NVIDIA T4 has only 15g of memory?

Hi everyone,
NVIDIA officially states that T4 memory is 16GB, but I see about 15GB. What’s the reason?
Tue Mar 23 18:36:05 2021
±----------------------------------------------------------------------------+
| NVIDIA-SMI 418.67 Driver Version: 418.67 CUDA Version: 10.1 |
|-------------------------------±---------------------±---------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Tesla T4 Off | 00000000:21:01.0 Off | 0 |
| N/A 33C P0 26W / 70W | 0MiB / 15079MiB | 0% Default |
±------------------------------±---------------------±---------------------+

±----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| No running processes found |
±----------------------------------------------------------------------------+

When you run “nvidia-smi -q”, what shows under the “Retired Pages” section?

Hard to believe that 1GB worth of pages have been failed though…

My first thought was that the “missing” memory is the memory needed for ECC, but best I can decipher the output from nvidia-smi displayed above, ECC is turned off on this GPU?

My recollection is that modern GPUs with ECC support need 6.25% of the memory for ECC. Which for 16GB of memory works out to exactly 1 GB.

[Later:]

Ah, here, from the Best Practices Guide:

Note: On GPUs with GDDR memory with ECC enabled the available DRAM is reduced by 6.25% to allow for the storage of ECC bits. Fetching ECC bits for each memory transaction also reduced the effective bandwidth by approximately 20% compared to the same GPU with ECC disabled, though the exact impact of ECC on bandwidth can be higher and depends on the memory access pattern. HBM2 memories, on the other hand, provide dedicated ECC resources, allowing overhead-free ECC protection

https://docs.nvidia.com/cuda/cuda-c-best-practices-guide/index.html