Hardware to run Qwen 3.6 35B-A3B (MoE)

Apr 2026 release. 35B / 3B active MoE — beats Gemma 4-31B on agentic coding, matches Sonnet on most vision tasks. Native 262 K context (extensible to 1 M), ~18 GB at Q4. The new local-coding king under 200 B.

Quantization Q2 Q2_K Q4 Q4_K_M Q5 Q5_K_M Q8 Q8_0

Availability ⚡ Only plug & play

Cheapest

Single Tesla P40 24 GB (used) build

NVIDIA · desktop tower

$750

tokens / secQ4

14B 15 t/s

30B 12 t/s

70B —

Memory24 GB · 23 usable

Bandwidth347 GB/s

Idle / Active10 W / 250 W

Sticker$750

Why: Lowest sticker that still fits Qwen 3.6 35B-A3B (MoE) ($750 USD).

AmazoneBay (used)Newegg

📺 Reviews on YouTube

▶ Dell PowerEdge R730 Tesla Nvidia P40 vs P100 AI/LLM Bake Off

Fastest

DGX B200 — 8× B200 server (1.44 TB HBM3e)

NVIDIA · 10U DGX server

$475,000

tokens / secQ4

14B 420 t/s

30B 270 t/s

70B 180 t/s

Memory1440 GB · 1404 usable

Bandwidth8000 GB/s

Idle / Active900 W / 10200 W

Sticker$475,000

Why: Highest measured tg/s — 810 t/s on Qwen 3.6 35B-A3B (MoE)-class models at Q4.

NVIDIA DGX B200SuperMicro HGX B200Lambda Labs

📺 Reviews on YouTube

▶ Inside a 1.44 TB HBM3e NVIDIA HGX B200 AI Server from ASRock Rack

All-rounder

2× Strix Halo cluster (256 GB unified)

AMD · mini-PC pair

$5,600

tokens / secQ4

14B 40 t/s

30B 24 t/s

70B 6.0 t/s

Memory256 GB · 192 usable

Bandwidth256 GB/s

Idle / Active16 W / 240 W

Sticker$5,600

Why: Top quartile across speed, value, memory headroom, and efficiency — the "buy this if unsure" pick.

Amazon (Strix Halo Mini PC)GMKtecFramework

📺 Reviews on YouTube

▶ AMD Strix Halo / Ryzen AI Max+ 395 — honest review (single-node baseline) ▶ Kimi-K2 (1T) / GLM 4.7 (355B) on a 4-node Strix Halo cluster — 512 GB unified memory

Best value

Tesla V100 32 GB SXM2 mod build

NVIDIA · desktop tower

$900

tokens / secQ4

14B 50 t/s

30B 27 t/s

70B —

Memory32 GB · 31 usable

Bandwidth900 GB/s

Idle / Active33 W / 300 W

Sticker$900

Why: Best $/tg-per-second — ~$11 per t/s.

AmazoneBay V100 SXM2 32 GBeBay SXM2 → PCIe adapter

📺 Reviews on YouTube

▶ Expensive RTX 5090 for LLMs? No — use V100 SXM2 + Z8 G4 instead ▶ I built my own AI — A.I.D.E.N. (32 GB V100 SXM2 + HP Z8 G4)

Best CUDA

DGX H200 — 8× H200 server (1.13 TB HBM3e)

NVIDIA · 8U DGX / HGX server rack

$380,000

tokens / secQ4

14B 390 t/s

30B 250 t/s

70B 170 t/s

Memory1128 GB · 1100 usable

Bandwidth4800 GB/s

Idle / Active700 W / 6500 W

Sticker$380,000

Why: Strongest CUDA-only software stack among fitting builds.

NVIDIA DGX H200SuperMicro HGXLambda Labs

📺 Reviews on YouTube

▶ Inside the SUPER NVIDIA H200 Server From Supermicro (8U HGX H200)

Most VRAM

12× RTX Pro 6000 Blackwell rack (1152 GB)

NVIDIA · 8U server rack (multi-node, 1-2 chassis)

$118,000

tokens / secQ4

14B 340 t/s

30B 250 t/s

70B 170 t/s

Memory1152 GB · 1116 usable

Bandwidth1792 GB/s

Idle / Active340 W / 7400 W

Sticker$118,000

Why: 1116 GB usable — most headroom for batching and longer contexts.

AmazonCoreWeaveSuperMicroSHI enterprise

Efficient

MacBook Pro M4 Pro 48 GB

Apple · laptop

$2,899

tokens / secQ4

14B 28 t/s

30B 14 t/s

70B 6.0 t/s

Memory48 GB · 40 usable

Bandwidth273 GB/s

Idle / Active5 W / 70 W

Sticker$2,899

Why: 70 W active — lowest power draw of the fitting builds.

