ASUS TUF RTX 3090 Used Review | Generated by AI

Home 2026.07 English 中文

---

Question: How good is this ASUS TUF RTX 3090 24GB second-hand GPU listing?

Answer:

This is actually one of the better RTX 3090 used listings, especially for AI/LLM work.

Specs:

• GPU: ASUS TUF Gaming GeForce RTX 3090 24GB
• VRAM: 24GB GDDR6X (the main reason to buy it)
• Architecture: Ampere
• Compute capability: 8.6
• Tensor cores: 328
• FP16 Tensor performance: ~142 TFLOPS
• Memory bandwidth: ~936 GB/s

For AI:

Model workload	RTX 3090 24GB
Llama/Qwen 7B inference	✅ Excellent
14B models quantized	✅ Good
32B Q4 inference	✅ Possible
LoRA fine-tuning 7B-14B	✅ Very good
Full fine-tuning	❌ Too small
GPT-2 760M / nanoGPT training	✅ Comfortable

Compared with your RTX 4070:

	RTX 4070	RTX 3090
VRAM	12GB	24GB
Memory bandwidth	504 GB/s	936 GB/s
FP16 Tensor	~60 TFLOPS	142 TFLOPS
Power	200W	350W
CUDA	same generation support	same

For LLM development, the 3090 is a big jump because VRAM is the bottleneck.

---

The seller claims:

3dmark测试99.3 烤鸡温度70度左右

Good signs:

✅ Stress temperature 70°C is very good for a 3090. 3090s often run:

• GPU core: 75-85°C
• Memory junction: 95-110°C

The missing information:

Need memory temperature.

3090 has hot GDDR6X memory on both sides of PCB.

Ask seller:

GPU-Z screenshot:
- GPU temperature
- Memory junction temperature
- Hot spot temperature
- ASIC quality
- BIOS version

The key number:

• Memory junction < 90°C → excellent
• 90-100°C → normal

• 105°C → avoid

---

Age:

20年的卡

Meaning 2020 card.

So it is ~6 years old.

Risk:

• Fan bearing aging
• Thermal pads drying
• Mining history possible

However ASUS TUF is among the better 3090 designs:

• large heatsink
• good VRM
• better cooling than many blower/reference cards

---

Price determines everything.

Approximate 2026 China used market:

• RTX 3090 normal: ¥2500-3200
• ASUS TUF / STRIX: ¥2800-3500
• suspicious cheap: <¥2200

My valuation:

Price	Judgment
<2500 CNY	🔥 Buy
2500-3000	👍 Good
3000-3500	OK only if excellent condition
>3500	Not attractive

---

For your use case (LLM training, ROCm experiments, local models):

I would rank:

1. RTX 3090 24GB ⭐⭐⭐⭐⭐
2. RTX 4090 24GB ⭐⭐⭐⭐⭐ (too expensive)
3. RTX 3090 Ti 24GB ⭐⭐⭐⭐
4. RTX 4070 Ti Super 16GB ⭐⭐⭐
5. RX 7900 XT 20GB ⭐⭐⭐ (ROCm advantage but ecosystem weaker)

The 3090’s killer feature is simply:

24GB CUDA VRAM at used-market price.

For AI engineering, a 3090 often beats a newer 12GB/16GB card because many models simply don’t fit.

Before buying, I would ask for:

1. GPU-Z full screenshot
2. GPU memory temperature under FurMark/OCCT
3. Run:
   nvidia-smi -q -d MEMORY
4. Any mining history?
5. Original invoice?

If the price is around ¥2800-3000, this is a strong candidate.

---

Generated by AI. Curating and sharing still takes effort. If you find it useful, feel free to donate. WeChat: @lzwjava · X: @lzwjava · Say hi 👋

Back Donate