AMD / discrete

AMD Instinct MI455X

432 GB decides what fits. 23300 GB/s decides how fast it runs once it does.

Quick answer

The AMD Instinct MI455X fits 297 of 320 sized models entirely in its 432 GB; the largest widely used one is DeepSeek-V3.2 (422.2 GB at Q4_K_M). Memory decides what fits; its 23300 GB/s of bandwidth decides how fast it answers.

Largest popular model that fitsDeepSeek-V3.2685B params · Q4_K_M · needs 422.2 GB~842 tok/s Faster than you readOpen in the calculator →
Best fast pickDeepSeek-V3.2-Exp685B params · Q4_K_M · needs 422.3 GB~842 tok/s Faster than you readOpen in the calculator →
Most downloaded that fitsGLM-5.3-Flash321B params · Q4_K_M · needs 198.3 GB~1921 tok/s Faster than you readOpen in the calculator →

At 8,192 tokens of context with the whole model in device memory. Speeds are estimates from memory bandwidth, not benchmarks run on this card.

AMD Instinct MI455X · 432 GB · 23300 GB/s · Q4_K_M where published · 8,192 tokensSpecificationpublished dataDecode speedestimate

297 of 320 fit entirely

297 of the 320 models the engine can size fit entirely in device memory at Q4_K_M where it is published, otherwise the nearest published format, and 8,192 tokens. 6 run with some layers on system memory (32 GB assumed), and 17 do not run at all.

93%of the models the engine can size fit entirely

432GB of device memory

23300GB/s memory bandwidth

6run with system memory

17do not run at 8,192 tokens

Featured models · Q4_K_M at 8,192 tokens, including offload

Model	Needs	Verdict	Decode	Calculator
Qwen3.5-2B2.3B parameters	2.32 GB	fits	~14717 tok/s	Open →
NVIDIA-Nemotron-3-Nano-4B-BF164.0B parameters	3.46 GB	fits	~6819 tok/s	Open →
Qwen3-VL-8B-Instruct8.8B parameters	7.04 GB	fits	~3084 tok/s	Open →
Qwen3.5-9B9.7B parameters	7.05 GB	fits	~3735 tok/s	Open →
gemma-4-12B-it12B parameters	9.54 GB	fits	~2221 tok/s	Open →
gpt-oss-20b21B parameters	13.8 GB	fits	~7258 tok/s	Open →
Qwen3.8-27B28B parameters	18.6 GB	fits	~1226 tok/s	Open →
Qwen3.6-27B28B parameters	18.6 GB	fits	~1226 tok/s	Open →
Kimi-Linear-48B-A3B-Instruct49B parameters	30.4 GB	fits	~9710 tok/s	Open →
Qwen3.8-Flash-Next180B parameters	111.6 GB	fits	~5046 tok/s	Open →
DeepSeek-V4-Flash-Vision-Exp305B parameters	188.1 GB	fits	~1612 tok/s	Open →
GLM-5.3-Flash321B parameters	198.3 GB	fits	~1921 tok/s	Open →

These examples are selected from prominent labs using the catalogue’s latest Hugging Face 30-day downloads and repository-creation freshness signal, with newer releases guaranteed a place. Offloaded rows assume 32 GB of system RAM, and a speed is only shown for a row that runs. Fitting in memory is not the same as loading: whether the runtime and version you have supports each architecture and format on this machine has not been tested here. Each “Open” link carries the same model, format, context and RAM into the calculator.

Fits entirely in AMD Instinct MI455X memory — 297 of 320 sized models

Every model that fitswhole model resident · offload off

The most demanding model that fits is DeepSeek-V3.2-Exp at Q4_K_M: 422.3 GB of the 432 GB, leaving 9.75 GB spare.

Fully resident at 8,192 tokens (or the model’s own maximum, where that is shorter), offload off, at 23300 GB/s. Each row is a run of the engine for this configuration; the rows start with current, prominent releases and “fits” is memory, not a tested runtime. Older or less prominent models remain available through this search and “Show all”.

