CogVideoX-5B · 4 GPUs measured first-party · text-to-video · Updated October 2026

What GPU Do You Need for CogVideoX-5B?

CogVideoX-5B on 4 GPUs, measured first-party: NVIDIA GeForce RTX 5090 leads at 0.26 frames/s, RTX 3090 trails at 0.083 frames/s, and it peaked at 17GB of VRAM.

Benchmarked weights: THUDM/CogVideoX-5b

Fastest we measured
NVIDIA GeForce RTX 5090

NVIDIA GeForce RTX 5090

0.26 frames/s on CogVideoX-5B, the ceiling. Measured on our bench. 32GB of VRAM, $1,999 at launch.

Pros
  • 0.26 frames/s on CogVideoX-5B
  • 32GB, clears the CogVideoX-5B floor
Cons
  • 575W board rating
Best consumer card
NVIDIA GeForce RTX 4090

NVIDIA GeForce RTX 4090

0.17 frames/s on CogVideoX-5B, fastest card you can buy at retail. Measured on our bench. 24GB of VRAM, $1,599 at launch.

Pros
  • 0.17 frames/s on CogVideoX-5B
  • 24GB, clears the CogVideoX-5B floor
Cons
  • 450W board rating
Cheapest card that runs it
NVIDIA GeForce RTX 4080

NVIDIA GeForce RTX 4080

0.11 frames/s on CogVideoX-5B, lowest launch price that still fits. Measured on our bench. 16GB of VRAM, $1,199 at launch.

Pros
  • 0.11 frames/s on CogVideoX-5B
  • 16GB, clears the CogVideoX-5B floor
Cons
  • 320W board rating
Best value
NVIDIA GeForce RTX 3090

NVIDIA GeForce RTX 3090

0.08 frames/s on CogVideoX-5B, most speed per dollar. Measured on our bench. 24GB of VRAM, $1,499 at launch. That is 0.06 frames/s per $1,000 of launch price.

Pros
  • 0.08 frames/s on CogVideoX-5B
  • 24GB, clears the CogVideoX-5B floor
Cons
  • 350W board rating
0.26frames/s
Fastest: NVIDIA GeForce RTX 5090
measured, 3-run average
~17GB
VRAM needed (measured peak)
GPU-independent, applies to every card
4
GPUs measured
same pinned harness
0.0frames/s/W
Most efficient: NVIDIA GeForce RTX 5090
real power sampling, not TDP

What GPU Do You Need for CogVideoX-5B?, frames/s by GPU

NVIDIA GeForce RTX 5090
0.26 frames/s
NVIDIA GeForce RTX 4090
0.17 frames/s
NVIDIA GeForce RTX 4080
0.11 frames/s
NVIDIA GeForce RTX 3090
0.08 frames/s

Efficiency: frames/s per 100W drawn

NVIDIA GeForce RTX 5090
0.05 frames/s / 100W
NVIDIA GeForce RTX 4080
0.04 frames/s / 100W
NVIDIA GeForce RTX 4090
0.04 frames/s / 100W
NVIDIA GeForce RTX 3090
0.03 frames/s / 100W

Power is the average pulled during the run, sampled at 1Hz. The fastest card is often not the one here, and for anything left running this is the number that shows up on the bill.

Value: frames/s per $1,000 of MSRP

NVIDIA GeForce RTX 5090
0.13 frames/s / $1k
NVIDIA GeForce RTX 4090
0.11 frames/s / $1k
NVIDIA GeForce RTX 4080
0.09 frames/s / $1k
NVIDIA GeForce RTX 3090
0.06 frames/s / $1k

Launch price, not street price, so it ages. A speed leaderboard always crowns the most expensive card; this is the counterweight.

CogVideoX-5B. Measured text-to-video speed by GPU

NVIDIA GeForce RTX 50900.26
NVIDIA GeForce RTX 40900.17
NVIDIA GeForce RTX 40800.11
NVIDIA GeForce RTX 30900.08
GPUframes/sPeak VRAMAvg power
NVIDIA GeForce RTX 50900.2627.3GB572.4 W
NVIDIA GeForce RTX 40900.1715.6GB423.0 W
NVIDIA GeForce RTX 40800.1115.4GB270.1 W
NVIDIA GeForce RTX 30900.0815.4GB327.2 W

What the numbers show. Across 4 GPUs measured on our own bench, RTX 5090 is fastest at 0.26 frames/s. The slowest, RTX 3090, manages 0.08, so the spread is 3.1x from top to bottom. The fastest card with 16GB or less is RTX 4080 at 0.11 frames/s.