AmazonAppleB&H Photo

📺 Reviews on YouTube

▶ FREE Local LLMs on Apple Silicon — Fast! (AZisk)

Cheapest

Single Intel Arc Pro B70 build

Intel · desktop tower

$1,799

tokens / secQ4

14B 40 t/s

30B 25 t/s

70B —

Memory32 GB · 31 usable

Bandwidth608 GB/s

Idle / Active18 W / 220 W

Sticker$1,799

Why: Lowest sticker that still fits Qwen 3.6 35B-A3B (MoE) ($1.8k USD).

AmazonNeweggB&H Photo

📺 Reviews on YouTube

▶ Should you buy Intel Arc Pro B70 for Local AI? Gemma 4 benchmark ▶ Intel Arc Pro B70 32GB for Local LLMs — llama.cpp & vLLM benchmarks

Fastest

Single AMD Instinct MI355X 288 GB workstation

AMD · 4U server (OAM, liquid-cooled)

$28,000

tokens / secQ4

14B 270 t/s

30B 160 t/s

70B —

Memory288 GB · 282 usable

Bandwidth8000 GB/s

Idle / Active140 W / 1400 W

Sticker$28,000

Why: Highest measured tg/s — 480 t/s on Qwen 3.6 35B-A3B (MoE)-class models at Q4.

Dell XE9712AMD MI350 seriesSupermicro UBB

📺 Reviews on YouTube

▶ AMD's MI350/355X Advancing AI Event Recap ▶ AMD AI Event w/ Craft Computing — MI350/355X Launch, ROCm 7 ▶ AMD Visit and Tour! Featuring ROCm 7 and AMD Instinct

All-rounder

MacBook Pro M5 Max 128 GB

Apple · laptop

$5,399

tokens / secQ4

14B 50 t/s

30B 30 t/s

70B 14 t/s

Memory128 GB · 108 usable

Bandwidth614 GB/s

Idle / Active6 W / 95 W

Sticker$5,399

Why: Top quartile across speed, value, memory headroom, and efficiency — the "buy this if unsure" pick.

AmazonAppleB&H Photo

📺 Reviews on YouTube

▶ Apple's New M5 Max Changes the Local AI Story (AZisk) ▶ I Linked 3 M5 MacBooks to Run "Impossible" AI (AZisk) ▶ I was WRONG about M5 MacBook Pro vs M4 Max & M3 Ultra for Local AI (xCreate)

Best value

Single RTX 5090 build

NVIDIA · desktop tower

$4,900

tokens / secQ4

14B 124 t/s

30B 70 t/s

70B —

Memory32 GB · 31 usable

Bandwidth1792 GB/s

Idle / Active30 W / 520 W

Sticker$4,900

Why: Best $/tg-per-second — ~$23 per t/s.

AmazonNeweggB&H Photo

📺 Reviews on YouTube

▶ RTX 3090 vs 4090 vs 5090 vs Mac M5 Max — Qwen3.6-27B benchmark ▶ RTX 3090 vs 4090 vs 5090 vs Mac M5 Max — Qwen3.6-35B-A3B benchmark ▶ Not even close — LLMs on RTX 5090 vs others (AZisk)

Best CUDA

Single B200 180 GB workstation

NVIDIA · workstation / 4U server

$47,000

tokens / secQ4

14B 225 t/s

30B 135 t/s

70B 75 t/s

Memory180 GB · 176 usable

Bandwidth8000 GB/s

Idle / Active100 W / 1000 W

Sticker$47,000

Why: Strongest CUDA-only software stack among fitting builds.

NVIDIA B200 partnersSHI enterpriseLambda Labs

📺 Reviews on YouTube

▶ AI Lab: NVIDIA B200 vs GB200 explained | GPU architecture for LLMs ▶ Inside a NEW AI Cluster — Tour with NVIDIA B200

Most VRAM

Mac Studio M3 Ultra 512 GB

Apple · small desktop

$14,199

tokens / secQ4

14B 70 t/s

30B 38 t/s

70B 18 t/s

Memory512 GB · 480 usable

Bandwidth819 GB/s

Idle / Active12 W / 220 W

Sticker$14,199

Why: 480 GB usable — most headroom for batching and longer contexts.

Apple RefurbB&H Photo (used)eBay (used)

📺 Reviews on YouTube

▶ M3 Ultra Mac Studio Review ▶ M3 Ultra vs RTX 5090 — The Final Battle

Efficient

MacBook Pro M4 Pro 48 GB

Apple · laptop

$2,899

tokens / secQ4

14B 28 t/s

30B 14 t/s

70B 6.0 t/s

Memory48 GB · 40 usable

Bandwidth273 GB/s

Idle / Active5 W / 70 W

Sticker$2,899

Why: 70 W active — lowest power draw of the fitting builds.

