New — Multi-GPU Architect is live

Never guess your AI hardware again.

Aliteq is the intelligence layer for AI machines. Describe the models and the budget — get the exact build, the buy-vs-rent math, and real-world tokens/sec from 41,000+ live benchmarks.

No credit card · Trusted by 9,400+ builders worldwide

aliteq — rig planner

run qwen3-72b @ 30 tok/s

budget $2,200 · sometimes fine-tune

GPU2× RTX 3090 · 48GB

CPURyzen 9 7900

RAM64GB DDR5-6000

PSU1000W · 80+ Gold

Cooling2× 360mm AIO

Throughput

27 tok/s

Llama-class 70B, Q4

VRAM

48 GB

fits w/ 8k context

Total cost

$1,940

$260 under budget

vs cloud

Breakeven in 5.2 months vs RunPod

Tuned for the models you actually run

Llama 4Qwen3 72BDeepSeek V4Mistral LargeGemma 3GLM-5Stable DiffusionFLUX.1WhisperPhi-4Llama 4Qwen3 72BDeepSeek V4Mistral LargeGemma 3GLM-5Stable DiffusionFLUX.1WhisperPhi-4

The platform

Everything between a model and a machine.

Six tools that turn “what should I buy?” into a decision you can trust — backed by real benchmarks, not vibes.

AI Rig Planner

Describe the models, the throughput you want, and your budget. Get a complete, compatible build tuned to hit your numbers — not a generic parts dump.

Buy-vs-Rent Engine

Live GPU-cloud pricing vs. owning hardware, with a breakeven in months. Sometimes renting wins — we say so.

Multi-GPU Architect

Tensor & pipeline splits, PSU headroom, thermals, PCIe lanes. The math single-GPU tools quietly skip.

Throughput Index

Crowdsourced, verified tokens/sec across thousands of model × quant × hardware combos. Updated daily.

Price Radar

New and used GPU prices tracked worldwide, with alerts when your target part drops.

Compatibility Guard

Socket, memory, clearance, wattage — flagged before you spend a cent on parts that don't fit together.

41,283

configs benchmarked

312

models tracked

28

GPUs profiled

$1.2M

saved by builders

How it works

From prompt to parts list in under a minute.

01

Describe your workload

“Run Qwen3-72B at ~30 tok/s, occasional fine-tunes, ~$2,200.” Plain language is fine.

02

Aliteq runs the numbers

We match your goal against the live Index, then size GPUs, power, cooling and lanes for it.

03

Build, buy, or rent

Get the exact parts list, the buy-vs-rent breakeven, and where to get each part cheapest.

The Aliteq Throughput Index

Real tokens/sec. Real hardware.

The first crowdsourced index that benchmarks multi-GPUsetups, not just single cards. Submit a run, sharpen everyone's recommendations.

Submit your benchmark

llama-70b · Q4_K_M

tok/s

RTX 5090 · 32GB58
2× RTX 3090 · 48GB41
RTX 4090 · 24GB38
Mac M5 Max · 128GB19
DGX Spark · 128GB11

Illustrative figures shown for demo purposes.

Pricing

Free to plan. Cheap to go deep.

One good recommendation pays for years of Pro.

Hobby

For your first build.

$0forever
Start free
  • Unlimited rig plans
  • Browse the Throughput Index
  • Compatibility Guard
  • Community submissions
Most popular

Pro

For serious builders.

$19/mo
Start 14-day trial
  • Everything in Hobby
  • Buy-vs-Rent modeling
  • Saved & shareable builds
  • Price Radar alerts
  • Multi-GPU Architect

Studio

For teams & shops.

$99/mo
Talk to us
  • Everything in Pro
  • API + embeddable widget
  • Bulk fleet planning
  • Priority Index access
  • Seats for your team
Talked me out of a 5090 and into two used 3090s. Same speed on 70B, €1,400 cheaper. Wild that nothing else does this.
DDevon R.r/LocalLLaMA
The buy-vs-rent breakeven killed my impulse build. Renting an H100 was genuinely the right call for my volume.
AAastha P.ML engineer
Finally a tool that knows MoE models don't need the VRAM everyone assumes. The Index is addictive.
MMarco V.Homelab builder

FAQ

Questions, answered.

You describe the models you run (or want to), your throughput target, and your budget. Aliteq maps that against its live benchmark index — real tokens/sec from thousands of community-submitted configs — and returns the build that hits your numbers, not a generic parts list.

That's our whole edge. The Multi-GPU Architect computes tensor/pipeline splits, PSU headroom, thermal load, and PCIe-lane allocation — the stuff single-GPU calculators quietly ignore. Two used 3090s or one 5090? Aliteq shows you the real trade-off.

The Buy-vs-Rent engine models your monthly token volume against current GPU-cloud pricing (RunPod, Vast, Lambda and friends) and shows the breakeven point in months. Sometimes the honest answer is 'just rent' — and we'll tell you.

The Aliteq Throughput Index is crowdsourced: anyone can submit a run (model × quantization × hardware → tokens/sec). We dedupe, sanity-check, and publish. The more the community runs, the sharper everyone's recommendations get.

Yes. Planning a rig, browsing the Index, and the compatibility checks are free forever. Pro unlocks saved builds, price alerts, and the buy-vs-rent modeling at scale.

Stop guessing your next GPU.

Spec the perfect AI machine in under a minute — free, no card, no sign-up wall.

Plan my rig — free