I am not a gamer. I bought this laptop to do small sample size deep learning training for transformer language models. I use cloud GPU and TPU for the main training with full sample size. But I really wanted to have something locally to run fine tuning and debugging with smaller sample sizes. I was OK just using the cloud for image deep learning, but language models get truly huge! I haven’t trained anything yet, but I’m really excited to test it out. NVIDIA RTX 3070 or 3070 Ti seems to be the sweet spot for a reasonable cost vs performance in June 2022. I also though about a desktop, but got tired of waiting for GPU prices for desktops to come down. One thing I was concerned about was the screen being FHD and not QHD and brightness only 300 nits. But the quality is excellent anyway and colors pop for Netflix and video watching. Pixels can be seen, but images are still very smooth. QHD will be better, but it cost $600 more for a similar laptop. Fan noise, heat, were pleasant surprise when running office type tasks. CPU temp was 50 C, body was luke warm, and fan was silent. Track pad is very smooth. Keys are springy and comfortable and not too loud. Audio is excellent too. Speakers are good quality on music and very loud. Headphone jack output is of high quality and can drive all of your best music. Blue tooth headphone paired easily and sounds OK. Not as good as plugged-in using high quality headset, obviously, but same as usual. This will also make a good media watching device since the screen quality and audio is so nice. Note, there is NO videocam. Can’t do video calling without a plug-in video camera. I already have a USB video cam, but just a reminder. There is a built-in microphone for voice commands, dictation, Cortana. It seems to be a good quality. It pickec up my voice no problem and sounded clear.
Update: I trained 200 epoch transformer language model, text generation task (Francois Cholet, 2nd ed, chapter 12 part 1). It took half the time per epoch as free tier Google Colab GPU. I thought it will be slower, maybe take 50-100% longer per epoch. Was pleasantly surprised. Now I can run all Colab free tier GPU/TPU training on my laptop locally, which is much easier than moving large data and model files around. Also I hate editing Jupyter notebook directly on Colab, which can seriously lag when a lot of people are using their services. CPU temps are 85C and GPU temps are 75C during training. I lifted that back up by 2" to help with air flow. I can use manual mode to up the fan to 100% and I can keep it 5% coller during heavy training on GPU/CPU. For light tasks, CPU is 65C and GPU os 0. Fan speed "Windows" seem to keep things quiet and cool at the same time for light tasks.
| Standing screen display size | 17.3 Inches |
|---|---|
| Screen Resolution | 1920 x 1080 pixels |
| Max Screen Resolution | 1920 x 1080 Pixels |
| Processor | 4.5 GHz ryzen_9 |
| RAM | 8 GB DDR4 |
| Memory Speed | 3200 MHz |
| Hard Drive | 1 TB M.2 NVMe PCIe 3.0 SSD |
| Graphics Coprocessor | NVIDIA GeForce RTX 3070 |
| Chipset Brand | NVIDIA |
| Card Description | Dedicated |
| Graphics Card Ram Size | 8 GB |
| Wireless Type | Bluetooth, 802.11ax |
| Number of USB 3.0 Ports | 4 |
