GPU rankings

Scan the GPU field before you buy.

This page ranks GPUs by real local AI speed, VRAM headroom, and value so you can separate field leaders, sensible upgrades, and expensive noise without guessing.

Tracked field64

1 measured anchors inside a live catalog that still compares NVIDIA, AMD, Intel, and Apple in one place.

16GB+ shelf23

These are the options with enough memory headroom to stay relevant once you move past the smallest local models.

Value leader61.1 T/s / $100

GeForce GTX 1080 Ti is the current proof that price discipline can still beat hype in this catalog.

Jump to ranking

Featured buying paths

A sane entry, a strong value, and the true ceiling.

Choose the path that fits your budget and ambitions before diving into the full table.

Best ValueRTX 4070 Ti Super

16GB VRAM Highest T/s per $100

$799 MSRPView dossier

Best EfficiencyApple M3 Max

48GB unified 92W best per watt

$1,999 MSRPView dossier

Best OverallRTX 4090

Runs everything, no compromises

$1,599 MSRPView dossier

Ranked catalog

Compare the full field

Use filters to narrow the board, then compare real speed, VRAM headroom, and value without the usual shopping noise.

#1Leader

NVIDIA • S-Tier • estimated

GeForce RTX 5090

32GB VRAM • 575W • $1,999 MSRP

50/50 tracked model fits stay inside this envelope.

The new consumer ceiling for local AI speed and memory headroom.

Relative throughput100.0%

Decode372.4T/s

Efficiency18.6T/s per $100

VRAM32GBGDDR7

Fits tracked50/50catalog models

Check price

+2 more retailer links inside details

#2Podium

NVIDIA • S-Tier • estimated

GeForce RTX 4090

24GB VRAM • 450W • $1,599 MSRP

47/50 tracked model fits stay inside this envelope.

The undisputed king of local LLM inference. Runs every major open model at full precision.

Relative throughput84.0%

Decode312.8T/s

Efficiency19.6T/s per $100

VRAM24GBGDDR6X

Fits tracked47/50catalog models

Check price

+2 more retailer links inside details

#3Podium

NVIDIA • S-Tier • estimated

GeForce RTX 3090 Ti

24GB VRAM • 450W • $1,999 MSRP

46/50 tracked model fits stay inside this envelope.

The older 24GB brute-force option that still carries serious local AI weight.

Relative throughput77.2%

Decode287.4T/s

Efficiency14.4T/s per $100

VRAM24GBGDDR6X

Fits tracked46/50catalog models

Check price

+2 more retailer links inside details

#4Ranked

NVIDIA • A-Tier • estimated

GeForce RTX 5080

16GB VRAM • 360W • $999 MSRP

42/50 tracked model fits stay inside this envelope.

A cleaner high-end step when 4090-level spend feels unnecessary.

Relative throughput66.8%

Decode248.6T/s

Efficiency24.9T/s per $100

VRAM16GBGDDR7

Fits tracked42/50catalog models

Check price

+2 more retailer links inside details

#5Ranked

NVIDIA • B-Tier • estimated

GeForce RTX 3080 Ti

12GB VRAM • 350W • $1,199 MSRP

29/50 tracked model fits stay inside this envelope.

A very fast older flagship card that still pays a memory tax at 12GB.

Relative throughput62.3%

Decode232.1T/s

Efficiency19.4T/s per $100

VRAM12GBGDDR6X

Fits tracked29/50catalog models

Check price

+2 more retailer links inside details

#6Ranked

NVIDIA • B-Tier • estimated

GeForce RTX 3080

10GB VRAM • 320W • $699 MSRP

25/50 tracked model fits stay inside this envelope.

A classic throughput-heavy card whose 10GB memory ceiling now shows sooner.

Relative throughput57.7%

Decode214.8T/s

Efficiency30.7T/s per $100

VRAM10GBGDDR6X

Fits tracked25/50catalog models

Check price

+2 more retailer links inside details

#7Ranked

NVIDIA • A-Tier • estimated

GeForce RTX 4080 Super

16GB VRAM • 320W • $999 MSRP

39/50 tracked model fits stay inside this envelope.

The sweet spot for serious local AI. 16GB VRAM handles all 7B13B models with headroom.

Relative throughput53.3%

Decode198.4T/s

Efficiency19.9T/s per $100

VRAM16GBGDDR6X

Fits tracked39/50catalog models

Check price

+2 more retailer links inside details

#8Ranked

NVIDIA • A-Tier • estimated

GeForce RTX 5070 Ti

16GB VRAM • 300W • $749 MSRP

39/50 tracked model fits stay inside this envelope.

A practical upper-midrange card for buyers who want 16GB without overspending.

Relative throughput52.7%

Decode196.4T/s

Efficiency26.2T/s per $100

VRAM16GBGDDR7

Fits tracked39/50catalog models

Check price

+2 more retailer links inside details