AmazonAppleB&H Photo

📺 Reviews on YouTube

▶ FREE Local LLMs on Apple Silicon — Fast! (AZisk)

Every other build that runs Qwen 3.6 35B-A3B (MoE)

52 additional builds fit Qwen 3.6 35B-A3B (MoE) at Q4_K_M (22 GB usable minimum), sorted by sticker price.

Build	Price	Memory	Bandwidth	tg/s (Q4)	Active W	5-yr power
Single AMD Instinct MI50 32 GB (used) buildAMD · desktop tower	$900	32 / 31 GB	1024 GB/s	—	300 W	$1.0k
Single RTX 3090 (used) buildNVIDIA · desktop tower	$1.8k	24 / 23 GB	936 GB/s	84 t/s	350 W	$1.2k
Single AMD Radeon AI Pro R9700 32 GB buildAMD · desktop tower	$2.0k	32 / 31 GB	640 GB/s	—	300 W	$1.1k
Mac Studio M4 Max 36 GBApple · small desktop	$2.5k	36 / 28 GB	546 GB/s	48 t/s	130 W	$453
Quad AMD MI50 32 GB (128 GB) homelab buildAMD · rack/large tower	$2.5k	128 / 122 GB	1024 GB/s	50 t/s	1200 W	$4.2k
Quad Tesla P40 (96 GB) homelab buildNVIDIA · rack/large tower	$2.7k	96 / 92 GB	347 GB/s	24 t/s	1000 W	$3.5k
AMD Ryzen AI Max+ 395 (128 GB)AMD · mini desktop / laptop	$2.8k	128 / 96 GB	256 GB/s	77 t/s	120 W	$420
Dual RTX 3090 (used) buildNVIDIA · desktop tower	$2.8k	48 / 46 GB	936 GB/s	105 t/s	700 W	$2.4k
MacBook Pro M5 Pro 48 GBApple · laptop	$3.1k	48 / 40 GB	307 GB/s	92 t/s	75 W	$263
Single RTX 4090 buildNVIDIA · desktop tower	$3.2k	24 / 23 GB	1008 GB/s	126 t/s	410 W	$1.4k
Dual Intel Arc Pro B70 buildIntel · desktop tower	$3.2k	64 / 62 GB	608 GB/s	55 t/s	380 W	$1.4k
Single AMD Radeon Pro W7800 32 GB buildAMD · workstation	$3.5k	32 / 31 GB	576 GB/s	54 t/s	260 W	$920
Dual AMD Radeon AI Pro R9700 build (64 GB)AMD · workstation	$3.7k	64 / 62 GB	640 GB/s	90 t/s	600 W	$2.1k
ASUS Ascent GX10 (128 GB)ASUS · small desktop	$4.0k	128 / 119 GB	273 GB/s	95 t/s	240 W	$903
MacBook Pro M4 Max 64 GBApple · laptop	$4.0k	64 / 54 GB	410 GB/s	107 t/s	90 W	$315
Dell Pro Max with GB10 (128 GB)Dell · small desktop	$4.1k	128 / 119 GB	273 GB/s	95 t/s	240 W	$887
MacBook Pro M5 Max 64 GBApple · laptop	$4.1k	64 / 54 GB	614 GB/s	131 t/s	95 W	$332
Lenovo ThinkStation PGX (128 GB)Lenovo · small desktop	$4.1k	128 / 119 GB	273 GB/s	95 t/s	160 W	$650
MSI EdgeXpert MS-C931 (128 GB)MSI · small desktop	$4.7k	128 / 119 GB	273 GB/s	95 t/s	240 W	$887
Mac Studio M4 Max 128 GBApple · small desktop	$4.7k	128 / 112 GB	546 GB/s	107 t/s	130 W	$453
NVIDIA DGX Spark (128 GB)NVIDIA · small desktop	$4.7k	128 / 119 GB	273 GB/s	95 t/s	240 W	$887
Single RTX A6000 48 GB (Ampere) buildNVIDIA · workstation	$4.7k	48 / 46 GB	768 GB/s	72 t/s	300 W	$1.0k
Single AMD Radeon Pro W7900 48 GB buildAMD · workstation	$5k	48 / 46 GB	864 GB/s	66 t/s	295 W	$1.0k
Mac Studio M3 Ultra 96 GBApple · small desktop	$5k	96 / 80 GB	819 GB/s	90 t/s	180 W	$624
Quad Intel Arc Pro B70 buildIntel · rack/large tower	$6k	128 / 124 GB	608 GB/s	55 t/s	700 W	$2.5k
Quad RTX 3090 (used) buildNVIDIA · rack/large tower	$6k	96 / 92 GB	936 GB/s	120 t/s	1400 W	$4.9k
