Can I deploy NVIDIA GH200 or NVIDIA RTX 4090 on Runcrate today?

NVIDIA GH200 is not in stock on Runcrate right now; leave your email on its pricing page to register interest. NVIDIA RTX 4090 is on demand in 4 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 GH200 has 96GB HBM3. NVIDIA RTX 4090 has 24GB GDDR6X.

GH200 or RTX 4090: which holds a 70B model in 8-bit weights?

It needs 70 GB of 8-bit weights. On the GH200: 1 GPU, 26 GB spare. On the RTX 4090: 4 GPUs, more than one 2x node. Pick the GH200: fits on one GPU.

GPU comparison

GH200 vs RTX 4090.

The GH200 is ahead on memory (96 GB vs 24 GB) and speed (6.0x the FP16 throughput and up to 4.0x the memory bandwidth). The RTX 4090 makes sense when your model fits in 24 GB and its hourly price is lower.

Launch RTX 4090 Reserve a cluster

Rent either on Runcrate

GH200Not in stock right now
Register interest GH200 pricing
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 RTX 4090 pricing

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.

JobGH200RTX 4090Pick

Hold a 70B model in 8-bit weights
70 GB of 8-bit weights
GH200: 1 GPU, 26 GB spare
RTX 4090: 4 GPUs, more than one 2x node
GH200. Fits on one GPU.
Serve an 8B model in 16-bit weights
16 GB of 16-bit weights
GH200: 1 GPU, 80 GB spare
RTX 4090: 1 GPU, 8 GB spare
GH200. Same GPU count, up to 4.0x the memory bandwidth.
Fine-tune a 70B model with QLoRA
About 46 GB with 4-bit weights
GH200: 1 GPU, 50 GB spare
RTX 4090: 2 GPUs (48 GB)
GH200. Fits on one GPU.
Train on a full node
FP16 tensor throughput per node
GH200: 1x, 990 TFLOPS
RTX 4090: 2x, 330 TFLOPS
GH200. 6.0x the FP16 throughput per GPU.
Throughput per watt
FP16 TFLOPS per 100 W of TDP
GH200: Superchip power, not a GPU TDP
RTX 4090: 37 TFLOPS per 100 W
Either. Not comparable.

Compare another pair

Hopper

NVIDIA GH200

Not in stock right now

Memory: 96 GB HBM3
FP16 tensor: 990 TFLOPS
Bandwidth: up to 4.0 TB/s

Full pricing details

Ada Lovelace

NVIDIA RTX 4090

from$0.57/hr per GPU

on an 8-GPU machine

Available now · 4 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 GH20096 GB

NVIDIA RTX 409024 GB

Memory bandwidth

NVIDIA GH200up to 4.0 TB/s

NVIDIA RTX 40901.0 TB/s

FP16 tensor

NVIDIA GH200990 TFLOPS

NVIDIA RTX 4090165.2 TFLOPS

FP32

NVIDIA GH20067 TFLOPS

NVIDIA RTX 409082.6 TFLOPS

Full specs

The complete spec sheet.

Spec	NVIDIA GH200	NVIDIA RTX 4090
Architecture	Hopper	Ada Lovelace
Memory	96 GB	24 GB
Memory type	HBM3	GDDR6X
Memory bandwidth	up to 4.0 TB/s	1.0 TB/s
FP32	67 TFLOPS	82.6 TFLOPS
FP16 tensor	990 TFLOPS	165.2 TFLOPS
INT8	1979 TOPS	661 TOPS
Power	up to 1000 W per superchip (CPU, GPU and memory)	450 W per GPU (TDP)
Form factor	Superchip	Consumer
NVLink	One GPU	No
GPUs per node	1	2
Hourly price (live)	Not in stock right now	from $0.57/hr per GPU on an 8-GPU machine

Questions

GH200 or RTX 4090. What buyers ask first.

Other comparisons