B	89.79 tok/s	122 W54°CQ4_K_M	✓ Measured
Qwen2.5-7B	89.68 tok/s	120 W58°CQ4_K_M	✓ Measured
dolphin-2.9-llama3-8b	87.98 tok/s	114 W56°CQ4_K_M	✓ Measured
DarkIdol-Llama-3.1-8B-Instruct-1.2-Uncensored	87.84 tok/s	113 W56°CQ4_K_M	✓ Measured
DeepSeek-R1 Distill Llama 8B	86.83 tok/s	122 W55°CQ4_K_M	✓ Measured
L3-8B-Stheno-v3.2	86.8 tok/s	118 W58°CQ4_K_M	✓ Measured
Meta-Llama-3.1-8B	86.69 tok/s	118 W57°CQ4_K_M	✓ Measured
Llama-3.1-8B	86.62 tok/s	120 W58°CQ4_K_M	✓ Measured
Dolphin X1 8B	86.61 tok/s	121 W61°CQ4_K_M	✓ Measured
Dolphin 3.0 Llama 3.1 8B	86.54 tok/s	122 W61°CQ4_K_M	✓ Measured
Qwen3 8B	84.09 tok/s	117 W41°CQ4_K_M	✓ Measured
Josiefied-Qwen3-8B-abliterated-v1	83.26 tok/s	121 W59°CQ4_K_M	✓ Measured
DeepSeek-R1-0528-Qwen3-8B	83.22 tok/s	122 W54°CQ4_K_M	✓ Measured
Qwen3-8B	83.04 tok/s	120 W60°CQ4_K_M	✓ Measured
Ornith-1.0-9B	73.78 tok/s	121 W55°CQ4_K_M	✓ Measured
NemoMix-Unleashed-12B	57.52 tok/s	118 W56°CQ4_K_M	✓ Measured
gemma-2-9b	57.37 tok/s	128 W60°CQ4_K_M	✓ Measured
Mistral-Nemo-Instruct-2407	57 tok/s	125 W59°CQ4_K_M	✓ Measured
Gemma 4 12B	52.67 tok/s	125 W58°CQ4_K_M	✓ Measured
Gemma 3 12B	52.07 tok/s	128 W52°CQ4_K_M	✓ Measured
Phi-4 14B	48.18 tok/s	128 W59°CQ4_K_M	✓ Measured
Qwen3 14B	48.12 tok/s	126 W46°CQ4_K_M	✓ Measured
Qwen3-14B	47.8 tok/s	124 W60°CQ4_K_M	✓ Measured
Hermes-4-14B	47.8 tok/s	130 W60°CQ4_K_M	✓ Measured
Uncensored	47.45 tok/s	124 W57°CQ4_K_M	✓ Measured
Qwen2.5-Coder-14B	47.07 tok/s	123 W57°CQ4_K_M	✓ Measured
Qwen2.5-14B	47.02 tok/s	125 W56°CQ4_K_M	✓ Measured
EVA-Qwen2.5-14B-v0.2	47 tok/s	130 W58°CQ4_K_M	✓ Measured
DeepSeek-R1 Distill 14B	46.96 tok/s	131 W59°CQ4_K_M	✓ Measured
Qwen2.5-Coder-14B-Instruct-abliterated	46.91 tok/s	127 W59°CQ4_K_M	✓ Measured
Gemma 3 12B (Q3_K_M)	46.29 tok/s	126 W57°CQ3_K_M	✓ Measured
Phi-4 14B (Q3_K_M)	43.56 tok/s	126 W57°CQ3_K_M	✓ Measured
StarCoder2 15B	42.19 tok/s	132 W62°CQ4_K_M	✓ Measured
DeepSeek-R1 Distill 14B (Q3_K_M)	41.72 tok/s	126 W58°CQ3_K_M	✓ Measured
Codestral 22B	32.09 tok/s	133 W61°CQ4_K_M	✓ Measured
Cydonia-24B-v4.3	31.04 tok/s	131 W58°CQ4_K_M	✓ Measured
Dolphin-Mistral-24B-Venice-Edition	31 tok/s	129 W59°CQ4_K_M	✓ Measured
Devstral Small 24B	30.98 tok/s	133 W60°CQ4_K_M	✓ Measured
