Datasheet for local AI49 models98 GPUsData read 2026-09-25
No adsNo tracking
Check my GPU →

Radeon 8060S (Strix Halo) 96 GB vs RTX 5090 32 GB for local AI

The Radeon 8060S (Strix Halo) 96 GB has 64 GB more memory, and that decides it for local AI: it runs a better file (or runs at all) for 22 of 49 models, and 49 run at 8-bit or better against 47. The RTX 5090 32 GB does have hardware FP8, so for models that fit on both it can be faster per step.

Radeon 8060S (Strix Halo) 96 GB
VRAM
96 GB
Memory
LPDDR5X (shared system memory)
Bandwidth
256 GB/s
Architecture
RDNA 3.5
FP8 compute
No
Power
—

All models on it →

VS
RTX 5090 32 GB
VRAM
32 GB
Memory
GDDR7
Bandwidth
1792 GB/s
Architecture
Blackwell
FP8 compute
Yes
Launch price
$1,999

All models on it →

01Model by model

ModelRadeon 8060S (Strix Halo) 96 GBRTX 5090 32 GB
image models
FLUX.1 devRuns well 16-bitRuns well 16-bit
FLUX.1 schnellRuns well 16-bitRuns well 16-bit
FLUX.1 KontextRuns well 16-bitRuns well 16-bit
FLUX.1 KreaRuns well 16-bitRuns well 16-bit
FLUX.1 FillRuns well 16-bitRuns well 16-bit
FLUX.2 devRuns well 16-bitRuns Q6_K
FLUX.2 klein 9BRuns well 16-bitRuns well 16-bit
FLUX.2 klein 4BRuns well 16-bitRuns well 16-bit
Krea 2Runs well 16-bitRuns well 16-bit
Qwen-ImageRuns well 16-bitRuns well FP8
Qwen-Image-EditRuns well 16-bitRuns well FP8
Qwen-Image 2.1Runs well 16-bitRuns well 16-bit
Z-Image TurboRuns well 16-bitRuns well 16-bit
Z-ImageRuns well 16-bitRuns well 16-bit
Ideogram 4Runs well Q8_0Runs well FP8
Boogu-ImageRuns well 16-bitRuns well 16-bit
ERNIE-ImageRuns well 16-bitRuns well 16-bit
HiDream-O1Runs well 16-bitRuns well 16-bit
Mage-FlowRuns well 16-bitRuns well 16-bit
Lumina 2.0Runs well 16-bitRuns well 16-bit
HiDream-I1 FullRuns well 16-bitRuns well FP8
HiDream-I1Runs well 16-bitRuns well FP8
SD 3.5 LargeRuns well 16-bitRuns well 16-bit
SD 3.5 MediumRuns well 16-bitRuns well 16-bit
Chroma1-HDRuns well 16-bitRuns well 16-bit
SDXLRuns well 16-bitRuns well 16-bit
Illustrious / PonyRuns well 16-bitRuns well 16-bit
SD 1.5Runs well 16-bitRuns well 16-bit
HunyuanImage 2.1Runs well 16-bitRuns well FP8
video models
Wan 2.1 14BRuns well 16-bitRuns well FP8
Wan 2.1 1.3BRuns well 16-bitRuns well 16-bit
Wan 2.1 I2V 480PRuns well 16-bitRuns well FP8
Wan 2.1 I2V 720PRuns well 16-bitRuns well FP8
Wan VACE 14BRuns well 16-bitRuns well Q8_0
Wan 2.2 T2VRuns well 16-bitRuns well FP8
Wan 2.2 I2VRuns well 16-bitRuns well FP8
Wan 2.2 5BRuns well 16-bitRuns well 16-bit
Wan 2.2 AnimateRuns well 16-bitRuns well FP8
Wan Animate 2Runs well 16-bitRuns well Q8_0
Wan 2.2 S2VRuns well 16-bitRuns well FP8
SCAIL-2Runs well 16-bitRuns well FP8
HunyuanVideo 13BRuns well 16-bitRuns well 16-bit
LTX-Video 13BRuns well 16-bitRuns well FP8
HunyuanVideo 1.5Runs well 16-bitRuns well 16-bit
LTX-2Runs well 16-bitRuns well Q8_0
LTX-2.3Runs well 16-bitRuns well Q8_0
LTX-2.5Runs well 16-bitRuns well Q8_0
MiniMax H3Runs well 16-bitRuns Q5_K_M
MiniMax H3 PrunedRuns well 16-bitRuns well FP8

Highlighted: the GPU that runs a better file for that model. Launch prices are the maker’s original list prices, not today’s street prices. Calculated from real file sizes; how the numbers work.

02Score

49run well on Radeon 8060S (Strix Halo) 96 GB
47run well on RTX 5090 32 GB
22better on Radeon 8060S (Strix Halo) 96 GB
0better on RTX 5090 32 GB