Intel has some Arrow Lake on "Intel 3"?

[Xebec]

I really see value of using an Intel CPU:

https://twitter.com/x/status/1888962597065822233

AMD did "remedy" this with the AM5 platform. Almost all Ryzen 7000 and 9000 CPUs now include an integrated GPU. The only exceptions are the rare chips like 7500F, Intel has the same of course.

Mobile AMD has had an iGPU for a while of course.

Example, Ryzen 7600 CPU: https://www.amd.com/en/products/processors/desktops/ryzen/7000-series/amd-ryzen-5-7600.html

Graphics Model AMD Radeon™ Graphics
Graphics Core Count 2
Graphics Frequency 2200 MHz
USB Type-C® DisplayPort™ Alternate Mode Yes

 

[XYang2023]

AMD did "remedy" this with the AM5 platform. Almost all Ryzen 7000 and 9000 CPUs now include an integrated GPU. The only exceptions are the rare chips like 7500F, Intel has the same of course.

Mobile AMD has had an iGPU for a while of course.

Example, Ryzen 7600 CPU: https://www.amd.com/en/products/processors/desktops/ryzen/7000-series/amd-ryzen-5-7600.html

Graphics Model AMD Radeon™ Graphics
Graphics Core Count 2
Graphics Frequency 2200 MHz
USB Type-C® DisplayPort™ Alternate Mode Yes

I think Intel iGPU is generally better. It has Quicksync. Maybe the software support is also better. It was quite impressive as the OBS was recording the training secession without affecting the performance of discrete GPU during training.

 

[Paul2]

I think Intel iGPU is generally better. It has Quicksync. Maybe the software support is also better. It was quite impressive as the OBS was recording the training secession without affecting the performance of discrete GPU during training.

Quicksync lives almost completely separate from GPU pipeline, and can be DMAed efficiently, while AMD's video encoding/decoding is either write-back, or it has to be scaled, post-processed, and displayed insider the GPU pipeline. This is why AMD GPUs are weak for video/image re-encoding.

Under Windows, it's hidden by the acceleration abstraction layers, and people normally don't notice until they see that re-encoding speed is just terrible vs. rendering to screen. In embedded applications, where you normally do your scaling, and post-processing in main memory under OS's control, this doesn't fly well, and here also comes the terrible power efficiency penalty from write-back.

 

[XYang2023]

Quicksync lives almost completely separate from GPU pipeline, and can be DMAed efficiently, while AMD's video encoding/decoding is either write-back, or it has to be scaled, post-processed, and displayed insider the GPU pipeline. This is why AMD GPUs are weak for video/image re-encoding.

Under Windows, it's hidden by the acceleration abstraction layers, and people normally don't notice until they see that re-encoding speed is just terrible vs. rendering to screen. In embedded applications, where you normally do your scaling, and post-processing in main memory under OS's control, this doesn't fly well, and here also comes the terrible power efficiency penalty from write-back.

I think Quicksync is quite cool, for example, OBS has dedicated Qucksync Encoder.

 

[siliconbruh999]

Intel's Quick Sync Media encoders are even better than NVIDIA's they support more codecs than NVIDIA like H.265 4:2:0 which NVIDIA started supported with Blackwell

 

[XYang2023]

On Intel Arrow Lake H:

Intel's Core Ultra 200 laptop CPUs deliver shocking performance gains

The performance of Intel's Core 285H mobile chip inside a MSI Prestige 16 AI Evo laptop is surprisingly impressive and overcame the "Arrow Lake" processor's flop on the desktop.

www.pcworld.com

 

[XYang2023]

https://twitter.com/x/status/1896898711994958106