Dolphin Mistral 24B Venice	30.95 tok/s	131 W61°CQ4_K_M	✓ Measured
Mistral Small 24B	30.94 tok/s	132 W61°CQ4_K_M	✓ Measured
Dolphin 3.0 R1 Mistral 24B	30.9 tok/s	131 W62°CQ4_K_M	✓ Measured
Codestral 22B (Q3_K_M)	28.01 tok/s	133 W62°CQ3_K_M	✓ Measured
Mistral Small 24B (Q3_K_M)	26.8 tok/s	131 W59°CQ3_K_M	✓ Measured
Gemma 3 27B	24.8 tok/s	133 W63°CQ4_K_M	✓ Measured
DeepSeek-R1-Distill-Qwen-32B-abliterated	22.14 tok/s	127 W58°CQ4_K_M	✓ Measured
Olmo-3.1-32B-Think	22.13 tok/s	132 W59°CQ4_K_M	✓ Measured
Qwen3-32B	22.04 tok/s	133 W61°CQ4_K_M	✓ Measured
Qwen2.5-32B	21.98 tok/s	132 W60°CQ4_K_M	✓ Measured
Qwen2.5-Coder 32B	21.92 tok/s	132 W61°CQ4_K_M	✓ Measured
QwQ 32B	21.87 tok/s	132 W62°CQ4_K_M	✓ Measured
DeepSeek-R1 Distill 32B	21.85 tok/s	135 W63°CQ4_K_M	✓ Measured
Dolphin 2.9.1 Yi 1.5 34B	21.12 tok/s	133 W63°CQ4_K_M	✓ Measured
Qwen2.5-Coder 32B (Q3_K_M)	18.78 tok/s	133 W61°CQ3_K_M	✓ Measured
Llama 3.3 70B	✕ Won't fit needs ~46 GB	VRAM-gated at this precision	✓ Measured
DeepSeek-R1-Distill-Llama-70B	✕ Won't fit needs ~54 GB	VRAM-gated at this precision	✓ Measured
Hermes-4-70B	✕ Won't fit needs ~54 GB	VRAM-gated at this precision	✓ Measured
Qwen3-Coder-Next-abliterated	✕ Won't fit needs ~61 GB	VRAM-gated at this precision	✓ Measured
Laguna-XS-2.1	✕ Won't fit needs ~26 GB	VRAM-gated at this precision	✓ Measured
Llama-3.3-70B-Instruct-abliterated	✕ Won't fit needs ~54 GB	VRAM-gated at this precision	✓ Measured
Meta-Llama-3.1-70B	✕ Won't fit needs ~54 GB	VRAM-gated at this precision	✓ Measured
Nanbeige4.2-3B	✕ Won't fit needs ~4 GB	VRAM-gated at this precision	✓ Measured
Nemotron-3-Nano-30B-A3B	✕ Won't fit needs ~31 GB	VRAM-gated at this precision	✓ Measured
Qwen3-Coder-Next	✕ Won't fit needs ~61 GB	VRAM-gated at this precision	✓ Measured
Qwen3-Next-80B-A3B-Thinking	✕ Won't fit needs ~61 GB	VRAM-gated at this precision	✓ Measured
Qwen2.5-72B	✕ Won't fit needs ~60 GB	VRAM-gated at this precision	✓ Measured
Qwen3-Coder-Next	✕ Won't fit needs ~61 GB	VRAM-gated at this precision	✓ Measured
Qwen3-Next-80B-A3B-Thinking	✕ Won't fit needs ~61 GB	VRAM-gated at this precision	✓ Measured
Qwen3-Next-80B-A3B	✕ Won't fit needs ~61 GB	VRAM-gated at this precision	✓ Measured

