internlm2 5 20b chat needs ~57.6 GB VRAM. NVIDIA DGX Spark 128GB has 0 MB. With F16 quantization, expect ~6 tok/s.
Operating mode
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.
Select quantization to explore
Fit status
Runs well
Decode
13.4 tok/s
TTFT
14420 ms
Safe context
562K
Memory
28.8 GB / 108.8 GB
The model fits in shared memory, but shared-memory bandwidth is now the real limiter.
Fit does not mean dedicated-VRAM speed
Unified or shared memory can make a model technically fit, but sustained tokens per second may still trail a discrete high-bandwidth GPU with less total memory.
Shared-memory contention still exists
The OS, browser, and inference runtime all compete for the same physical memory pool, so real-world headroom is less forgiving than raw capacity suggests.
Prioritize bandwidth, not only capacity
If this workload feels slow, the next useful step is often a GPU tier with materially faster memory bandwidth rather than only a small bump in capacity.
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | C | Runs well | 13.4 tok/s | 7865 ms | 562K |
| Coding | F | Too heavy | 2.4 tok/s | 80109 ms | 4K |
| Agentic Coding | C | Runs well | 13.4 tok/s | 20974 ms | 562K |
| Reasoning | C | Runs well | 13.4 tok/s | 17041 ms | 562K |
| RAG | C | Runs well | 13.4 tok/s | 26217 ms | 562K |
Inference speed
Estimated decode speed (tokens/sec) for internlm2 5 20b chat at Q4_K_M across popular GPUs and Apple Silicon, using the fastest local runtime per device. Fastest is RTX 5090 32GB at ~98 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 | Q4_K_M | 98.4 | Fits |
| RTX 4090 24GB | 24 GB | Q4_K_M | 62.8 | Fits |
RX 7900 XTX 24GB | 24 GB | Q4_K_M | 56.7 | Fits |
| RTX 3090 24GB | 24 GB | Q4_K_M | 53.7 | Fits |
Mac Studio M3 Ultra 256GB | 256 GB | Q4_K_M | 45.6 | Fits |
Mac Studio M2 Ultra 128GB | 128 GB | Q4_K_M | 38.0 | Fits |
Mac Studio M1 Ultra 128GB | 128 GB | Q4_K_M | 36.1 | Fits |
MacBook Pro M4 Max 128GB | 128 GB | Q4_K_M | 35.6 | Fits |
MacBook Pro M4 Max 64GB | 64 GB | Q4_K_M | 35.6 | Fits |
| RTX 4080 Super 16GB | 16 GB | Q4_K_M | 31.7 | Heavy offload |
MacBook Pro M4 Pro 48GB | 48 GB | Q4_K_M | 22.4 | Fits |
MacBook Pro M3 Max 64GB | 64 GB | Q4_K_M | 19.7 | Fits |
MacBook Pro M1 Max 64GB | 64 GB | Q4_K_M | 18.0 | Fits |
| RTX 4070 12GB | 12 GB | Q4_K_M | 11.2 | Too big |
| RTX 3060 12GB | 12 GB | Q4_K_M | 7.1 | Too big |
| RTX 4060 8GB | 8 GB | Q4_K_M | 2.6 | Too big |
Estimates for single-stream decoding at Q4_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.
How internlm2 5 20b chat (20B params) fits at each quantization level on NVIDIA DGX Spark 128GB (92.2 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 7.8 GB | Low | D39 |
Q3_K_S | 3 | 9.8 GB | Low | D39 |
NVFP4 | 4 | 11.2 GB | Medium | D39 |
Q4_K_M | 4 | 12.2 GB | Medium | D39 |
Q5_K_M | 5 | 14.4 GB | High | D40 |
Q6_K | 6 | 16.4 GB | High | D40 |
Q8_0 | 8 | 21.4 GB | Very High | C41 |
F16Best for your GPU | 16 | 41.0 GB | Maximum | C45 |
Copy-paste commands to run internlm2 5 20b chat on your machine.
Run
lms load hf-bartowski--internlm2-5-20b-chat-gguf && lms server startUpgrade-Optionen
NVIDIA H200 141GBGünstige WahlRaises estimated decode speed by about 1990%.
Adds memory headroom for longer context windows and future model growth.
ca. $30,000 MSRP
NVIDIA H200 PCIe 141GBBestes Preis-Leistungs-VerhältnisRaises estimated decode speed by about 1990%.
Adds memory headroom for longer context windows and future model growth.
ca. $30,000 MSRP
NVIDIA B200 180GBNVIDIA-UpgradeRaises estimated decode speed by about 1990%.
Adds memory headroom for longer context windows and future model growth.
ca. $30,000 MSRP
Yes, NVIDIA DGX Spark 128GB can run internlm2 5 20b chat at F16 quantization (Runs well). The recommended Q4_K_M requires 15.7 GB which exceeds available memory, but at F16 it needs only 57.6 GB. Expected decode speed: 5.6 tok/s.
internlm2 5 20b chat (20B parameters) requires approximately 15.7 GB at Q4_K_M quantization. On NVIDIA DGX Spark 128GB, it fits at F16 using 57.6 GB.
The recommended quantization is Q4_K_M, but on NVIDIA DGX Spark 128GB the best fitting quantization is F16, which uses 57.6 GB.
On NVIDIA DGX Spark 128GB, internlm2 5 20b chat achieves approximately 5.6 tokens per second decode speed with a time-to-first-token of 34614ms using F16 quantization.
For coding workloads, internlm2 5 20b chat on NVIDIA DGX Spark 128GB receives a F grade with 2.4 tok/s and 4K context.
On NVIDIA DGX Spark 128GB, internlm2 5 20b chat can safely use up to 366K tokens of context at F16 quantization. The model's official context limit is —, but available memory constrains the safe maximum.
Prioritize bandwidth, not only capacity. If this workload feels slow, the next useful step is often a GPU tier with materially faster memory bandwidth rather than only a small bump in capacity.
Not always. NVIDIA DGX Spark 128GB can often fit larger models thanks to unified memory, but a discrete GPU with dedicated high-bandwidth VRAM may still decode faster once the model fits. For this combination, the important distinction is capacity versus sustained throughput.
Paste this snippet into any page to show a live fit card.
<iframe src="https://willitrunai.com/embed/hf-bartowski--internlm2-5-20b-chat-gguf-on-dgx-spark-128gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>
Preview: