Hows performance after patch 11 ?

[Morbad]

Have you manually tuned those Morbad?

Extensively.

The cheap Timetec CJR kit that I replaced with the Team B-die kit, with the settings I was using, was appreciably faster than the second kit's XMP timings.

I've just gotten some faster ram in an effort to help with Odyssey, might even be the same as yours.

What's the part number?

Sorry if I've missed it here but do you think resizeable bar with ryzen/radeon makes an appreciable difference in Odyssey vs ryzen/nvidia?

When I tested it, it did make a small positive difference on my AMD GPUs, at least if you can find appropriately GPU limited situations. However, I haven't tested ReBAR on vs. off enough with my NVIDIA GPUs that support it to get a good idea if it's meaningful on those platforms. I normally just leave it enabled as there are very few situations where it hurts performance at this point, and it's a useful GPU stability diagnostic tool itself. VRAM errors with ReBAR enabled will often crash the entire system; without it, such crashes are rare because the odds of any given segment of VRAM being directly mapped to the main virtual address space is low.

 

[sgurr]

Extensively.

The cheap Timetec CJR kit that I replaced with the Team B-die kit, with the settings I was using, was appreciably faster than the second kit's XMP timings.

That's some impressive knowledge.
I don't want to spend too much time on it but are there some simple things i could try beyond 1usmus or is it best left alone unless I'm willing to put the time in?

What's the part number?

tdppd416g3600hc16gbk

When I tested it, it did make a small positive difference on my AMD GPUs, at least if you can find appropriately GPU limited situations. However, I haven't tested ReBAR on vs. off enough with my NVIDIA GPUs that support it to get a good idea if it's meaningful on those platforms. I normally just leave it enabled as there are very few situations where it hurts performance at this point, and it's a useful GPU stability diagnostic tool itself. VRAM errors with ReBAR enabled will often crash the entire system; without it, such crashes are rare because the odds of any given segment of VRAM being directly mapped to the main virtual address space is low.

Good to know, thanks for that o7

 

[Morbad]

I don't want to spend too much time on it but are there some simple things i could try beyond 1usmus or is it best left alone unless I'm willing to put the time in?

There are some presets that tend to work well with Samsung B-die based modules that should let you dial in meaningfully better-than-stock settings in relatively short order. However, you'll still want to do some overnight testing to be sure they work.

tdppd416g3600hc16gbk

This seems to be a middle of the pack Samsung B-die bin, and probably has a fair bit of margin for tuning.

Platform matters a lot in terms of what settings are likely to optimal. What motherboard and CPU are you using?

 

[sgurr]

Platform matters a lot in terms of what settings are likely to optimal. What motherboard and CPU are you using?

3700x and asus strix x470-f

Not planning on changing the board when i do the cpu

 

[Morbad]

3700x and asus strix x470-f

Not planning on changing the board when i do the cpu

Board should be fine, though I'd recommend not going past firmware version 5861, unless you're sure you'll be getting a 5800X3D. All of AMD's AGESA versions from 1.2.0.4 through 1.2.0.6c have more issues than advantages; keep 1.2.0.3c until you actually have a 5800X3D ordered, then jump to (the yet unreleased) 1.2.0.7.

First step to figuring out what memory settings to use would be to find the maximum fully stable FCLK your CPU can handle. Once you know this you can decide on what memory clock to aim for.

 

[sgurr]

Board should be fine, though I'd recommend not going past firmware version 5861, unless you're sure you'll be getting a 5800X3D. All of AMD's AGESA versions from 1.2.0.4 through 1.2.0.6c have more issues than advantages; keep 1.2.0.3c until you actually have a 5800X3D ordered, then jump to (the yet unreleased) 1.2.0.7.

First step to figuring out what memory settings to use would be to find the maximum fully stable FCLK your CPU can handle. Once you know this you can decide on what memory clock to aim for.

Much appreciated Morbad.
It's only at 1600 just now but hoping i can up that, I'll see how it goes on the weekend. Cheers

 

[Postal]

That's because you are even more CPU limited.

I finally figured out what it is. Apparently oculus is 40/80 FPS depending on conditions apparently. Which is kind of dumb. But still. Settlements absolutely kill frames when I am in a ship or SRV.
In horizons though I am at a solid 80 FPS with HMD quality at 2.0 no matter what. So odyssey still performs like hot garbage in VR.

 

[Mogen Teras]

meanwhile in VR I get 80 FPS in space, 40ish in stations, and 25 to 40 flying over settlements with my 3080ti, ryzen 7 2700x, and 64gb of ddr4

weird thing is with oculus ASW off it doesnt matter what my HMD quality is set at. Could be .5 or 2.0. I am still locked on those frame rates.

This just highlights the CPU optimization issues. Your GPU is clearly overkill for ANY game right now, but a 2700x is going to be cutting it pretty close for most modern games. I'm not surprised your framerates tank so hard in a sloppy game like Odyssey.

 

[Morbad]

Some settlements have inexplicably good performance (playing at 4k with maximum quality FSR as I'm not GPU limited anyway):

Source: https://www.youtube.com/watch?v=KnffgIpFUX8


Will need to compare that with other examples of medium-sized settlements to see where the main differences are. Illumination being a factor is a distinct possibility.

