Previously, we retested the Intel Arc A770 and A750 graphics cards using the new driver, and the performance of these two graphics cards of Intel under the blessing of the new driver is obviously much better than when they first came out, in fact, Arc A series graphics cards are also a very good productivity tool, the Xe-HPG architecture used in the graphics card is equipped with dual Xe media engines, and also supports Deep Link technology to work with Intel core graphics, Since the new driver optimizes the game, Are there performance optimizations for video manufacturing? Let's test it today.
Let's first review the media engine on the Xe architecture used in Arc A-series graphics cards:
XE Media Engine
Although the Xe-HPG architecture is mainly oriented to gaming applications, Intel has strengthened the Xe-LP on the basis of the Xe-LP is already very comprehensive, and the Xe-HPG architecture can now support hardware codecs in VP9, AVC, HEVC and AV1 formats, up to 8K60 12-bit HDR hard decoding, and 8K 10-bit HDR hard coding.
Among them, AV1 hardware encoding is Intel's first technology to provide, in the Arc A series graphics card released NVIDIA RTX 40 series and AMD Radeon RX 7000 did not come out, this video format has a better compression ratio and picture quality, the efficiency is 50% higher than the common H.264 format, even compared to HEVC is 20% higher, more suitable as the future Internet ultra-high-definition video transmission format, and AV1 format is open source without license fees, At present, well-known software such as HandBrake, Adobe Premiere Pro and FFMPEG already support Intel's AV1 hardware encoding capabilities.
Deep Link technology
Since Intel also has Xe core graphics in its own CPUs above the 12th generation Core, they thought of making Arc series discrete graphics cards work with core graphics, which is called Deep Link technology, which currently provides Stream Assist, super coding and XMX engine three functions, of which Stream Assist is friendly to game live streamers, and can call the core graphics to process game screen streaming, virtual green screen and automatic capture of game highlights, allowing the discrete graphics card to focus on delivering maximum gaming performance.
As for super coding, because the Iris Xe core itself also comes with a hardware codec, this function allows the core display to process the encoding and decoding of video together with the media engine of the Arc series discrete graphics card, and the XMX engine is to integrate the computing power of the Iris Xe core and Arc series discrete graphics cards to speed up the processing speed of some AI projects. However, at present, Deep Link technology cannot be used in games, and it does not allow the core graphics to help increase the game frame rate a little.
Test platform with instructions
The test platform uses the current Intel Core i9-13900KS processor, turns on the core graphics, with the Asus ROG MAXIMUS Z690 EXTREME motherboard, the Grape Flame Halberd DDR5-6000 16GB*2 memory, the test uses the Blue Halberd Arc A750 Photon 8G OC W Asian Games co-branded graphics card, using the latest 4824 version driver and the 3490 WHQL version driver launched on the day of the first release. The reference is NVIDIA's strongest RTX 4090, in fact, NVIDIA RTX 4070 Ti and above have dual video encoders with similar video processing performance.
The graphics card used is the Blue Halberd Arc A750 Photon 8G OC W Asian Games co-branded model, the Blue Halberd has really put a lot of effort into this A750, on the basis of its white appearance added the design concept of the Asian Games, more aesthetic than pure white color matching, the left side of the graphics card back panel added a badge composed of six elements, including the fan, track, Qiantang River, Qianjiang tide head, Internet symbols and sun graphics, and on the right side added the same Qiantang River and Qianjiang Chaotou as the image core composed of an extended design, The overall image symbolizes the Asian Games spirit of unity and cooperation, close embrace and always moving forward.
Video transcoding performance test
The test software includes the popular video manufacturing software Davinci Resolve and Handbrake Nightly transcoding software, both of which support Deep Link technology.
Davinci Resolve
Let's take a look at the video export of Davinci Resolve Studio 18, using our dedicated test project, which exports AV1, H.265, and H.264 formats respectively. Arc A750 encoding transcoding speed after using the latest driver is indeed significantly improved, AV1 and H.265 video export time reduced by 1/3, while H.264 encoding has not changed, you can see that A750's AV1 and H.265 video export performance is already stronger than RTX 4090, and it can also work with the Xe architecture of the core display, the core display is responsible for decoding the independent display is responsible for encoding, further improving work efficiency. The biggest improvement is when exporting H.264 video, and the independent GPU + core graphics achieve the reverse of RTX 4090.
8K video export does not have the results of the old driver because the old driver can not run at all, and because the UHD 770 core graphics can not process 8K video, so the deep link is also invalid, so the above results are Arc A750 running under the latest driver, it can be seen that the A750 whether AV1 or H.265 8K video export speed is faster than RTX 4090.
Handbrake Nightly
Next is Handbrake Nightly's transcoding test, this software actually supports Deep Link, but this thing is off by default, need to be opened manually, but when transcoding, the task manager can not see the load of the kernel at all, but in fact the transcoding rate is increased, the test is to use a 10-minute 4K H.265 video to another format, using H.265 QSV 2160p 4K presets to ensure that the video quality parameters are one to, Export video in an unacceptable format by changing the video encoder format.
At present, the official version of Handbrake does not support NVEnc's AV1 video encoding, and the beta version has a problem and cannot call NVIDIA's video decoder, so RTX 4090's H.264 and H.265 video is measured with the official version, the encoder of these two presets is Fastest, AV1 video transcoding is used in the beta version, because the CPU will be responsible for video decoding at this time, the CPU load is significantly higher than others, and the encoder presets Medium.
Intel's graphics card with the test version or the official version is no difference, the function is complete, the encoder preset is the default Speed, from the test results the new and old drivers have little impact on the transcoding performance, but the 3490 driver to AV1 will report an error, can not be converted at all, the new driver does not have this problem. It can be seen that the transcoding rate is improved after using Deep Link to make the independent GPU and the core graphics work together.
If you talk about graphics card transcoding performance alone, RTX 4090's H.265/H.264 performance is better than A750, but after A750 with UHD770 to open Deep Link, the gap narrows a lot, as for AV1 transcoding, due to software problems RTX 4090 needs to use CPU to decode, so the speed is slower than A750 using hardware decoding.
Full text summary
From the test results, Arc A750's transcoding performance is comparable to the opponent's flagship RTX 4090, in fact, in the early stage of Arc A series of drivers can play most of the transcoding performance of the graphics card, of course, at that time it was not perfect, the latest version of the performance is indeed better, in addition, Intel's independent graphics can also open the Deep Link function with the 12th generation or newer Core processor core graphics, so that the transcoding efficiency is on a higher level, in terms of compatible video formats, It supports mainstream video formats such as AV1, H.265, and H.264, which is a very efficient independent display for video workers.
After talking about the performance and then look at the price factor, because the Arc A series of graphics cards have a complete two sets of Xe media engines, so in theory, A380, A750, A770 graphics card transcoding efficiency are similar, of course, GPU frequency, memory bandwidth, memory capacity will have some impact, comprehensive these things, 1699 yuan Arc A750 is actually a better choice, the price is not high, the overall performance is much better than A380, transcoding efficiency can be compared with the opponent's flagship RTX 4090 board, the difference between the two is close to 10 times, in fact, RTX 4070 Ti's transcoding performance is also similar to RTX 4090, but the price is also more than three times the A750, it can be seen that A750 as a video acceleration card is quite cost-effective, with Intel's own processor can assemble a set of cost-effective and efficient video workstations.