Intel ARC - General Discussion of Newest GPU Competitor

RogerS wrote on 3/25/2023, 10:09 PM

Could you please do the benchmarks in my signature so we can see the performance in VEGAS vs AMD and NVIDIA?

I think ARC is an exciting platform as price to performance is very reasonable.

Howard-Vigorita wrote on 3/25/2023, 10:49 PM

I have an a770 and an a380 in my 11900k system. The a770 is the quicker of the two encoding qsv. Where they excel in Vegas is at decoding. What they're not so good at is timeline and fx processing... I would expect your 5600xt would be better. I have a 5700xt in a different system and it's quite a bit faster at that. My 11900k system also has a 6900xt and that's it's main gpu.

I've just been experimenting with Intel Hyper encoding qsv with QsvEncC64 and it knocked a test encode down from 51 sec to 16 sec. Unfortunately, Vegas doesn't have that natively yet.

wwaag wrote on 3/26/2023, 12:46 AM

As the HOS developer, I installed an A380 a couple of months ago primarily for testing GPU encoding using AV1 and VP9 formats. Recently, I've been capturing quite a lot of Hero 11 5.3K 60P 10bit footage which wouldn't even "play" on my 8700K system using the A380 with such players as MPC-HC and VLC. Last week I installed an A770 LE and it is working reasonably well. (I initially ordered an RTX 4070ti but learned on delivery that it was way too long for my case--my bad for not checking.) The Hero footage even previews in Vegas although use of a proxy is still warranted. One thing of interest to HOS users--both the A380 and A770 can successfully create Vegas-style proxies, something which I was unable to do with the native Intel 630 GPU in the 8700. More importantly, speed. Just tested a 1:04 sec clip. Proxy creation in Vegas took roughly 4:05--the KwikProxy script using the A770 took 51 sec--a little more than 5 times as fast.

Dexcon wrote on 3/26/2023, 4:36 AM

I have ... an a380 in my 11900k system ...

I installed an A380 a couple of months ago ...

I am very impressed. I tried installing an A380 -

... but I think that I need a bigger computer case 😄.

walter-i. wrote on 3/26/2023, 5:10 AM

I have ... an a380 in my 11900k system ...

I installed an A380 a couple of months ago ...

I am very impressed. I tried installing an A380 -

... but I think that I need a bigger computer case 😄.

Joker! 😂

RogerS wrote on 3/26/2023, 5:24 AM

I'd love to see more benchmarks of the Intel ARC cards in different systems, but the VP sample project with Howard's a770 is GPU heavy for timeline and Fx and it holds its own with decent AMD and NVIDIA cards (including my RTX 2080 Super which is somewhere around a 3060Ti for general performance).

IAM4UK wrote on 3/26/2023, 1:23 PM

Could you please do the benchmarks in my signature so we can see the performance in VEGAS vs AMD and NVIDIA?

I think ARC is an exciting platform as price to performance is very reasonable.

I ran both 4K and HD, and filled in the online forms. Here were the key results:
4K render time: 1 minute, 12 seconds
HD render time: 39 seconds

IAM4UK wrote on 3/26/2023, 1:25 PM

As the HOS developer, I installed an A380 a couple of months ago primarily for testing GPU encoding using AV1 and VP9 formats. Recently, I've been capturing quite a lot of Hero 11 5.3K 60P 10bit footage which wouldn't even "play" on my 8700K system using the A380 with such players as MPC-HC and VLC. Last week I installed an A770 LE and it is working reasonably well. (I initially ordered an RTX 4070ti but learned on delivery that it was way too long for my case--my bad for not checking.) The Hero footage even previews in Vegas although use of a proxy is still warranted. One thing of interest to HOS users--both the A380 and A770 can successfully create Vegas-style proxies, something which I was unable to do with the native Intel 630 GPU in the 8700. More importantly, speed. Just tested a 1:04 sec clip. Proxy creation in Vegas took roughly 4:05--the KwikProxy script using the A770 took 51 sec--a little more than 5 times as fast.

Just wanted to tell you, Wayne, that your work on Happy Otter Scripts is amazing. I am thankful I purchased them, and you have transformed the Vegas Pro experience. Thank you very much.

animagic wrote on 3/26/2023, 1:52 PM

I use my PC for multiple programs, some of which require NVidia-specific features. That said, the introduction of a new brand of GPU may help lower prices of the competitors, so that's always a good thing.

RogerS wrote on 3/26/2023, 8:52 PM

@IAM4UK Thanks for filling out the benchmark!

Can you please check all settings with the [EDIT] HD one and try it again? 9 seconds isn't likely based on the playback framerate and your UHD render time (it's the fastest render recorded for HD... by about 15 seconds or so).

Overall the Arc card holds its own even on this GPU heavy benchmark (which stresses computational ability, not decoding where we know the Arc shines).