Image Generation images/min 10

Sana 1.6B14.13

PixArt-Sigma XL9.93

Stable Diffusion XL6.38

Z-Image Turbo (1024px)6.32

Stable Diffusion 3.5 Medium4.11

Z-Image Turbo3.45

AuraFlow v0.31.39

Workload	Result	Telemetry	Data
Sana 1.6B	14.13 images/min	149 W48°C	✓ Measured
PixArt-Sigma XL	9.93 images/min	149 W55°C	✓ Measured
Stable Diffusion XL	6.38 images/min	15.8 GB peak149 W58°C9.4 s/img	✓ Measured
Z-Image Turbo (1024px)	6.32 images/min	149 W59°C	✓ Measured
Stable Diffusion 3.5 Medium	4.11 images/min	149 W66°C	✓ Measured
Z-Image Turbo	3.45 images/min	21.9 GB peak148 W63°C17.4 s/img	✓ Measured
AuraFlow v0.3	1.39 images/min	150 W68°C	✓ Measured
FLUX.1 dev	✕ Won't fit needs ~26 GB	VRAM-gated at this precision	✓ Measured
FLUX.1 Schnell	✕ Won't fit needs ~26 GB	VRAM-gated at this precision	✓ Measured
Stable Diffusion 3.5 Large	✕ Won't fit needs ~24 GB	VRAM-gated at this precision	✓ Measured

Fine-Tuning train tok/s 4

TinyLlama 1.1B LoRA5499.6

Qwen2.5 1.5B LoRA4214.4

SmolLM2 1.7B LoRA3668.3

Qwen2.5 7B LoRA1239

Workload	Result	Telemetry	Data
TinyLlama 1.1B LoRA	5499.6 train tok/s	147 W66°C	✓ Measured
Qwen2.5 1.5B LoRA	4214.4 train tok/s	150 W67°C	✓ Measured
SmolLM2 1.7B LoRA	3668.3 train tok/s	150 W67°C	✓ Measured
Qwen2.5 7B LoRA	1239 train tok/s	149 W68°C	✓ Measured

LLM Serving serve tok/s 4

TinyLlama 1.1B served3558.2

Qwen2.5 1.5B served2702.5

SmolLM2 1.7B served2242.7

Qwen2.5 7B served858.7

Workload	Result	Telemetry	Data
TinyLlama 1.1B served	3558.2 serve tok/s	154 W34°C	✓ Measured
Qwen2.5 1.5B served	2702.5 serve tok/s	160 W37°C	✓ Measured
SmolLM2 1.7B served	2242.7 serve tok/s	171 W39°C	✓ Measured
Qwen2.5 7B served	858.7 serve tok/s	203 W46°C	✓ Measured

Image to Video clips/min 3

Workload	Result	Telemetry	Data
CogVideoX-5B I2V	✕ Won't fit needs ~24 GB	VRAM-gated at this precision	✓ Measured
Stable Video Diffusion XT	✕ Won't fit needs ~16 GB	VRAM-gated at this precision	✓ Measured
Wan 2.2 TI2V-5B (image to video)	✕ Won't fit needs ~28 GB	VRAM-gated at this precision	✓ Measured

Image Editing images/min 2

Workload	Result	Telemetry	Data
FLUX.1 Kontext dev	✕ Won't fit needs ~26 GB	VRAM-gated at this precision	✓ Measured
Qwen-Image-Edit	✕ Won't fit needs ~42 GB	VRAM-gated at this precision	✓ Measured

Depth Estimation images/min 2

Workload	Result	Telemetry	Data
Depth Anything V2 Small	941.18 images/min	55 W33°C	✓ Measured
Depth Anything V2 Large	740.74 images/min	55 W33°C	✓ Measured

Segmentation images/min 2

Workload	Result	Telemetry	Data
SAM ViT-Base	398.72 images/min	65 W36°C	✓ Measured
SAM ViT-Huge	84.22 images/min	142 W40°C	✓ Measured

Vision Language images/min 2

Workload	Result	Telemetry	Data
Florence-2 Base	210.79 images/min	63 W36°C	✓ Measured
Florence-2 Large	117.94 images/min	86 W37°C	✓ Measured

Video Generation frames/s 1

Workload	Result	Telemetry	Data
Wan 2.2 5B (720p)	0.18 frames/s	18.4 GB peak133 W70°C266.2 s/clip	✓ Measured

