Hardware · AMD
What AI models can the AMD EPYC 7002 / 7003 series NAS (512 GB) run?
- RAM
- 512 GB
- Usable for models
- 471 GB
- Memory bandwidth
- 205 GB/s
- Runs models on
- CPU cores
- Released
- 2019
A NAS with this chip runs models on its processor cores, so memory speed sets the pace: small models (about 1-4B) answer at a usable speed, bigger ones are slow. We plan around 471 GB for models, what is left after the NAS system and its storage services. Models run in a container (Docker) or the maker’s app store, and speeds assume the NAS isn’t busy with storage work.
More about this hardware
| Maker | AMD |
|---|---|
| Type | NAS (network attached storage) processor |
| Share of memory for models | 92% (what the NAS system and its storage services typically leave free at this size) |
| Chip | AMD EPYC 7002 / 7003 series |
| AI accelerator | None. Models run on the processor cores. |
| NAS models with this chip | QNAP QAI-h1290FX · QNAP TS-h1290FX · QNAP TS-h2490FU · QNAP TS-h1090FU · Synology FS6420 · Synology SA6400 · Synology RS11626xs+ · 45HomeLab HL15 2.0 (Fully Built & Burned In) · 45HomeLab HL15 Beast (Fully Built, ASRock Rack / EPYC) |
| Memory type | System RAM, shared with the NAS operating system and its storage services |
| Bandwidth figure | 204.8 GB/s, estimated, see the note below |
| Note | One profile for every NAS built on this chip. Memory speed is an estimate from the chip’s supported memory in a typical NAS fitting; a unit with one memory module runs slower than one with two. Speed assumes the NAS is not busy with storage work. |
| Specifications | www.qnap.com |
Apps that run models well on the AMD EPYC 7002 / 7003 series NAS (512 GB)
9 products with the AMD EPYC 7002 / 7003 series NAS (512 GB)
QAI-h1290FXProduct page (opens in a new tab)
TS-h1290FXProduct page (opens in a new tab)
TS-h2490FUProduct page (opens in a new tab)
TS-h1090FUProduct page (opens in a new tab)
FS6420Product page (opens in a new tab)
SA6400Product page (opens in a new tab)Discontinued
RS11626xs+Product page (opens in a new tab)
HL15 2.0 (Fully Built & Burned In)Product page (opens in a new tab)
HL15 Beast (Fully Built, ASRock Rack / EPYC)Product page (opens in a new tab)
Best LLMs for the AMD EPYC 7002 / 7003 series NAS (512 GB)
| Quality | Model | Quant | Memory | Speed | Runs |
|---|---|---|---|---|---|
| 96Excellent quality (A+) · 96 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. | MiniMax M3.1 Flash (428B-A23B) | Q6_K | 391 GB | 4.2 tok/s | Slow |
| 92Excellent quality (A+) · 92 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. | Ling 3.1 Flash 1019 (560B-A25B) | Q4_K_M | 387 GB | 5.5 tok/s | Slow |
| 93Excellent quality (A+) · 93 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. | K2-Horizon (375B-A23B) | Q6_K | 351 GB | 4.2 tok/s | Slow |
| 93Excellent quality (A+) · 93 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. | MiniMax M3 (427B) | Q6_K | 389 GB | 3.8 tok/s | Slow |
| 91Excellent quality (A+) · 91 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. | Beam 1.0 (501B-A23B) | Q5_K_M | 394 GB | 4.8 tok/s | Slow |
| 90Excellent quality (A+) · 90 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. | Ling 3.1 Flash (560B-A25B) | Q4_K_M | 387 GB | 5.5 tok/s | Slow |
| 94Excellent quality (A+) · 94 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. | Step 5 Preview (600B-A27B) | Q5_K_M | 468 GB | 4.2 tok/s | Slowtight fit |
| 90Excellent quality (A+) · 90 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. | GLM-4.7 (355B-A32B) | Q5_K_M | 296 GB | 3.6 tok/s | Slow |
| 91Excellent quality (A+) · 91 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. | Qwen3.8 Flash Next (125B-A6B) | Q4_K_M | 148 GB | 21.6 tok/s | Runs well |
| 88Excellent quality (A+) · 88 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. | Qwen3 Coder 480B-A35B | Q5_K_M | 381 GB | 3.3 tok/s | Slow |
| 82Great quality (A) · 82 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. | DeepSeek V4.1 Flash 1017 (552B-A16B) | Q3_K_M | 412 GB | 10.9 tok/s | Runs welltight fit |
