ERNIE 5.5 Flash Base has 286.3B parameters (14B active per token). In BF16 its weights alone take 521 GiB; quantized to 4-bit, about 157 GiB. Each 32K-token request adds 0.25 GiB of KV cache. Cheapest way to run it today: 8 × GeForce RTX 3090 (24 GB) at $0.864 per hour.
vLLM needs one extra install first: uv pip install vllm==0.32.0 git+https://github.com/baidu/vLLM-ERNIE.git (model card (opens in a new tab))
VRAM by precision and context length
Weights plus the KV cache of one request (BF16 cache) and the model's fixed recurrent state, in GiB. Add 1 to 3 GiB for the inference engine itself.
Weights format
Weights only
+ 8K context
+ 32K context
+ 128K context
+ 1M context
FP8 (native)
262
263
263
263
270
BF16 / FP16
521
521
521
522
529
FP8 / INT8
262
262
262
263
270
4-bit (GGUF Q4_K_M)
157
157
158
158
165
How many GPUs to run ERNIE 5.5 Flash Base
Fewest GPUs that fit the weights plus one 8K-token request. Data-center GPUs use vLLM (90% of memory usable), the others llama.cpp. Several GPUs split the model with tensor or pipeline parallelism.
GPU
FP8 (native)
BF16 / FP16
8-bit
4-bit
GeForce RTX 4090 24 GB
Not on this GPU
32 (several machines)
16 (several machines)
8
GeForce RTX 5090 32 GB
Not on this GPU
24 (several machines)
16 (several machines)
6
RTX PRO 6000 Blackwell 96 GB
Not on this GPU
6
3
2
Mac M5 Max 128 GB 96 GB usable
Not on this GPU
6
3
2
Ryzen AI Max+ 395 128 GB 96 GB usable
Not on this GPU
6
3
2
DGX Spark (GB10) 128 GB
Not on this GPU
5
3
2
NVIDIA H100 80 GB
4
8
4
3
NVIDIA H200 141 GB
3
5
3
2
NVIDIA B200 (HGX) 180 GB
2
4
2
2
8-bit: FP8 with vLLM (INT8 on GPUs without FP8), Q8_0 with llama.cpp. 4-bit: AWQ / GPTQ with vLLM, Q4_K_M with llama.cpp, both counted at the Q4_K_M size. Not on this GPU: the official FP8 weights do not run there (llama.cpp needs a GGUF version, and vLLM needs a GPU with FP8 support).
Cheapest way to run ERNIE 5.5 Flash Base today
For each GPU: the fewest cards that fit ERNIE 5.5 Flash Base with one 8K-token request, the most faithful weight format at that count, and the cheapest on-demand price (2026-11-06).
On-demand prices from AWS, Google Cloud, Azure, RunPod, Vast.ai, Oracle Cloud, OVHcloud, Scaleway, Verda, Akash and UpCloud, checked every hour. Serving many users needs more KV cache, so more memory: size it in the calculator. Some links are affiliate links: StudioTV may earn a commission, at no extra cost to you.
vLLM and SGLang apply these defaults unless a request sets its own. With llama.cpp, pass them yourself: --temp 1 --top-p 0.95
KV cache: how memory grows with context
Architecture: MoE 512 latent experts + 1 shared · 37 compressed-attention layers (128-token window + 128× compression, 512-wide) · 6 sparse MLA layers (512 + 64 latent, indexer keys). Its KV cache is 6.1× smaller than if every layer used full attention, because it keeps only a short window of tokens at full detail and compresses the rest of the sequence. Only 14B parameters are active per token, but all 286B must sit in memory: VRAM depends on the total, speed on the active part.
Context per request
BF16 cache
FP8 cache
8K tokens
0.07 GiB
0.04 GiB
32K tokens
0.25 GiB
0.15 GiB
128K tokens
0.98 GiB
0.60 GiB
1M tokens
7.82 GiB
4.78 GiB
Plus 18.5 MiB of recurrent state per request, whatever the context.
ERNIE 5.5 Flash Base VRAM FAQ
How much VRAM does ERNIE 5.5 Flash Base need?
In BF16 the weights alone take 521 GiB (559 GB). In FP8 that is 262 GiB, and about 157 GiB with 4-bit quantization (Q4_K_M); the official FP8 checkpoint is 262 GiB. Each request then adds KV cache: 0.07 GiB at 8K tokens and 0.25 GiB at 32K (BF16 cache).
Can ERNIE 5.5 Flash Base run on a single RTX 4090 (24 GB)?
No. Even in 4-bit it needs 8 RTX 4090s. In 4-bit it needs 3 H100 GPUs.
Can ERNIE 5.5 Flash Base run on an RTX 5090 (32 GB)?
No. Even in 4-bit it needs 6 RTX 5090s.
Can ERNIE 5.5 Flash Base run on a Mac?
Yes. In 4-bit (Q4_K_M with llama.cpp) it needs a Mac with 256 GB of unified memory or more, of which macOS gives the GPU about 192 GB, for one request with an 8K-token context. In 8-bit (Q8_0) it needs 512 GB.
Can ERNIE 5.5 Flash Base run on a DGX Spark or a Ryzen AI Max+ 395 (128 GB)?
No. Even in 4-bit it needs 2 DGX Spark units or 2 Ryzen AI Max+ 395 machines (128 GB), linked over the network.
What is the cheapest way to run ERNIE 5.5 Flash Base?
On 2026-11-06, the cheapest on-demand setup is 8 × GeForce RTX 3090 (24 GB) with Q4_K_M weights on Vast.ai, at $0.864 per hour (about $630.72 per month). Next: 4 × RTX A6000 (48 GB) with Q4_K_M weights on Vast.ai, at $1.29 per hour. Sized for one 8K-token request; prices are checked every hour.
How many H100 GPUs does ERNIE 5.5 Flash Base need?
With one request and an 8K-token context: 8 in BF16, 4 in FP8 and 3 in 4-bit (vLLM, 90% of memory usable). Serving many users at once needs more memory for their KV caches: the calculator sizes that for you.
What temperature should I use with ERNIE 5.5 Flash Base?
The official defaults are temperature 1, top-p 0.95.
ERNIE 5.5 Flash Base VRAM badge
For a model card or a README: the 4-bit size and the smallest GPU it fits on, linked to this page. Data as JSON: /api/models/ernie-5-5-flash-base.json.
[](https://studiotvai.com/vram-requirements/ernie-5-5-flash-base)
Estimates, not guarantees: computed from the official config.json with the same engine as the LLM VRAM Calculator. Real usage depends on your engine version and settings.
Baidu, China · Released October 21, 2026