# Realizing Native INT8 Compute for Diffusion Transformers on Consumer GPUs: A Fused INT8 GEMM Kernel for Ideogram 4.0

arXiv: https://arxiv.org/abs/2606.14598 · HTML: https://mlsanity.com/paper/2606.14598/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models.
- Builds on 21 of its 24 references that are in the map.
- First posted on arXiv 12 Jun 2026.

## Placed in

- Diffusion models / [Diffusion foundations and sampling](https://mlsanity.com/diffusion/diffusion-foundations-and-sampling/)

## Builds on (most cited first, 21 of 21)

- [SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models](https://mlsanity.com/paper/2411.05007/) (2024)
- [SageAttention: Accurate 8-Bit Attention for Plug-and-play Inference Acceleration](https://mlsanity.com/paper/2410.02367/) (2024)
- [Efficient Diffusion Models: A Survey](https://mlsanity.com/paper/2502.06805/) (2025)
- [Prefixing Attention Sinks can Mitigate Activation Outliers for Large Language Model Quantization](https://mlsanity.com/paper/2406.12016/) (2024)
- [1.58-bit FLUX](https://mlsanity.com/paper/2412.18653/) (2024)
- [ABQ-LLM: Arbitrary-Bit Quantized Inference Acceleration for Large Language Models](https://mlsanity.com/paper/2408.08554/) (2024)
- [AutoKernel: Autonomous GPU Kernel Optimization via Iterative Agent-Driven Search](https://mlsanity.com/paper/2603.21331/) (2026)
- [QQQ: Quality Quattuor-Bit Quantization for Large Language Models](https://mlsanity.com/paper/2406.09904/) (2024)
- [Outliers and Calibration Sets have Diminishing Effect on Quantization of Modern LLMs](https://mlsanity.com/paper/2405.20835/) (2024)
- [Exploring Quantization for Efficient Pre-Training of Transformer Language Models](https://mlsanity.com/paper/2407.11722/) (2024)
- [MixDiT: Accelerating Image Diffusion Transformer Inference with Mixed-Precision MX Quantization](https://mlsanity.com/paper/2504.08398/) (2025)
- [LiquidGEMM: Hardware-Efficient W4A8 GEMM Kernel for High-Performance LLM Serving](https://mlsanity.com/paper/2509.01229/) (2025)
- [PQD: Post-training Quantization for Efficient Diffusion Models](https://mlsanity.com/paper/2501.00124/) (2024)
- [DMQ: Dissecting Outliers of Diffusion Models for Post-Training Quantization](https://mlsanity.com/paper/2507.12933/) (2025)
- [Mixed-Precision Quantization for Language Models: Techniques and Prospects](https://mlsanity.com/paper/2510.16805/) (2025)
- [DiRotQ: Rotation-Aware Quantization for 4-bit Diffusion Transformers](https://mlsanity.com/paper/2605.16732/) (2026)
- [SQ-DM: Accelerating Diffusion Models with Aggressive Quantization and Temporal Sparsity](https://mlsanity.com/paper/2501.15448/) (2025)
- [Tilus: A Tile-Level GPGPU Programming Language for Low-Precision Computation](https://mlsanity.com/paper/2504.12984/) (2025)
- [FireQ: Fast INT4-FP8 Kernel and RoPE-aware Quantization for LLM Inference Acceleration](https://mlsanity.com/paper/2505.20839/) (2025)
- [Multi-Scale Dequant: Eliminating Dequantization Bottleneck via Activation Decomposition for Efficient LLM Inference](https://mlsanity.com/paper/2605.13915/) (2026)
- [Holding the FP8 Quality Ceiling at 8-Bit Weights and Activations: INT8 and GGUF Post-Training Quantization of Ideogram 4.0 for Consumer GPUs](https://mlsanity.com/paper/2606.12280/) (2026)

## Abstract (arXiv)

Post-training INT8 (W8A8) quantization of diffusion transformers is widely deployed as a speed optimization, yet on consumer Ampere GPUs it is frequently slower than the FP8 and NF4 alternatives it is meant to beat. We trace this to a software artifact: the production "INT8" forward quantizes weights and activations only to immediately dequantize them back to bf16 and run a bf16 matrix multiply, never engaging the GPU's INT8 tensor cores, so the hardware's compute advantage is left entirely unrealized. We close this gap with a single fused Triton INT8 GEMM (int8xint8->int32 on Ampere tensor cores, with per-token x per-channel dequantization and bias folded into the epilogue, autotuned per GEMM shape) dropped into the Ideogram 4.0 diffusion transformer's linear layers in place of the dequantize-to-bf16 path. In the kernel, the int8xint8->int32 accumulation is bit-exact against torch._int_mm and the dequantized output matches the reference at cosine similarity 1.0 with no NaNs, running 2.8-4.2x faster than bf16 per GEMM. End to end it delivers a ~1.1x (~9-10%) speedup at 768px, and at 1024px it generates an image in 156.5 s on a single RTX 3090, faster than the single-card NF4 (164.5 s) and FP8 (172.9 s) baselines, at no measurable quality cost on these point estimates (PickScore/CLIPScore). INT8 thus goes from the slowest variant to the fastest, and 1024px becomes single-GPU feasible. The primary speed criterion (beat FP8, by ~9.5%) is comfortably met; the NF4 margin (~4.9%, single-run n=4) is within run-to-run variance we did not quantify and is best read as consistent with meeting the stretch target. We close with an honest deployment map: the win is specific to consumer Ampere, and on A100 and B200 the same kernel loses to those cards' fast native bf16/FP8 paths.