Skip to content

GPU comparison

RTX 3090 vs RTX 4080.

Pick the RTX 3090 to fit bigger models on one GPU (24 GB vs 16 GB). Pick the RTX 4080 for speed: 37% more FP16 throughput.

Rent either on Runcrate

Live hourly price, then per GB of memory and per 100 FP16 TFLOPS.

Which is better for

Which to pick for each job. Memory fit and throughput, from the datasheets.

  • Hold a 70B model in 8-bit weights

    70 GB of 8-bit weights

    RTX 3090: 4 GPUs, more than one 2x node

    RTX 4080: 5 GPUs, more than one 2x node

    Either. Needs a multi-node cluster on either.

  • Serve an 8B model in 16-bit weights

    16 GB of 16-bit weights

    RTX 3090: 1 GPU, 8 GB spare

    RTX 4080: 2 GPUs (32 GB)

    RTX 3090. Fits on one GPU.

  • Fine-tune a 70B model with QLoRA

    About 46 GB with 4-bit weights

    RTX 3090: 2 GPUs (48 GB)

    RTX 4080: 3 GPUs, more than one 2x node

    RTX 3090. Fits in one node (2 GPUs).

  • Train on a full node

    FP16 non-tensor throughput per node

    RTX 3090: 2x, 142 TFLOPS, NVLink

    RTX 4080: 2x, 195 TFLOPS

    RTX 4080. 37% more FP16 throughput per GPU.

  • Throughput per watt

    FP16 TFLOPS per 100 W of TDP

    RTX 3090: 20 TFLOPS per 100 W

    RTX 4080: 30 TFLOPS per 100 W

    RTX 4080. 50% more FP16 per watt.

Compare another pair

vs

Ampere

NVIDIA RTX 3090

from$0.42/hr
Available now · 1 region
Memory
24 GB GDDR6X
FP16 non-tensor
71 TFLOPS
Bandwidth
936 GB/s
Full pricing details

Ada Lovelace

NVIDIA RTX 4080

Not in stock right now
Memory
16 GB GDDR6X
FP16 non-tensor
97.5 TFLOPS
Bandwidth
736 GB/s
Full pricing details

At a glance

Spec by spec. The longer bar is the better number.

Memory

NVIDIA RTX 309024 GB
NVIDIA RTX 408016 GB

Memory bandwidth

NVIDIA RTX 3090936 GB/s
NVIDIA RTX 4080736 GB/s

FP16 non-tensor

NVIDIA RTX 309071 TFLOPS
NVIDIA RTX 408097.5 TFLOPS

FP32

NVIDIA RTX 309035.6 TFLOPS
NVIDIA RTX 408048.7 TFLOPS

Full specs

The complete spec sheet.

SpecNVIDIA RTX 3090NVIDIA RTX 4080
ArchitectureAmpereAda Lovelace
Memory24 GB16 GB
Memory typeGDDR6XGDDR6X
Memory bandwidth936 GB/s736 GB/s
FP3235.6 TFLOPS48.7 TFLOPS
FP16 non-tensor71 TFLOPS97.5 TFLOPS
INT8284 TOPS390 TOPS
TDP350 W320 W
Form factorConsumerConsumer
NVLinkYesNo
GPUs per node22
Hourly price (live)from $0.42/hrNot in stock right now

Questions

RTX 3090 or RTX 4080. What buyers ask first.

Other comparisons

More pairs with the RTX 3090 or the RTX 4080.