What's the price difference between NVIDIA L40S and NVIDIA RTX 4090?

On Runcrate today, NVIDIA L40S starts at $0.54/hr per GPU on a 2-GPU machine and NVIDIA RTX 4090 starts at $0.57/hr per GPU on an 8-GPU machine; NVIDIA L40S is cheaper by 5%.

Can I deploy NVIDIA L40S or NVIDIA RTX 4090 on Runcrate today?

NVIDIA L40S is on demand in 6 regions from $0.54/hr per GPU on a 2-GPU machine, billed by the second on most GPUs. NVIDIA RTX 4090 is on demand in 3 regions from $0.57/hr per GPU on an 8-GPU machine, billed by the second on most GPUs.

How much VRAM does each GPU have?

NVIDIA L40S has 48GB GDDR6. NVIDIA RTX 4090 has 24GB GDDR6X.

L40S or RTX 4090: which holds a 70B model in 8-bit weights?

It needs 70 GB of 8-bit weights. On the L40S: 2 GPUs (96 GB). On the RTX 4090: 4 GPUs, more than one 2x node. Pick the L40S: fits in one node (2 GPUs).

GPU comparison

L40S vs RTX 4090.

The L40S is ahead on memory (48 GB vs 24 GB) and speed (2.2x the FP16 throughput). The RTX 4090 makes sense when your model fits in 24 GB and its hourly price is lower.

Launch L40S Reserve a cluster

Rent either on Runcrate

L40SCheaperfrom $0.54/hr per GPU
on a 2-GPU machine
$0.011 per GB/hr$0.15 per 100 TFLOPS/hr
Launch L40S
RTX 4090from $0.57/hr per GPU
on an 8-GPU machine
$0.024 per GB/hr$0.35 per 100 TFLOPS/hr
Launch RTX 4090

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.

JobL40SRTX 4090Pick

Hold a 70B model in 8-bit weights
70 GB of 8-bit weights
L40S: 2 GPUs (96 GB)
RTX 4090: 4 GPUs, more than one 2x node
L40S. Fits in one node (2 GPUs).
Serve an 8B model in 16-bit weights
16 GB of 16-bit weights
L40S: 1 GPU, 32 GB spare
RTX 4090: 1 GPU, 8 GB spare
RTX 4090. Same GPU count, 17% more memory bandwidth.
Fine-tune a 70B model with QLoRA
About 46 GB with 4-bit weights
L40S: 1 GPU, 2 GB spare
RTX 4090: 2 GPUs (48 GB)
L40S. Fits on one GPU.
Train on a full node
FP16 tensor throughput per node
L40S: 4x, 1,448 TFLOPS
RTX 4090: 2x, 330 TFLOPS
L40S. 4 GPUs per node, not 2.
Throughput per watt
FP16 TFLOPS per 100 W of TDP
L40S: 103 TFLOPS per 100 W
RTX 4090: 37 TFLOPS per 100 W
L40S. 2.8x the FP16 per watt.

Compare another pair

Ada Lovelace

NVIDIA L40S

from$0.54/hr per GPUCheaper

on a 2-GPU machine

Available now · 6 regions

Memory: 48 GB GDDR6
FP16 tensor: 362 TFLOPS
Bandwidth: 864 GB/s

Launch L40S Reserve a cluster

Full pricing details

Ada Lovelace

NVIDIA RTX 4090

from$0.57/hr per GPU

on an 8-GPU machine

Available now · 3 regions

Memory: 24 GB GDDR6X
FP16 tensor: 165.2 TFLOPS
Bandwidth: 1.0 TB/s

Launch RTX 4090 Reserve a cluster

Full pricing details

At a glance

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

Memory

NVIDIA L40S48 GB

NVIDIA RTX 409024 GB

Memory bandwidth

NVIDIA L40S864 GB/s

NVIDIA RTX 40901.0 TB/s

FP16 tensor

NVIDIA L40S362 TFLOPS

NVIDIA RTX 4090165.2 TFLOPS

Hourly price, from (live)

NVIDIA L40S$0.54/hr

NVIDIA RTX 4090$0.57/hr

Full specs

The complete spec sheet.

Spec	NVIDIA L40S	NVIDIA RTX 4090
Architecture	Ada Lovelace	Ada Lovelace
Memory	48 GB	24 GB
Memory type	GDDR6	GDDR6X
Memory bandwidth	864 GB/s	1.0 TB/s
FP32	91.6 TFLOPS	82.6 TFLOPS
FP16 tensor	362 TFLOPS	165.2 TFLOPS
INT8	733 TOPS	661 TOPS
TDP	350 W	450 W
Form factor	PCIe	Consumer
NVLink	No	No
GPUs per node	4	2
Hourly price (live)	from $0.54/hr per GPU on a 2-GPU machine	from $0.57/hr per GPU on an 8-GPU machine

Questions

L40S or RTX 4090. What buyers ask first.

Other comparisons