8 minutes ago, hanhamreds said:

Glad I waited

 

 

https://www.tweaktown.com/news/75679/nvidia-will-shift-over-to-tsmc-for-new-7nm-ampere-gpus-in-2021/index.html?utm_source=dlvr.it&utm_medium=facebook

I saw a report that suggested this might just be a switch to 7nm for laptop parts.  It may not mean a 7nm replacement for high-end.  It could, but no guarantees.  The performance of AMDs next offering may dictate what Nvidia does more than anything.

Edited October 18, 20205 yr by Virtual-Chris

26 minutes ago, captain420 said:

What filter are you referring to? I will give it a try. I have lowered my rendering scale back to 100% because I get better performance this way.

Here is a guide.

8 minutes ago, hanhamreds said:

https://www.tweaktown.com/news/75679/nvidia-will-shift-over-to-tsmc-for-new-7nm-ampere-gpus-in-2021/index.html?utm_source=dlvr.it&utm_medium=facebook

Seems to be a false rumour. Reputable leakers do not know anything about that, and the source is DigiTimes who also leaked that Zen 3 will be on 5nm and coming in 2021.

Even if it were true, the result would likely be underwhelming without something to reduce the memory's power draw, and it would probably still be no match for 5nm RDNA 3 coming in about a year from now. It would be in their best of interests to focus on Hopper from now on.

14 minutes ago, Virtual-Chris said:

I saw a report that suggested this might just be a switch to 7nm for laptop parts.  It may not mean a 7nm replacement for high-end.  It could, but no guarantees.  The performance of AMDs next offering may dictate what Nvidia does more than anything.

The issue with Samsung 8 is low yield and to some extent thermals.

Without doing any serious research, my uninformed hunch is the TSMC 7nm will NOT instantly give better performance but you WILL  get less heat and/or cheaper products if they go that way. 

Edited October 18, 20205 yr by Glenn Fitzpatrick

19 hours ago, Glenn Fitzpatrick said:

 

I am aware you can SLI the 3090 but what non-server motherboard lets you physically fit three triple slot GPU cards ?

They will all be on its on board. 

Hi Rob,

I do apologies for not being clear and not totally understanding.

If I follow this topic, its saying as you say that +200 render scale does make MSFS have SSAA at 8xSSAA but its also been pointed out it scales you resolution to 8k also.

I was referring to the test you did showing MSFS can use more VRAM than P3D, i meant to say that in that test you would of had to used NCP and set DRS to make P3D run also at 8k.

Therefore its not simply in MSFS 8xSSAA but also upscaling to 8K as ChaoticBeauty has pointed out said. SSAA may work with any resolution but as he says within MSFS also makes the sim run at a resolution of 8k.

 

 

Edited October 18, 20205 yr by Nyxx

On 10/17/2020 at 4:24 AM, Rob_Ainscough said:

I don't believe MSFS support VRR like P3D does

What do you mean by that? MSFS supports G-sync for sure. Unless you mean something else.

6 hours ago, Rob_Ainscough said:

P3D 8X SSAA

That seems like a huge waste of power consumption for the GPU with minimal and even degraded visual quality with the extreme down sampling needed to normal monitor resolutions. 

Right now I'm more interested in about what AND will bring to the table.

15 hours ago, Rob_Ainscough said:

The testing I did was with MSFS (not with P3D) using UserCfg.opt for LoDRadius values (both = 9.0), and render scaling 100 (no SSAA) and another test with render scaling at 200 (4X SSAA) in and around the KLAX area.  If the 3090 can handle MSFS Render Scaling of 200 (equivalent to my current P3D graphics settings) at 30 FPS in areas like KLAX, then that's certainly better than what my Titan RTX can do. 

I tested with LOD 4 around LA downtown. With scale 120 I got around 25-32fps, VRAM around 9GB used in 3090. At scale 200 the fps halved to 15 around, and VRAM around 13GB. All other settings ultra.

At LOD 5 my CPU (10850K @5.1GHz all cores) gets really hammered and MSFS becomes totally CPU limited at rendering scale 120. I can't even imagine how possibly LOD 9 is going to even work on my rig. Maybe I should give it a try just out of curiosity.

Edited October 18, 20205 yr by FlyIce

Why use render scale > 100?

24 minutes ago, FlyIce said:

I tested with LOD 4 around LA downtown. With scale 120 I got around 25-32fps, VRAM around 9GB used in 3090. At scale 200 the fps halved to 15 around, and VRAM around 13GB. All other settings ultra.

At LOD 5 my CPU (10850K @5.1GHz all cores) gets really hammered and MSFS becomes totally CPU limited at rendering scale 120. I can't even imagine how possibly LOD 9 is going to even work on my rig. Maybe I should give it a try just out of curiosity.

In a double-blinded test could you discern the difference between render scale of 100 versus 120?  I'm not seeing anything myself when I've done short tests, at 3440x1440.

Also, given what you're seeing now, what do you think the minimum VRAM will be for MSFS over the next couple of years?  I know the DX12 implementation should affect that but even so, what's enough?  The 9Gb around downtown LA w/ LOD of 4 seems to suggest maybe 16Gb will be ample for most purposes, at 3440x1440? 

Honestly I'm not sure whether the perceived differences between scale 100 and 120 are real or not. It APPEARS to be sharper, that's all I can say. It's like that I'm becoming more CPU limited with 3090 so what the heck just throw a bit more work to 3090 so I can feel a bit better about the money I irrationally and shamefully overspent...

As far as I have seen, with LOD 4, I haven't seen VRAM usage over 10GB when rendering scale set to 100 - 120, even at mega cities like LA. Rendering scale 200 + LOD 9 probably is going to give 24GB a run for its money, but I feel that we'll long be bottlenecked by CPU and possibly MSFS server bandwidth before breaking the 24GB. No idea about what DX12 will bring to the table though. 

Now, here is the strange thing, I tested MSFS at a remote Japan island Shimojishima (RORS) and you'd bet my fps (unlimited) would be out of roof. Nope, with exactly same setting as I'm flying over LA, I'm still around 35fps over these tiny island in nowhere. Have no idea why. 

 

Edited October 18, 20205 yr by FlyIce

OK, just tested LOD 9 in around downtown LA. It's actually not too bad.

LOD 9 + rendering scale 100: 20-24fps, 24GB RAM, 9.3GB VRAM

LOD 9 + rendering scale 200: 15fps, 24GB RAM, 14.3 GB VRAM

All settings ultra. Clear sky, JMB spawn directly above downtown LA. MSFS is mostly CPU limited at LOD 9.

[email protected], 32GB RAM (3600), 3090 FE.

I would add that visually LOD 9 improvement over LOD 4 is kind of subtle. Not worth the loss of fps.

Edited October 18, 20205 yr by FlyIce

I get 18GB VRAM usage at EGLC in my settings and yes I can tell a difference between render scaling at 100 and 120. There are less jaggies, things are smoother. But for me, it's not worth it. I'll stick to 100 and keep my frames.

13 minutes ago, captain420 said:

I get 18GB VRAM usage at EGLC in my settings and yes I can tell a difference between render scaling at 100 and 120. There are less jaggies, things are smoother. But for me, it's not worth it. I'll stick to 100 and keep my frames.

You can up scale render scaling then turn down your AA settings - it works great - nice tradeoff - its like using the render scaling to handle you AA - try it it works

31 minutes ago, Richard Sennett said:

You can up scale render scaling then turn down your AA settings - it works great - nice tradeoff - its like using the render scaling to handle you AA - try it it works

Are you referring to Antisoptric Filtering and Texture Supersampling? Mine's currently set to the max for both.