About CogVideoX-5B. CogVideoX-5B: from zai-org, 5.6B parameters, on Hugging Face since August 2024. 16,792 downloads in the last 30 days.

How it compares. RTX 4090: CogVideoX-5B 0.17 frames/s, LTX-Video (distilled) 6.7 (2B), CogVideoX-2B 0.52 (2B), Wan 2.1 1.3B 0.67 (1B), Wan 2.2 5B (720p) 0.43. All 4 beat CogVideoX-5B here.

Cost on a rented GPU. 1 minute of 24fps video of CogVideoX-5B: $0.59 on a RTX 3090 ($0.12/hr, 4.8 hours), $0.60 on a RTX 5090 ($0.39/hr, 93 min, 1.0x the cost).

CogVideoX-5B: cost per 1 minute of 24fps video on rented GPUs

NVIDIA GeForce RTX 3090$0.12/hr
NVIDIA GeForce RTX 5090$0.39/hr
NVIDIA GeForce RTX 4080$0.20/hr
NVIDIA GeForce RTX 4090$0.34/hr
GPUCheapest rateSpeed (frames/s)Cost per 1 minute of 24fps video
NVIDIA GeForce RTX 3090$0.12/hr0.08$0.59
NVIDIA GeForce RTX 5090$0.39/hr0.26$0.60
NVIDIA GeForce RTX 4080$0.20/hr0.11$0.73
NVIDIA GeForce RTX 4090$0.34/hr0.17$0.79

Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.

Speed tiers for CogVideoX-5B. under 1 frames/s: 4 (RTX 5090, RTX 4090, RTX 4080). 10 frames/s turns out a minute of 24fps video in under 2.5 minutes.

VRAM for CogVideoX-5B. Measured peak 15.4GB, so 24GB is the smallest common card size; smallest card it ran on: RTX 4080 (16GB).

Power on CogVideoX-5B. Most efficient: RTX 5090, 572W, 0.88 kWh per 1 minute of 24fps video. At $0.15/kWh: $0.13 per 1 minute of 24fps video.

Our verdict

Fastest on CogVideoX-5B: NVIDIA GeForce RTX 5090, 0.26 frames/s. Cheapest consumer card that ran it: NVIDIA GeForce RTX 4080 ($1,199, 0.11 frames/s). Cheapest to rent per job: NVIDIA GeForce RTX 3090, $0.59 per 1 minute of 24fps video.

FAQ

What GPU do I need to run CogVideoX-5B?
About 16GB. Cheapest consumer card that ran it: NVIDIA GeForce RTX 4080 (16GB, 0.11 frames/s).
How fast is CogVideoX-5B on the NVIDIA GeForce RTX 5090?
0.26 frames/s at 572W, faster than every datacenter card we ran it on.
How much does it cost to run CogVideoX-5B in the cloud?
$0.59 per 1 minute of 24fps video on a NVIDIA GeForce RTX 3090 at $0.12/hr, cheapest of 4 rentable cards we measured.
Can I run CogVideoX-5B on a 12GB, 16GB or 24GB card?
It used 15.4GB at the precision we tested. 12GB: no; 16GB: no; 24GB: yes.
Is the RTX 4090 or the RTX 3090 faster for CogVideoX-5B?
The RTX 4090: 0.17 vs 0.08 frames/s, 105% faster on our bench.
Should I buy or rent a RTX 5090 for CogVideoX-5B?
Its $1,999 launch price buys 5,139 rented hours at $0.39/hr, enough for about 3,327x 1 minute of 24fps video of CogVideoX-5B. Buy only if you'll run more than that.

How we test

CogVideoX-5B in diffusers, bf16, one warmup clip and three timed clips from the same prompt, measured as frames of finished video per second, with CPU offload only on cards below the model's full-GPU size. Text-to-video is compute-bound and memory-hungry; cards that need CPU offload lose far more speed than their raw compute suggests.