Ascend910系列:修订间差异
跳到导航
跳到搜索
无编辑摘要 |
无编辑摘要 |
||
第5行: | 第5行: | ||
Main features: | Main features: | ||
1. The Huawei Ascend 910 has 32 DaVinci AI cores. | # 1. The Huawei Ascend 910 has 32 DaVinci AI cores. | ||
2. The chip combines these DaVinci AI cores with high-speed HBM2 memory. | # 2. The chip combines these DaVinci AI cores with high-speed HBM2 memory. | ||
3. The main chip also has additional logic blocks such as 128 channel video decoding engines. | # 3. The main chip also has additional logic blocks such as 128 channel video decoding engines. | ||
4. There is a Nimbus V3 chip that handles most of the I/O and is co-packaged alongside the main chip and the HBM2 stacks. | # 4. There is a Nimbus V3 chip that handles most of the I/O and is co-packaged alongside the main chip and the HBM2 stacks. | ||
5. The Huawei Ascend 910 is designed to run at a higher power envelope (350W) and higher performance compared to the Ascend 310. | # 5. The Huawei Ascend 910 is designed to run at a higher power envelope (350W) and higher performance compared to the Ascend 310. | ||
6. Huawei claims 256 TFLOPs of FP16 performance for the Ascend 910. | # 6. Huawei claims 256 TFLOPs of FP16 performance for the Ascend 910. | ||
7. The Tesla V100 has 112-125 TFLOPs for deep learning thanks to its tensor cores. | # 7. The Tesla V100 has 112-125 TFLOPs for deep learning thanks to its tensor cores. |
2023年11月3日 (五) 14:12的版本
Architecture of Ascend 910 Chips
Main features:
- 1. The Huawei Ascend 910 has 32 DaVinci AI cores.
- 2. The chip combines these DaVinci AI cores with high-speed HBM2 memory.
- 3. The main chip also has additional logic blocks such as 128 channel video decoding engines.
- 4. There is a Nimbus V3 chip that handles most of the I/O and is co-packaged alongside the main chip and the HBM2 stacks.
- 5. The Huawei Ascend 910 is designed to run at a higher power envelope (350W) and higher performance compared to the Ascend 310.
- 6. Huawei claims 256 TFLOPs of FP16 performance for the Ascend 910.
- 7. The Tesla V100 has 112-125 TFLOPs for deep learning thanks to its tensor cores.