Find a model in this list

Showing 40 of 297 models that fit.

Model	Format	Needs	Spare	Decode	Downloads / 30d	Calculator
GLM-5.3-FlashNEWzai-org · 321B params	Q4_K_M	198.3 GB	233.7 GB	~1921 tok/s	5.4M	Open →
Qwen3.8-27BQwen · 28B params	Q4_K_M	18.6 GB	413.4 GB	~1226 tok/s	6.9M	Open →
DeepSeek-V4-Flash-0731DeepSeek · 304B params	Q4_K_M	187.5 GB	244.5 GB	~1612 tok/s	4.5M	Open →
Qwen3.8-Flash-NextNEWQwen · 180B params	Q4_K_M	111.6 GB	320.4 GB	~5046 tok/s	1.4M	Open →
gemma-4-26B-A4B-itGoogle · 26B params	Q4_K_M	16.8 GB	415.2 GB	~6818 tok/s	13M	Open →
DeepSeek-V4-Flash-Vision-ExpNEWDeepSeek · 305B params	Q4_K_M	188.1 GB	243.9 GB	~1612 tok/s	915.3K	Open →
gemma-4-31B-itGoogle · 31B params	Q4_K_M	22.2 GB	409.8 GB	~882 tok/s	9.9M	Open →
Qwen3.5-9BQwen · 9.7B params	Q4_K_M	7.05 GB	425.0 GB	~3735 tok/s	9M	Open →
Qwen3.5-4BQwen · 4.7B params	Q4_K_M	3.98 GB	428.0 GB	~6721 tok/s	7.8M	Open →
Qwen3.6-35B-A3BQwen · 36B params	Q4_K_M	23.1 GB	408.9 GB	~9791 tok/s	3.3M	Open →
gemma-4-12B-itGoogle · 12B params	Q4_K_M	9.54 GB	422.5 GB	~2221 tok/s	1.9M	Open →
Inkling-Smallthinkingmachines · 266B params	Q4_K_M	165.5 GB	266.5 GB	~2549 tok/s	657.4K	Open →
Qwen3.6-27BQwen · 28B params	Q4_K_M	18.6 GB	413.4 GB	~1226 tok/s	2.5M	Open →
Qwen3.5-2BQwen · 2.3B params	Q4_K_M	2.32 GB	429.7 GB	~14717 tok/s	4.9M	Open →
gemma-4-E4B-itGoogle · 8.0B params	Q4_K_M	5.86 GB	426.1 GB	~6001 tok/s	4.4M	Open →
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16NVIDIA · 32B params	Q4_K_M	20.3 GB	411.7 GB	~931 tok/s	530K	Open →
NVIDIA-Nemotron-3-Nano-4B-BF16NVIDIA · 4.0B params	Q4_K_M	3.46 GB	428.5 GB	~6819 tok/s	3.4M	Open →
gemma-4-E2B-itGoogle · 5.1B params	Q4_K_M	4.02 GB	428.0 GB	~15853 tok/s	3M	Open →
Qwen3.5-0.8BQwen · 873M params	Q4_K_M	1.46 GB	430.5 GB	~32240 tok/s	2.6M	Open →
DeepSeek-V4-FlashDeepSeek · 284B params	Q4_K_M	175.9 GB	256.1 GB	~1612 tok/s	1.1M	Open →
Qwen3-VL-8B-InstructQwen · 8.8B params	Q4_K_M	7.04 GB	425.0 GB	~3084 tok/s	14.6M	Open →
MiniMax-M2.7MiniMaxAI · 229B params	Q4_K_M	141.2 GB	290.8 GB	~2113 tok/s	1.1M	Open →
Qwen3.5-27BQwen · 28B params	Q4_K_M	18.6 GB	413.4 GB	~1226 tok/s	1.9M	Open →
North-Micro-Vision-InstructCohereLabs · 2.5B params	Q4_K_M	2.99 GB	429.0 GB	~9678 tok/s	180.7K	Open →
Hy3Tencent · 299B params	Q4_K_M	181.2 GB	250.8 GB	~1286 tok/s	304.9K	Open →
Qwen3.5-35B-A3BQwen · 36B params	Q4_K_M	23.1 GB	408.9 GB	~9791 tok/s	1.6M	Open →
NVIDIA-Nemotron-3-Super-120B-A12B-BF16NVIDIA · 124B params	Q4_K_M	76.9 GB	355.1 GB	~238 tok/s	1.2M	Open →
NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16NVIDIA · 561B params	Q4_K_M	345.4 GB	86.6 GB	~53 tok/s	358K	Open →
granite-4.2-8bIBM · 8.8B params	Q4_K_M	7.49 GB	424.5 GB	~2809 tok/s	128.4K	Open →
DeepSeek-V3.2DeepSeek · 685B params	Q4_K_M	422.2 GB	9.77 GB	~842 tok/s	3.6M	Open →
GLM-4.7-Flashzai-org · 31B params	Q4_K_M	20.4 GB	411.6 GB	~925 tok/s	1.8M	Open →
granite-4.1-3bIBM · 3.4B params	Q4_K_M	3.72 GB	428.3 GB	~6554 tok/s	521.8K	Open →
LFM2.5-2.6BLiquidAI · 2.7B params	Q4_K_M	2.61 GB	429.4 GB	~10511 tok/s	108.1K	Open →
MiMo-V2.6-Flash-RLNEWXiaomiMiMo · 309B params	Q4_K_M	192.2 GB	239.8 GB	~1872 tok/s	48.6K	Open →
MiniMax-M3MiniMaxAI · 427B params	Q4_K_M	264.0 GB	168.0 GB	~1125 tok/s	177.2K	Open →
MiMo-V2.5XiaomiMiMo · 311B params	Q4_K_M	192.2 GB	239.8 GB	~1872 tok/s	245.2K	Open →
Qwen3-VL-4B-InstructQwen · 4.4B params	Q4_K_M	4.72 GB	427.3 GB	~4902 tok/s	3.5M	Open →
granite-4.1-30bIBM · 29B params	Q4_K_M	20.4 GB	411.6 GB	~924 tok/s	301.5K	Open →
Qwen3.5-122B-A10BQwen · 125B params	Q4_K_M	77.9 GB	354.1 GB	~3584 tok/s	512.6K	Open →
Qwen3-VL-2B-InstructQwen · 2.1B params	Q4_K_M	3.05 GB	429.0 GB	~8885 tok/s	2.8M	Open →

