Can InternLM 20B run on RTX 5090 32GB?

BARELY — Tight on Memory

C49Usable

Estimated from fit model

InternLM 20B needs ~36.8 GB VRAM. RTX 5090 32GB has 32.0 GB. With Q5_K_M quantization, expect ~45 tok/s.

Runtime: llama.cppCapacity: OffloadBandwidth: HighStack: StandardBottleneck: Host offload

Operating mode

Choose the run profile you care about

Interactive favors responsiveness, while light API and scale-out lean harder on serving readiness. The fit stays the same, but the recommendation lens changes.

Current mode

Balanced

Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.

Capabilities:

Select quantization to explore

Q5_K_M (High quality) — 36.8 GB, 44.5 tok/s, Very compromised (needs ~1.9 GB host RAM)

36.8 GB required32.0 GB available

115% VRAM needed

4.8 GB over capacity — needs offload or smaller quantization

Fit status

Very compromised (needs ~1.9 GB host RAM)

Decode

44.5 tok/s

TTFT

4346 ms

Safe context

Memory

36.8 GB / 32.0 GB

Offload

10%

Memory breakdown

Weights14.4 GB

KV Cache18.3 GB

Runtime0.9 GB

Headroom3.2 GB

See how fast it feels

See how fast it feelsInternLM 20B on RTX 5090 32GB
1st promptCold start — includes initialization
>Write a Python function to sort a list
Here's a Python function that sorts a list using quicksort:

def quicksort(arr):
    if len(arr) <= 1:
        return arr
    pivot = arr[len(arr) // 2]
    left   = [x for x in arr if x < pivot]
    middle = [x for x in arr if x == pivot]
    right  = [x for x in arr if x > pivot]
    return quicksort(left) + middle + quicksort(right)

# Example usage
numbers = [3, 6, 8, 10, 1, 2, 1]
print(quicksort(numbers))  # [1, 1, 2, 3, 6, 8, 10]

The algorithm picks a pivot, partitions the list into smaller, equal, and larger elements, then recursively sorts each partition and concatenates the results.
2nd promptWarm — model ready, no init overhead
>Now add type hints and a docstring
def quicksort(arr: list[int]) -> list[int]:
    """Sort a list using the quicksort algorithm.

    Args:
        arr: List of integers to sort.

    Returns:
        A new sorted list.
    """
    if len(arr) <= 1:
        return arr
    pivot = arr[len(arr) // 2]
    left   = [x for x in arr if x < pivot]
    middle = [x for x in arr if x == pivot]
    right  = [x for x in arr if x > pivot]
    return quicksort(left) + middle + quicksort(right)

Added type hints for the input and return type, plus a Google-style docstring.
Estimated: 44.5 tok/s decode · 4.3s TTFT (warm) · 111 tok/s prefill

What limits this setup

It fits through host-memory offload, and offload is the main reason performance drops.

CPU or host-memory offload is active

About 10% of the working set spills out of accelerator memory, which usually hurts latency and sustained decode throughput.

Very little memory headroom

You can run the model, but there is not much room left for longer context, bigger batches, extra apps, or future model updates.

Best improvement path

Remove offload with more accelerator memory

Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.

Buy headroom, not only minimum fit

A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.

Increase host RAM if you keep offloading

This setup may need roughly 1.9 GB of extra host RAM just for the offloaded portion, before OS and other tools.

Performance by workload

Workload	Grade	Fit	Decode	TTFT	Context
Chat	B	Tight fit	77.5 tok/s	1362 ms	8K
Coding	C	Very compromised (needs ~1.9 GB host RAM)	44.5 tok/s	4346 ms	8K
Agentic Coding	F	Too heavy	19.5 tok/s	14405 ms	8K
Reasoning	C	Very compromised (needs ~1.9 GB host RAM)	44.5 tok/s	5136 ms	8K
RAG	F	Too heavy	19.5 tok/s	18006 ms	8K

Inference speed

InternLM 20B inference speed — tokens per second by GPU & Mac

Estimated decode speed (tokens/sec) for InternLM 20B at Q5_K_M across popular GPUs and Apple Silicon, using the fastest local runtime per device. Fastest is RTX 5090 32GB at ~45 tok/s. Speed is memory-bandwidth bound, so cards that fit the whole model in VRAM run far faster than ones that offload to system RAM.

