← All GPUs
NVIDIA · Consumer · Blackwell (GB203)
NVIDIA GeForce RTX 5070 Ti
No Founders Edition; length and slot width vary by partner card.
VRAM
16 GB
GDDR7
Memory bandwidth
896 GB/s
FP16 compute
87.9 TFLOPS
dense tensor
Price
~$749
new, maker or suggested price
What fits at 4-bit
The 4-bit file plus 8,192 tokens of context and 1.5 GB of overhead, against the VRAM of one card. Speeds are measured medians where someone has published a run at that card count; a blank speed means no run, not a slow one. Machine pages carry estimates for the gaps.
| Model | Needs | 1 card (16 GB) |
|---|---|---|
| Qwen3.5 4B4B · Q4_K_M | 4.3 GB | fits |
| Llama 3.1 8B Instruct8B · Q4_K_M | 7.1 GB | 50 tok/smedian of 2 runs |
| Qwen3 8B8.2B · Q4_K_M | 7.3 GB | 121 tok/smedian of 1 run |
| Gemma 4 E4B IT8B · Q4_K_M | 6.5 GB | fits |
| Granite 4.2 8B8.8B · Q4_K_M | 7.9 GB | fits |
| Qwen3.5 9B9B · Q4_K_M | 7.0 GB | fits |
| Gemma 4 12B IT11.95B · Q4_K_M | 8.2 GB | fits |
| Ministral 3 14B Reasoning (2512)13.9B · Q4_K_M | 10.4 GB | fits |
| Phi-414.7B · Q4_K_M | 11.3 GB | fits |
| gpt-oss-20b21B, 3.6B active · MXFP4 (native) | 12.5 GB | 173 tok/smedian of 2 runs |
Measured speeds
One card
| Model | Quant | Backend | Generation | Prompt | Context | Source |
|---|---|---|---|---|---|---|
| gpt-oss-20b | MXFP4 MoE | llama.cpp b6210 CUDA | 189 tok/s | 6,340 tok/s2,048-token prompt | 0 | llama.cpp benchmark discussion2025-08 |
| gpt-oss-20b | MXFP4 | llama.cpp (build not stated) | 156 tok/s | 6,178 tok/s4,096-token prompt | 4,096 | review2026-03 |
| Llama 3.1 8B Instruct | Q4_K_M | LocalScore (llamafile; version not stated) | 66 tok/s | 3,657 tok/s | 0 | LocalScore2026-10 |
| Llama 3.1 8B Instruct | Q4_K_M | LocalScore (llamafile; version not stated) | 35 tok/s | 2,575 tok/s | 0 | LocalScore2026-10 |
| Qwen3 8B | Q4_K | llama.cpp (build not stated) | 121 tok/s | 5,557 tok/s4,096-token prompt | 4,096 | review2026-03 |
Speculative decoding, several users, or long context
Kept apart because each measures a different workload from one person chatting near the start of a conversation.
| Model | Quant | Backend | Generation | Prompt | Context | Measured on | Source |
|---|---|---|---|---|---|---|---|
| gpt-oss-20b | MXFP4 | llama.cpp (build not stated) | 133 tok/s | 4,941 tok/s16,384-token prompt | 16,384 | RTX 5070 Ti | review2026-03 |
| gpt-oss-20b | MXFP4 | llama.cpp (build not stated) | 116 tok/s | 3,904 tok/s32,768-token prompt | 32,768 | RTX 5070 Ti | review2026-03 |
| Qwen3 8B | Q4_K | llama.cpp (build not stated) | 88 tok/s | 3,654 tok/s16,384-token prompt | 16,384 | RTX 5070 Ti | review2026-03 |
| Qwen3 8B | Q4_K | llama.cpp (build not stated) | 63 tok/s | 2,269 tok/s32,768-token prompt | 32,768 | RTX 5070 Ti | review2026-03 |
Specifications
- Memory
- 16 GB GDDR7
- FP16 basis
- Dense FP16/BF16 tensor with FP32 accumulate; FP16 accumulate 175.8 dense (351.5 sparse).
- Board power
- 300 W
- PCIe
- Gen 5 x16
- NVLink
- No
- Cooling
- Open-air
- Software
- CUDA, Vulkan
- Released
- 2025
- Availability
- sold new
- Price basis
- NVIDIA launch MSRP; no Founders Edition (partner cards only, dimensions vary). Jan 2026 reports say ASUS stopped making new 16 GB 5070 Ti batches; street prices ran above MSRP during the 2026 memory shortage.
- Maker page
- nvidia.com
Sources
- nvidia.com: 16 GB GDDR7 256-bit, 300 W, dimensions vary by manufacturer, PCIe Gen 5
- images.nvidia.com: NVIDIA RTX Blackwell whitepaper appendix tables: memory, bandwidth, TGP, PCIe gen, dense/sparse FP16 tensor TFLOPS
- nvidia.com: NVIDIA GeForce compare page: RTX 50 series memory bandwidth row; no RTX 50 SUPER listed
- nvidianews.nvidia.com: NVIDIA launch pricing: RTX 5090 $1,999, RTX 5080 $999, RTX 5070 Ti $749
- hostzealot.com: Jan 2026 report (ASUS / Hardware Unboxed) that 16GB RTX 5060 Ti and 5070 Ti production stopped