RTX 3090 or RTX 4090: same 24 GB, different speed

Both cards have the same memory, so for models that fit in 24 GB the deciding factors are speed and hourly price. The RTX 4090 is newer (Ada Lovelace), faster and supports FP8; the RTX 3090 is cheaper and enough for many tasks.

Specifications

	RTX 3090 24 GB	RTX 4090 48 GB
Video memory	24 GB	48 GB
Memory type	GDDR6X	GDDR6X
Architecture	Ampere	Ada Lovelace
Price from	21₽/h	62₽/h
Servers now	250	66

Prices and availability come from the live catalog and update automatically.

Which one to pick

Pick RTX 3090 if

budget generation and experiments
LLMs up to 32B at int4
simple jobs where speed is not critical

Pick RTX 4090 if

frequent generation where pace matters
FP8 and newer frameworks
LoRA training with short runs

Cost example

RTX 3090, per hour: about 21 ₽
RTX 4090, per hour: about 62 ₽

Estimate at the current per-GPU hourly price. Billed per minute, the price is fixed at launch.

Questions and answers

Does the 3090 support FP8?: Ampere has no native FP8, so FP8 model variants run slower or need workarounds.
What should I compare?: The hourly price and the time one job takes. If the 4090 finishes twice as fast for a price less than twice as high, it is the cheaper option per job.

RTX 3090 or RTX 4090: same 24 GB, different speed

Specifications

Which one to pick

Pick RTX 3090 if

Pick RTX 4090 if

Cost example

Questions and answers

Compare with others

RTX 3090 or RTX 4090: same 24 GB, different speed

Specifications

Which one to pick

Pick RTX 3090 if

Pick RTX 4090 if

Cost example

Questions and answers

Compare with others