| 89Excellent quality (A+) · 89 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. | DeepSeek V4 Flash (284B-A13B) | Q4_K_M | 224 GB | 10.4 tok/s | Runs well |
Best Image models for the AMD EPYC 7002 / 7003 series NAS (512 GB)
| Quality | Model | Quant | Memory | Speed | Runs |
|---|---|---|---|---|---|
| 90Excellent quality (A+) · 90 of 100: how capable the model is at FP16, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. | HunyuanImage 3.0 (80B-A13B) | FP16 | 170 GB | 41.1 min | Very slow |
| 92Excellent quality (A+) · 92 of 100: how capable the model is at FP16, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. | Cosmos3-Super Text2Image (64B) | FP16 | 133 GB | 27.4 min | Very slow |
| 92Excellent quality (A+) · 92 of 100: how capable the model is at FP16, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. | FLUX.2 [dev] | FP16 | 116 GB | 24.7 min | Very slow |
| 92Excellent quality (A+) · 92 of 100: how capable the model is at FP16, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. | Krea 2 Turbo | FP16 | 37.4 GB | 2.9 min | Very slow |
| 92Excellent quality (A+) · 92 of 100: how capable the model is at FP16, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. | Ideogram 4 | FP16 | 57.3 GB | 17.1 min | Very slow |
| 93Excellent quality (A+) · 93 of 100: how capable the model is at FP16, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. | Qwen-Image 2.1 | FP16 | 57.0 GB | 41.1 min | Very slow |
Best Video models for the AMD EPYC 7002 / 7003 series NAS (512 GB)
| Quality | Model | Quant | Memory | Speed | Runs |
|---|---|---|---|---|---|
| 88Excellent quality (A+) · 88 of 100: how capable the model is at FP16, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. | Cosmos3-Super Image2Video (64B) | FP16 | 139 GB | 25.7 h | Very slow |
| 90Excellent quality (A+) · 90 of 100: how capable the model is at FP16, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. | MiniMax H3 (33B, video + audio) | FP16 | 113 GB | 11.9 h | Very slow |
| 88Excellent quality (A+) · 88 of 100: how capable the model is at FP16, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. | LTX-2.5 (22B, video + audio) | FP16 | 82.5 GB | 62.3 min | Very slow |
| 88Excellent quality (A+) · 88 of 100: how capable the model is at FP16, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. | Kandinsky 6.0 Pro (29B, video + audio) | FP16 | 85.4 GB | 33.6 h | Very slow |
| 86Excellent quality (A+) · 86 of 100: how capable the model is at FP16, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. | LTX-2.3 | FP16 | 75.4 GB | 4.6 h | Very slow |
Best Decision models for the AMD EPYC 7002 / 7003 series NAS (512 GB)
| Quality | Model | Quant | Memory | Speed | Runs |
|---|---|---|---|---|---|
| 78Great quality (A) · 78 of 100: how capable the model is at F16, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. | OpenAI Privacy Filter | F16 | 3.8 GB | 44 ms | Runs great |
| 68Good quality (B) · 68 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. | Llama Prompt Guard 2 22M | Q8_0 | 0.7 GB | 103 ms | Runs great |
| 80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. | GLiNER2.5 Decide | Q8_0 | 1.2 GB | 345 ms | Runs well |
| 86Excellent quality (A+) · 86 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. | Jina Reranker v3.5 | Q8_0 | 39.2 GB | 780 ms | Slow |
| 84Great quality (A) · 84 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. | Jina Reranker v3 | Q8_0 | 39.2 GB | 780 ms | Slow |
| 90Excellent quality (A+) · 90 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. | Qwen3 Reranker 4B | Q8_0 | 42.8 GB | 5.2 s | Very slow |
| 66Good quality (B) · 66 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. | GLiClass Base v3 | Q8_0 | 0.8 GB | 136 ms | Runs well |
| 80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. | FRIDA-Decisions | Q8_0 | 1.7 GB | 574 ms | Slow |
Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.