GPU / Mac	Memory	Quant	Speed (tok/s)	Fits?
RTX 5090 32GB	32 GB	Q5_K_M	44.5	Heavy offload
Mac Studio M3 Ultra 256GB	256 GB	Q5_K_M	39.4	Fits
Mac Studio M2 Ultra 128GB	128 GB	Q5_K_M	32.9	Fits
Mac Studio M1 Ultra 128GB	128 GB	Q5_K_M	31.2	Fits
MacBook Pro M4 Max 128GB	128 GB	Q5_K_M	30.7	Fits
MacBook Pro M4 Max 64GB	64 GB	Q5_K_M	30.7	Tight
MacBook Pro M3 Max 64GB	64 GB	Q5_K_M	17.0	Tight
MacBook Pro M4 Pro 48GB	48 GB	Q5_K_M	16.1	Heavy offload
RX 7900 XTX 24GB	24 GB	Q5_K_M	15.6	Too big
MacBook Pro M1 Max 64GB	64 GB	Q5_K_M	15.6	Tight
RTX 4090 24GB	24 GB	Q5_K_M	14.0	Too big
RTX 3090 24GB	24 GB	Q5_K_M	12.9	Too big
RTX 4080 Super 16GB	16 GB	Q5_K_M	5.3	Too big
RTX 4070 12GB	12 GB	Q5_K_M	3.3	Too big
RTX 3060 12GB	12 GB	Q5_K_M	2.2	Too big
RTX 4060 8GB	8 GB	Q5_K_M	2.0	Too big

Estimates for single-stream decoding at Q5_K_M; real tokens/sec varies with prompt length, context, batch size, and runtime build. Prompt processing (prefill) is faster than the decode figures shown here.

Quantization options

How InternLM 20B (20B params) fits at each quantization level on RTX 5090 32GB (32.0 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	7.8 GB	Low	C53
Q3_K_S	3	9.8 GB	Low	C54
NVFP4	4	11.2 GB	Medium	C54
Q4_K_M	4	12.2 GB	Medium	C55
Q5_K_M	5	14.4 GB	High	B56
Q6_K	6	16.4 GB	High	B57
Q8_0Best for your GPU	8	21.4 GB	Very High	B57
F16	16	41.0 GB	Maximum	F0

Get started

Copy-paste commands to run InternLM 20B on your machine.

Run

docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
  --hf-repo "internlm/internlm2_5-20b-chat" \
  --hf-file "internlm2_5-20b-chat-Q5_K_M.gguf" \
  -c 4096 -ngl 99

Upgrade-Optionen

Hardware, die InternLM 20B gut ausführt

RTX A6000 48GBGünstige Wahl

48 GB VRAM (+16)

Removes host-memory offload, which is usually the single biggest latency and throughput win.41.3 tok/s Dekodierung

Removes host-memory offload, which is usually the single biggest latency and throughput win.

Adds memory headroom for longer context windows and future model growth.

ca. $4,650 MSRP

RTX PRO 5000 Blackwell 48GBBestes Preis-Leistungs-Verhältnis

48 GB VRAM (+16)

Removes host-memory offload, which is usually the single biggest latency and throughput win.80 tok/s Dekodierung

Removes host-memory offload, which is usually the single biggest latency and throughput win.

Raises estimated decode speed by about 80%.

ca. $4,999 MSRP

Quadro RTX 8000 48GBNVIDIA-Upgrade

48 GB VRAM (+16)

Removes host-memory offload, which is usually the single biggest latency and throughput win.32.8 tok/s Dekodierung

Removes host-memory offload, which is usually the single biggest latency and throughput win.

Adds memory headroom for longer context windows and future model growth.

ca. $5,800 MSRP

Frequently asked questions

Can RTX 5090 32GB run InternLM 20B?

Yes, RTX 5090 32GB can run InternLM 20B with a C grade (Very compromised (needs ~1.9 GB host RAM)). Expected decode speed: 44.5 tok/s.

How much VRAM does InternLM 20B need?

InternLM 20B (20B parameters) requires approximately 36.8 GB of memory with Q5_K_M quantization.

What is the best quantization for InternLM 20B?

The recommended quantization for InternLM 20B is Q5_K_M, which balances quality and memory efficiency.

What speed will InternLM 20B run at on RTX 5090 32GB?

On RTX 5090 32GB, InternLM 20B achieves approximately 44.5 tokens per second decode speed with a time-to-first-token of 4346ms using Q5_K_M quantization.

Can RTX 5090 32GB run InternLM 20B for coding?

For coding workloads, InternLM 20B on RTX 5090 32GB receives a C grade with 44.5 tok/s and 8K context.

What context window can InternLM 20B use on RTX 5090 32GB?

On RTX 5090 32GB, InternLM 20B can safely use up to 8K tokens of context. The model's official context limit is 8K, but available memory constrains the safe maximum.

What should I upgrade first if InternLM 20B feels slow on RTX 5090 32GB?

Remove offload with more accelerator memory. Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.

See all results for RTX 5090 32GB See all hardware for InternLM 20B

Embed this result▼

Paste this snippet into any page to show a live fit card.

<iframe src="https://willitrunai.com/embed/internlm-20b-on-rtx-5090-32gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview: