Graphics Cards Use Compared In MEP 2021 to MMS 2023

AAProds wrote on 5/22/2023, 7:50 AM

This is just an after-dinner ramble about some observations I've made since upgrading my system. As most of you know, I had an old i5 with no effective graphics card. Going from MEP 2021 to MMS 2023 didn't do anything for me, performance-wise. But now that I have the full kit and kaboodle with a grunty CPU/iGPU (UHD 770) and a decent GPU, things are all so different.

In MEP 2021 (and all previous versions) there was a tick box for "calculate video effects on GPU" (as well as no hardware accel dropdowns in Device Options)

With it unticked, the UHD770 did all the work, timing out at 3min 51sec for a 9:50 HEVC movie I'm working on.

With it ticked, the 3060Ti was used as well, but the time blew out to 4min 36sec, a 17% slowdown. I found that interesting because I would have thought that pressing the faster dedicated GPU into action would speed things up.

In MMS 2023, with the 3060Ti selected all the hardware accel dropdowns (and I note that that tickbox has been removed) the time was 1min 54sec, a whopping 200% faster export time compared to the fastest time in MEP 2021.

In MMS 2023 with the UHD 770 selected, the export took 3min 58sec, very similar to the MEP 2021 time.

So Magix' claim that things are happening much faster with the "Infusion engine" in MMS 2023 with an discrete GPU stands up to scrutiny.

And the message for those of you sitting on the fence about whether to upgrade, either your program or computer, the performance kick is significant with MMS 2023 if you have a good GPU.

In the week I was waiting for the 3060Ti GPU (so I was only using the iGPU UHD 770 in the CPU), the thought did cross my mind that I didn't really need the 3060 at all; my new system scored around 9 minutes on CubeAce's benchmark test, which wasn't bad. So it would be feasible to use a computer with only a fast, recent CPU with integrated iGPU.

Comments

Former user wrote on 5/22/2023, 8:07 AM

@AAProds 😁 You sound a bit happier 👍 I've said it before, MMS uses my GPU really well so if you're buying a new PC get a good GPU 👍

me_again wrote on 5/22/2023, 8:27 AM

@AAProds

... the thought did cross my mind that I didn't really need the 3060 at all

Tell you what Al, I'll give you a couple of quid for the 3060 to take it off your hands. I'll even pay for the postage 😇

AndyW

AAProds wrote on 5/22/2023, 8:43 AM

@me_again

Andy, the operative word being "did". 😉 Not any more! As @Former user said somewhere here, you can never have a fast-enough computer! 😁

Scenestealer wrote on 5/23/2023, 7:51 PM

@AAProds

".....you can never have a fast-enough computer!"

Geez Al - talk about a leopard changing its spots - not long ago there was nothing an aging Quad core couldn't do!😉

Good to see you are "onboard" at last.

A couple of points:-

"In MEP 2021 (and all previous versions) there was a tick box for "calculate video effects on GPU" (as well as no hardware accel dropdowns in Device Options)

With it unticked, the UHD770 did all the work, timing out at 3min 51sec for a 9:50 HEVC movie I'm working on.

Prior to the Infusion engine, this setting just invoked Direct X acceleration of various MEP Effects via the Direct 3D processors (shaders) on the discreet video card, and most of us with more powerful systems found that ticking the box made for slower renders, possibly caused by the slower pipeline of passing out data processing to the GPU which then had to return to the CPU / memory, compared to keeping it on the fast CPU (with or without Quick Sync). I believe only weaker CPU's without Quick Sync would have benefited from this setting.

AFAIK 2021 did not use NVENC (ie. separate hardware encoding via the chip on the Nvidia card) and the method used in the later MEP versions that keeps all the calculations within the discreet GPU and VRAM, so with you new system you would be in the situation described above, even in MEP2023 if you did not invoke the NVENC encoding by selecting the 3060Ti in the Display settings.

Peter

AAProds wrote on 5/23/2023, 8:13 PM

@Scenestealer

Thanks Peter, I never really understood all that stuff so that's interesting. 👍