Single RTX Pro 5000 Blackwell 48 GB buildNVIDIA · workstation	$7k	48 / 46 GB	1344 GB/s	—	300 W	$1.1k
Single RTX 6000 Ada 48 GB buildNVIDIA · workstation	$8k	48 / 46 GB	960 GB/s	120 t/s	300 W	$1.0k
Mac Studio M3 Ultra 256 GBApple · small desktop	$8k	256 / 232 GB	819 GB/s	90 t/s	180 W	$624
Dual AMD Radeon Pro W7900 buildAMD · workstation	$9k	96 / 92 GB	864 GB/s	99 t/s	600 W	$2.1k
2× DGX Spark cluster (256 GB unified, CUDA)NVIDIA · two desktops, 200 G interconnect	$10k	256 / 240 GB	273 GB/s	150 t/s	460 W	$1.7k
Dual RTX 5090 buildNVIDIA · rack/large tower	$10k	64 / 62 GB	1792 GB/s	270 t/s	1050 W	$3.6k
Octuple Intel Arc Pro B70 clusterIntel · rack/large tower	$11k	256 / 248 GB	608 GB/s	126 t/s	1450 W	$5k
4× Strix Halo cluster (512 GB unified)AMD · rack of 4 mini-PCs, 10 GbE fabric	$12k	512 / 384 GB	256 GB/s	96 t/s	480 W	$1.7k
Single RTX Pro 6000 Blackwell 96 GB buildNVIDIA · workstation	$12k	96 / 93 GB	1792 GB/s	255 t/s	600 W	$2.1k
Tinybox Red (6× 7900 XTX, 144 GB)tinycorp · 12U pedestal	$15k	144 / 138 GB	960 GB/s	165 t/s	1500 W	$5k
4× DGX Spark cluster (512 GB unified, CUDA)NVIDIA · rack of 4 desktops	$20k	512 / 488 GB	273 GB/s	165 t/s	920 W	$3.4k
8× Strix Halo cluster (1024 GB unified)AMD · rack of 8 mini-PCs, 10/25 GbE fabric	$23k	1024 / 768 GB	256 GB/s	120 t/s	960 W	$3.4k
Dual RTX Pro 6000 Blackwell buildNVIDIA · workstation	$24k	192 / 188 GB	1792 GB/s	300 t/s	1100 W	$3.8k
Single AMD Instinct MI325X 256 GB workstationAMD · workstation / 4U server (OAM)	$25k	256 / 250 GB	6000 GB/s	390 t/s	1000 W	$3.6k
Tinybox Green (6× RTX 4090, 144 GB)tinycorp · 12U pedestal	$25k	144 / 138 GB	1008 GB/s	225 t/s	2200 W	$8k
2× Mac Studio M3 Ultra 512 GB cluster (TB5 / MLX)Apple · two desktops, Thunderbolt 5 RDMA	$28k	1024 / 960 GB	819 GB/s	90 t/s	440 W	$1.5k
Single AMD Instinct MI300X 192 GB workstationAMD · workstation	$30k	192 / 188 GB	5300 GB/s	300 t/s	750 W	$2.8k
Single H100 80 GB workstationNVIDIA · workstation	$32k	80 / 78 GB	3350 GB/s	270 t/s	700 W	$2.5k
Quad RTX Pro 6000 Blackwell build (384 GB)NVIDIA · workstation / 4U pedestal	$38k	384 / 372 GB	1792 GB/s	450 t/s	2200 W	$8k
Single H200 141 GB workstationNVIDIA · workstation / 2U server	$40k	141 / 138 GB	4800 GB/s	375 t/s	700 W	$2.5k
Tinybox Pro (8× RTX 4090, 192 GB)tinycorp · 12U pedestal	$40k	192 / 184 GB	1008 GB/s	285 t/s	3200 W	$11k
8× DGX Spark cluster (1024 GB unified, CUDA)NVIDIA · rack of 8 desktops, 200 GbE fabric	$44k	1024 / 976 GB	273 GB/s	216 t/s	1840 W	$7k
Tinybox Green v2 (4× RTX 5090, 128 GB)tinycorp · 12U pedestal	$45k	128 / 124 GB	1792 GB/s	330 t/s	2300 W	$8k
8× RTX Pro 6000 Blackwell server (768 GB)NVIDIA · 4U server (e.g. SuperMicro AS-4125GS-TNRT)	$78k	768 / 744 GB	1792 GB/s	660 t/s	4800 W	$16k
8× H100 80 GB serverNVIDIA · server rack	$280k	640 / 620 GB	3350 GB/s	540 t/s	5600 W	$20k
NVIDIA RTX Spark (128 GB)NVIDIA · OEM laptops + small desktops	—	128 / 119 GB	300 GB/s	—	— W	$756

Last updated 2026-07-11