AMD Instinct MI455X
VRAM
432 GB
Memory
HBM4
Bandwidth
19,600 GB/s
FP16
10,000 TF
TDP
-
Memory and compute
Architecture and power
Notes
Launched 2026-07-23 at Advancing AI. 40 PF FP4 / 20 PF FP8 dense per AMD (FP16 derived at half FP8). Helios rack platform; shipments ramp from Q4 2026 into H1 2027.
Models that may fit on AMD Instinct MI455X
Estimates based on INT8 quantization. Actual fit depends on framework and batch size.
More GPUs by AMD
The manufacturer's other cards in the catalog, largest memory first, with bandwidth and power where published.
Frequently asked questions about AMD Instinct MI455X
Answered from the entry's own figures: memory, which models fit, bandwidth, power and the card's class.
AMD Instinct MI455X has 432 GB of HBM4 at 19,600 GB/s. At INT8 a model's weights take about one byte per parameter, so the card holds roughly a 345B-parameter model with room for context; larger models need several cards or a lower precision.
From the catalog's estimates at INT8, DeepSeek V4 Flash Vision Exp, Gemma 4 31B IT, Qwen3 235B A22B Thinking 2507 and Llama Nemotron Embed VL 1B V2 fit on a single AMD Instinct MI455X with their default context. The list on this page shows the estimated memory each takes; the capacity planner sizes any model against this card for your own context length, batch size and request rate.
19,600 GB/s. Bandwidth bounds how fast a model generates tokens, because every token reads the whole set of weights and the KV cache from memory; at this rate a 345B model at INT8 could be read on the order of 57 times a second on one card, before batching and framework efficiency.
AMD has not published a TDP for AMD Instinct MI455X; the card ships in the OAM form factor. The catalog shows the figure as not published rather than estimating it.
AMD Instinct MI455X is a datacenter accelerator, meant for servers and multi-GPU serving, built on the CDNA 5 architecture and released in 2026. It supports a multi-GPU interconnect, so a model larger than one card can be split across several.
Specifications from the manufacturer's published material; a figure that is not published is shown as such, and a derived one is explained in the notes.
Added Jul 30, 2026 · Last updated Jul 30, 2026
Explore more
Companion tools that draw on the same catalog and routing engine.
Put intelligence in the middle.
One gateway in front of every model, with your policies applied and every decision on record. Start with $5 of credit and 5,000 routing decisions a month, no card required.