Tier 2 survived on INT4 7B. Tier 3 is the workstation infer box: 24–48 GB VRAM (RTX 4090 24 GB, RTX 6000 Ada / A6000 48 GB), running 13B–34B in FP16 / BF16—or larger models still quantized. Volume 12 KV cache and FlashAttention decide whether that 24 GB feels like 18 GB usable. This is still mostly inference + light PEFT experiments, not multi-GPU QLoRA farms (Tier 4) or NCCL clusters (Tier 5).
Learning Objectives
By the end of this lesson, students should be able to:
- Map 24 GB vs 48 GB to 13B/34B FP16 fit, including KV headroom.
- Explain why BF16 is preferred on Ampere+ workstation/datacenter vs FP16 on some consumer paths.
- Size vLLM / TensorRT-LLM / HF serve for a handful of concurrent users.
- Know when 70B INT4 on 48 GB is in-tier vs when 70B FP16 is not.
- Contrast 4090 (fast consumer) vs A6000 (capacity, ECC, workstation).
- Draw the line to Tier 4 when LoRA/QLoRA becomes the job, not chat.
Tier 3 (Local Advanced) is a single high-end GPU with 24–48 GB of VRAM used primarily for high-quality local inference of mid-size LLMs in FP16/BF16 (13B class comfortably; 34B class on 48 GB or with quant on 24 GB) plus embeddings, vision towers, and modest speculative-decode drafts. Host RAM should be 64 GB-class so dataloaders and tokenizers are not fighting the GPU. Multi-GPU NVLink training is out of scope unless you are peeking at Tier 4.
Fit Table (weights FP16/BF16 ≈ 2 bytes/param)
| Model | FP16 weights | 4090 24 GB | A6000 48 GB | Notes |
|---|---|---|---|---|
| 7B FP16 | ~14 GB | Easy + long KV / batch | Easy; overkill | Tier 2 could already INT4 this |
| 13B FP16 | ~26 GB | No FP16; yes INT8/INT4 or 2×GPU | Yes + KV room | 4090 often runs 13B quantized |
| 34B FP16 | ~68 GB | No | No FP16; INT4/INT8 maybe | 34B FP16 wants 80 GB or multi-GPU |
| 34B INT4 | ~18–22 GB | Tight yes (watch KV) | Comfortable | Quality vs 13B FP16 is a product call |
| 70B INT4 | ~35–40 GB | No | Maybe + short ctx | 70B FP16 is Tier 5 / 2×80 GB |
Read the table slowly: the module title’s “13B–34B FP16/BF16” is the ambition band, not a promise that every 34B FP16 fits a 4090. 48 GB is the card that makes 13B FP16 honest; 34B FP16 usually still needs quant or more silicon. That honesty is the playbook.
4090 vs A6000 vs “just quantize”
RTX 4090 (24 GB)
- Huge Ada tensor-core throughput
- Best $/infer for 7B FP16, 13B INT4, SDXL
- No ECC; 24 GB ceiling; noisy desktop
A6000 / 6000 Ada (48 GB)
- 13B FP16 + real KV/batch
- ECC, quieter, multi-GPU PCIe possible
- Slower than 4090 on some infer kernels
Stay quantized anyway
- 34B/70B on one card almost always INT4/INT8
- FP16 is for quality-sensitive 7–13B
- Vol. 19 will make you measure, not assume
Serve Sketch (vLLM-minded)
Tier 3 is where continuous batching, paged KV, and FlashAttention stop being trivia and start being capacity. Pin dtype explicitly.
When Tier 3 Is the Wrong Spend
Buy/rent Tier 3 when
- Local 13B FP16 quality or 34B INT4 privacy sandbox
- SDXL / FLUX FP16 (Vol. 17) plus a small LLM on 24–48 GB
- A few internal users, OpenAI-compatible vLLM on-LAN
- You outgrew 3060 KV limits, not just “want RGB”
Do not
- Expect 70B FP16 on one 4090
- Run serious multi-epoch QLoRA on 34B as if this were Tier 4
- Ignore PSU/thermals (4090 450 W class)
- Skip Tier 1 if API quality still dominates TCO
Related Lectures
| Lecture | Why it sits beside Tier 3 |
|---|---|
| Vol. 06 FP16 / BF16 | The dtypes in the tier name |
| Vol. 06 VRAM | 24 vs 48 GB is the whole game |
| KV Cache / FlashAttention | Usable capacity vs brochure GB |
| TensorRT | Optional compile path on this SKU |
| Vol. 12 LoRA | Light adapters possible; heavy FT → Tier 4 |
| Tier 4 | When training is the job |
“24 GB VRAM runs any 13B FP16.” 13B × 2 bytes ≈ 26 GB weights before KV. A 4090 runs 13B well in quant or with offload; 13B FP16 is an A6000-class story. Second: “34B FP16 on 48 GB.” 34B × 2 ≈ 68 GB. Third: “BF16 and FP16 are interchangeable everywhere.” Vol. 06: BF16 has FP32-range exponents; consumer FA/TensorRT builds differ—measure.
Knowledge Check
- Short Answer: What VRAM band defines Tier 3? Answer: About 24–48 GB (4090 / A6000 class).
- True/False: Llama-13B FP16 weights alone fit in 24 GB with room for 8k KV. Answer: False—weights ~26 GB already exceed 24 GB.
- Multiple Choice: Honest 13B FP16 single-GPU home: (a) 3060 12 GB, (b) A6000 48 GB, (c) phone NPU. Answer: (b).
- Short Answer: ~FP16 size of a 34B model? Answer: ~68 GB (34e9 × 2 bytes).
- True/False: 34B INT4 can be in-tier on a 24–48 GB card with short/medium context. Answer: True (tight on 24 GB).
- Multiple Choice: Why FlashAttention matters more here than on Tier 1: (a) CSS, (b) usable activation/KV headroom on a finite GPU, (c) API keys. Answer: (b).
- Short Answer: Name one 4090 vs A6000 trade. Answer: 4090 faster/24 GB vs A6000 48 GB + ECC/workstation (or thermals/PSU).
- True/False: 70B FP16 is a Tier 3 single-card workload. Answer: False—needs ~140 GB weights or multi-80 GB / quant+pipeline (Tier 5-ish).
- Multiple Choice: Preferred dtype on A100/A6000 infer: (a) FP64, (b) BF16 when supported, (c) INT32. Answer: (b).
- Short Answer: When do you leave Tier 3 for Tier 4? Answer: When LoRA/QLoRA (or heavier FT) is the primary job, especially multi-GPU 48–80 GB.
Key Takeaways
- Tier 3 = 24–48 GB infer workstation; 13B FP16 wants 48 GB; 34B FP16 is not a 4090 promise.
- Quant still matters; FP16/BF16 is for the models that actually fit.
- vLLM + FA + paged KV turn brochure GB into concurrent users.
- 4090 = speed; A6000 = capacity/ECC.
- Continue with Tier 4 Fine-Tuning Workstation.
Lab: Students compute usable VRAM at 80% for 24 vs 48 GB, then place 8B/13B/34B/70B at FP16 vs INT4 with 4k and 16k KV. Debate one purchase: 4090 vs used A6000 vs stay API.
Discussion: Is “34B FP16 on Tier 3” marketing or math? Make them say quant, multi-GPU, or smaller model.
Recap: Tier 3 is 24–48 GB local infer for mid-size FP16/BF16 (and quantized larger) models. Continue with Tier 4.