~/gpu/rtx-4080 vs rtx-4090

RTX 4080 manufacturerRTX 4080vsRTX 4090 manufacturerRTX 4090

Head-to-head for local LLM inference. The honest comparison: VRAM, bandwidth, compute, and which of the 30 catalog models actually fit on each.

The specs.

$ diff specs rtx-4080 rtx-4090
Stat
rtx-4080
rtx-4090
Δ
VRAM
16 GB
24 GB
+50%
Memory bandwidth
717 GB/s
1,008 GB/s
+41%
FP16 compute
390 TFLOPS
330 TFLOPS
-15%
Weights budget at 8K ctx
12 GB
18 GB
+50%

Model fit difference.

$ models that change with the card
Fits on both
16of 30
Only on rtx-4080
0
Only on rtx-4090
5

// showing 12 of 30 models; differing fits first

Model
rtx-4080
rtx-4090
Qwen 2.5 32B32.5B
overQ4_K_M
fitsAWQ 4-BIT
Qwen 2.5 Coder 32B32.5B
overQ4_K_M
fitsAWQ 4-BIT
Qwen3 30B A3B30.5B
overQ4_K_M
fitsQ4_K_M
Qwen 3.6 35B A3B35B
overQ4_K_M
fitsQ3_K_M
Yi 34B34B
overQ4_K_M
fitsQ3_K_M
Llama 3.2 1B1.23B
fitsFP16/BF16
fitsFP16/BF16
Llama 3.2 3B3.21B
fitsFP16/BF16
fitsFP16/BF16
Llama 3.1 8B8B
fitsQ8_0
fitsFP16/BF16
Llama 3.3 70B70.6B
overQ4_K_M
overQ4_K_M
Llama 3.1 405B405B
overQ4_K_M
overQ4_K_M
Qwen 2.5 7B7B
fitsQ8_0
fitsFP16/BF16
Qwen 2.5 72B72.7B
overQ4_K_M
overQ4_K_M

Which one wins for…

$ ./recommend --by-workload
More VRAM headroom

RTX 4090 has 8 GB more.

Faster decode (bandwidth)

RTX 4090 by +41%.

Faster prefill (compute)

RTX 4080 by +18% TFLOPS.

Catalog models that fit

RTX 4090: 21 fit · RTX 4080: 16.

Drill into either card.

$ ./vrambudget --gpu
$ openRTX 408016GBada$ openRTX 409024GBada · flagship

Discussion.

$ gh discussion list

// sign in with github to leave a comment. threads live in the repo's discussions tab.