GPU Between RTX 4070 and A6000

GPU Between RTX 4070 and A6000 | Generated by AI

Home 2026.07

Question: what GPU is between RTX 4070 and RTX A6000 Pro?

If you mean AI / CUDA compute / LLM workload, the ladder is roughly:

RTX 4070 12GB
      |
      |  (more VRAM / workstation)
      v
RTX 3080 Ti 12GB
RTX 3090 24GB
      |
      v
RTX A5000 24GB
      |
      v
RTX A5500 24GB
      |
      v
RTX A6000 48GB

([NVIDIA][1])

For your use case (LLM training, inference, ROCm experiments, local models), the interesting “middle” cards are:

GPU	VRAM	Architecture	AI value
RTX 4070	12GB GDDR6X	Ada	Fast, but VRAM limited
RTX 3090	24GB GDDR6X	Ampere	Best cheap AI card
RTX A5000	24GB GDDR6 ECC	Ampere workstation	3090 slower but professional
RTX A5500	24GB GDDR6 ECC	Ampere workstation	Between A5000/A6000
RTX A6000	48GB GDDR6 ECC	Ampere workstation	Big model card

([NVIDIA][1])

My recommendation for your situation

Since you already have:

RTX 4070
AMD MI50 16GB
interested in GPT training / Qwen / local LLM

I would look at:

1. RTX 3090 24GB (best price/performance)

Probably the sweet spot.

Why:

2x VRAM of RTX 4070
CUDA ecosystem
can run:
- 7B/14B/32B quant models easily
- fine-tune small LLMs
- LoRA on bigger models

The downside:

350W power
used cards may have mining history

2. RTX A5000 24GB

A workstation version.

Advantages:

ECC memory
blower-style cards exist
lower power (~230W)
professional drivers

But:

For pure AI training:

RTX 3090 > RTX A5000

because 3090 has much higher compute.

3. RTX A6000 48GB

This is the real jump.

It is not “a little faster”.

The important thing is:

RTX 4070:
12GB VRAM

RTX A6000:
48GB VRAM

4x memory capacity.

([NVIDIA][1])

You can move from:

Qwen 7B / 14B
        |
        v
Qwen 32B
        |
        v
70B quant inference

without multi-GPU tricks.

For your “one person lab” idea, I would rank:

Price/performance:
RTX 3090 24GB
        >
RTX A5000 24GB
        >
RTX A6000 48GB

Capability:
RTX A6000
        >
RTX 3090
        >
RTX 4070

The missing GPU between your RTX 4070 and A6000 is basically RTX 3090 24GB. It is the most common “AI hacker GPU”. ([TechSpot][2])

For used-market prices, a rough target:

RTX 3090: ¥4000-6000
RTX A5000: ¥5000-8000
RTX A6000: ¥8000-15000+

(depending heavily on China second-hand market).

[1]: https://www.nvidia.com/en-us/products/workstations/rtx-a6000/?utm_source=chatgpt.com “NVIDIA RTX A6000 For Powerful Visual Computing	NVIDIA”
[2]: https://www.techspot.com/specs/gpu/254404-nvidia-geforce-rtx-4070.html?utm_source=chatgpt.com “Nvidia GeForce RTX 4070 Specs	TechSpot”

Back Donate