Sized at the format most people actually download, not at FP16. The catalogue ordered by downloads →

Under the hood

The specification behind every figure

What the manufacturer publishes for this device, and the pages it was read from.

Manufacturer specification

Memory scope	dedicated
Capacity	432 GB
Published options	432 GB
Bandwidth	23300 GB/s
Memory type	HBM4
Bus width	Not published
FP32 peak	315 TFLOPS
Dense matrix peak	5000 TFLOPS (FP16 Matrix, dense)
Power	Not published

Source ledger

AMD Instinct MI455X GPUs ↗

Retrieved 2026-09-13

Caveats

Per-module figures for an Enhanced Accelerator Module (EAM) that AMD sells for its Helios rack; the rack's aggregate memory is not used as a device. AMD lists a launch date of 7/23/2026; that is not a statement of retail availability.
AMD's page publishes no board power for this module, so none is recorded and no energy figure is computed.
AMD publishes no memory interface width for this part, so none is recorded.
The FP16 matrix figure is the dense one; AMD's separate "with Structured Sparsity" row (10.1 PFLOPs) is not used.

Published capacity is a hardware ceiling, not guaranteed free runtime memory. The calculator shows the runtime reserve separately rather than folding it into a single number.

Run the diagnostic on the AMD Instinct MI455X →