IAM4UK wrote on 3/27/2023, 8:00 AM

@IAM4UK Thanks for filling out the benchmark!

Can you please check all settings with the UHD one and try it again? 9 seconds isn't likely based on the playback framerate and your UHD render time (it's the fastest render recorded... by about 15 seconds or so).

Overall the Arc card holds its own even on this GPU heavy benchmark (which stresses computational ability, not decoding where we know the Arc shines).

@RogerS
Sorry for the mistake. The lower time was for HD; the higher time was for UHD/4K.
The lower time should not be 9 seconds, but 39 seconds. (I think my mistake was having the "render loop region only" checked on that specific render, after testing timeline performance.
Can you alter the spreadsheet?

RogerS wrote on 3/27/2023, 8:31 AM

Yes, you had the HD with the lower time... it was just too low at 9 seconds! I fixed it and resorted it.

If you have time to try the VP 20 benchmark it tests Vegas quite differently. You're also guaranteed to have the highest Intel ARC score (as you'll be the first to complete it).

IAM4UK wrote on 3/27/2023, 10:39 AM

@RogerS
I did the VP20 Ad benchmark. It only used 20% of CPU and only 5% of GPU, and thus took a very long time (12:45). What is different about that veg file that causes it to not use resources in rendering?

RogerS wrote on 3/27/2023, 10:42 AM

Hard to say but very interesting! I assume this is with the AI section off. Could you share a screenshot of your GPU's activity in task manager/performance during the render (so we can see more than just the total %)?

RogerS wrote on 3/27/2023, 10:45 AM

I noticed you set dynamic ram preview to 0. For the purposes of this test could you also test it at the default of 5%? It may make a big difference (see my laptop near the bottom where the time more than doubled with DRP disabled).

How was playback of the preview section? Try that with DRP at defaults, too.

IAM4UK wrote on 3/27/2023, 12:56 PM

@RogerS
I experimented with the Dynamic RAM Preview, and found that it made a HUGE difference. With 5% Dynamic RAM, the CPU got used about 60%. WIth 25% Dynamic RAM (in this 64GB RAM system), the CPU got fully used. In both of those cases, the GPU got used a widely-varying amount throughout the rendering process...from 1% to about 30%. The rendering time was less than half of what it was with Dynamic RAM set to 0.
The attached screenshots are with Dynamic RAM set to 0.

RogerS wrote on 3/27/2023, 6:46 PM

Yes, it's a really important preview buffer. Thanks for filling out a second one with it on as it lets others know the difference between on and off.

For preview does NA mean it was <1fps?

GPU usage should vary as the project itself is more and less demanding in different places. Overall the project is a good GPU workout! I'm surprised the Arc didn't do better here as it's got a lot of vram and did well on the other benchmark. Perhaps try it again as drivers update in case there's more performance to be had.

IAM4UK wrote on 3/27/2023, 7:10 PM

@RogerS
na means I did not do the preview test. If I were guessing, it would be about 12-15 fps

RogerS wrote on 3/27/2023, 7:14 PM

Just select that part of the timeline and hit the spacebar. I would guess it's ~3fps fps based on your render performance but hard to say as the render times and preview times can be quite different and we have no other ARC examples.

IAM4UK wrote on 3/27/2023, 7:50 PM

Just select that part of the timeline and hit the spacebar. I would guess it's ~3fps fps based on your render performance but hard to say as the render times and preview times can be quite different and we have no other ARC examples.

From Marker 1-2, the preview framerate was a rock-solid 24.000 fps when I kept Dynamic RAM Preview at 25%, and used the "Good/Quarter" setting I typically use; however, at the specified 0 MB RAM and "Best/Full" the preview framerate was a pitiful 1.6 fps.

RogerS wrote on 3/27/2023, 7:52 PM

Thanks, dynamic ram lets it cheat : ) If everyone scored 24fps it wouldn't be useful as a benchmark!

Howard-Vigorita wrote on 3/29/2023, 4:42 PM

Ran a new AI-TX Torture Test on my Intel Arc a770 with vp20 build 370 on my 11900k system. The test I used is a little different from the Original... it uses 4k AVC media I found on the Vegas Hub which turns out to be tougher than the generated media built into the original. The media length and AI-FX applied are the same however. I was surprised to see performance just under that of the original test run with build 326 on an Amd 5700xt. Full results, download, and submission form are available here.

Btw, my 11900k system has multiple gpus... I usually use the 6900xt as my main one. For this test I switched them. So the Intel Arc was doing everything but the render. I always see a performance increase in general usage with multiple gpus. But for AI processing alone, I think inserting only 1 gpu might actually perform better. Filling both x16 slots drops them down to x8. That probably makes a difference with all the vram data being moved about during AI processing. @IAM4UK, if you have a chance, give it a try.