Image to 3D assets/hour 1

Workload	Result	Telemetry	Data
TRELLIS Image-to-3D	293.6 assets/hour	197 W51°C	✓ Measured

Background Removal images/min 1

Workload	Result	Telemetry	Data
BiRefNet	432.33 images/min	68 W32°C	✓ Measured

Upscaling images/min 1

Workload	Result	Telemetry	Data
Swin2SR 4x Upscaler	22.31 images/min	148 W47°C	✓ Measured

Text to Speech x realtime 1

Workload	Result	Telemetry	Data
Kokoro TTS 82M	101.11 x realtime	66 W36°C	✓ Measured

Music Generation x realtime 1

Workload	Result	Telemetry	Data
MusicGen Small	1.2 x realtime	104 W48°C	✓ Measured

Speech to Text x realtime 1

Workload	Result	Telemetry	Data
Whisper large-v3	86.07 x realtime	116 W41°C	✓ Measured

Speculative Decoding x vs solo 1

Workload	Result	Telemetry	Data
Qwen2.5 1.5B + 0.5B draft	0.92 x vs solo		✓ Measured

How we measured this. Every result comes from our own pinned, reproducible AI suite, 12 workloads: the Qwen3-4B to Llama-70B LLM ladder (llama.cpp, Q4_K_M), SDXL / Z-Image / FLUX-dev generation, FLUX-Kontext / Qwen-Edit editing, and LTX / Wan video, run first-party on rented hardware with under 0.5% run-to-run variance. Peak VRAM, power draw, temperature and tokens-per-watt are captured per workload. “Won’t fit” rows are real data: where a model exceeds the card’s VRAM at the tested precision we record a hard gate rather than silently dropping to a smaller quant. Measured 2026-07-10 · harness 2.0.0.

NVIDIA A10G specifications

Architecture	Ampere
CUDA cores	9,216
VRAM	24GB GDDR6
Memory bus	384-bit
Memory bandwidth	600 GB/s
Boost clock	1,710 MHz
TDP	150 W
Process	Samsung 8nm
Interface	PCIe 4.0 x16
Release date	2021-11-01
Launch MSRP	$2,800

Verdict, capable, but 24GB sets the ceiling

NVIDIA A10G scores 6.4/100, #39 of 102. It ran 6 of 12; 5 exceeded its 24GB. Every figure here is our own measurement.

Relative performance: where the NVIDIA A10G lands

100% = this card, AI & Machine Learning headline metric (AI Score). #19 of 21 desktop cards in this vertical.

GPU	%	AI Score
NVIDIA L40	305%	19.5
NVIDIA A40	277%	17.7
NVIDIA A100 40GB SXM4	266%	17
NVIDIA A100 40GB PCIe	261%	16.7
NVIDIA A10G	100%	6.4
NVIDIA L4	78%	5
NVIDIA T4	45%	2.9

One tier up: NVIDIA A100 40GB PCIe (+161%)
One tier down: NVIDIA L4 (-22%) · compare head-to-head

← All AI & Machine Learning GPU rankings

What this card can build

Whole-job timings, composed from our measured per-model results on this card.

Workflow	Time	Energy	Basis
50-image depth pass	7 s	0.06 Wh	measured
30-minute podcast pass	77 s	2.03 Wh	all 3 stages measured
500-image masking run	6 min	14.07 Wh	measured
20-asset 3D game kit	7.6 min	21.16 Wh	all 2 stages measured
24-frame storyboard	9 min	19.54 Wh	all 2 stages measured
200-product catalogue cutout	9.6 min	22.58 Wh	all 2 stages measured
Full codebase review	21.3 min	43.41 Wh	measured
10 short social clips	50.7 min	108.73 Wh	all 3 stages measured

Can't run: 40-product photo shoot (needs FLUX.1 Kontext dev), 6-panel comic page (needs FLUX.1 dev), 20 long-form articles (needs Llama 3.3 70B), Character sheet, 12 poses (needs FLUX.1 dev), 100-photo restoration batch (needs FLUX.1 Kontext dev), 100-photo restore and enlarge (needs FLUX.1 Kontext dev), 40-product shoot, start to finish (needs FLUX.1 Kontext dev).