Anyway on the encoder tangent, I've been comparing video capture performance as well and I think I've sold myself on a 5950X over the 5800X3D. It turns out that b-frames aren't working in AMD's AMF encoder yet, it's just that AMD's HEVC encoder is so terrible that it produces virtually the same bitrate as their AVC encoder at the same quality. This in turn isn't evidence of the AVC encoder being any good either, as I compared that against software x264 (AVC) and x264's veryfast preset produces more than five times the compression at the same video quality. This is even more ridiculous of a differential than I would have guessed and more than justifies getting the extra cores.

Edit: Tested NVENC (which I know has actual b-frame support) on my RTX 3080 as well, it's significantly better than AMD's VCN/AMF, but is still producing four times the bit rate of x264 veryfast. I guess AMD isn't as terrible in comparison to NVIDIA as I feared (and should be equalized one b-frames actually start working), it's just that at high resolutions and frames rates x264 pulls much further ahead than I was expecting, or rather the hardware encoders choke much harder.

 

[Frop]

Some settlements have inexplicably good performance (playing at 4k with maximum quality FSR as I'm not GPU limited anyway):

There are a few small and medium layouts (seeing that some medium layouts are actually smaller than the small layouts) that have significantly better performance, even for me. There is also one single layout where I can get 50-60 FPS in a CZ, that's the weirdest one (usually 30-40). I'll let you know which configuration that is.

 

[Frop]

Some settlements have inexplicably good performance (playing at 4k with maximum quality FSR as I'm not GPU limited anyway)

The layout in question is the small Industrial settlement with a HAB connected to the PWR-building and a single IND-building. Again, I have a hunch that certain (animated) assets like the spin cycle apparatus in for instance High Tech settlements drag down performance, regardless of settlement size.

 

[CMDR The Riz]

Benchmarks are one thing, but to get an idea if the system is able to handle sustained loads without throwing errors, one should run some stress tests able to max out load for protracted periods of time, and are able to flag errors that might otherwise go unnoticed.

I'd recommend TestMem5 using the 1usmus_v3 profile for system memory testing and y-cruncher (loops of the HNT test, specifically) for both memory and CPU testing.

GPU testing is a bit more tricky as few applications report errors short of those serious enough to produce visual artifacting or outright crashes. MSI Kombustor and OCCT have artifact checkers, but their loads aren't equivalent to most gaming tests and it's hard to find settings that won't just be throttled to uselessness by power limiters...they generally won't produce false positives, but they can miss some fairly obvious errors. That said, running OCCT's GPU memory test simultaneously with y-cruncher (limiting the number of processors to use on the latter to your number of logical cores, minus 2 and keeping memory utilization to about 75% of your total) should reveal any blatant stability issues very quickly.

Amazing info - thanks! I've downloaded the tools and will give them a go. Funnily enough, I was using MSI's Kombustor for testing

Performance isn't bad, all things considered, but it is curious to see that shutting down the settlement did nothing at all to improve frame rate. What timestamp on the video corresponds precisely to that dip in GPU utilization?

Judging by the timestamps, it looks like I had FrameView running about 2 minutes less than the video, and after a bit of comparison that GPU utilization dip should correspond roughly to around the 24-25 minute mark when I changed from my Dominator suit to my Maverick suit at the mining Settlement. At that point I had eliminated all of the NPCs there. I had MSI's Afterburner and FrameView running last night while I was running some more missions and noticed that the GPU utilization would typically drop and CPU utilization would increase when I hopped into my ship to change suits.

There are a few small and medium layouts (seeing that some medium layouts are actually smaller than the small layouts) that have significantly better performance, even for me. There is also one single layout where I can get 50-60 FPS in a CZ, that's the weirdest one (usually 30-40). I'll let you know which configuration that is.

The layout in question is the small Industrial settlement with a HAB connected to the PWR-building and a single IND-building. Again, I have a hunch that certain (animated) assets like the spin cycle apparatus in for instance High Tech settlements drag down performance, regardless of settlement size.

Likewise, I've found those small Industrial Settlements to be some of the best for performance. The small Agricultural Settlements (the ones with one HAB, one small Power Center and a single Agri Building) also perform abnormally well for me.

 

[Morbad]

All the small settlements give me pretty solid performance, and most of the large ones are borderline.

It's the medium settlements (one "+") that have the real spread, for me.

Amazing info - thanks! I've downloaded the tools and will give them a go. Funnily enough, I was using MSI's Kombustor for testing

You'll probably also want a quick bootable test if you're messing with memory. Using something like Memtest86 tests #7 & #8 before booting to Windows can save a lot of trouble.

 

[Frop]

It's the medium settlements (one "+") that have the real spread, for me.

Yep, another "good" layout for me is the "medium" Agriculture settlement, which has a similar HAB/PWR-building and one STO-building (none of the circular research buildings with little biodomes).

 

[Postal]

That's because you are even more CPU limited.

I know my CPU isnt great but that isnt the reason at all. The game is still terribly optimized for VR. I can literally run every other VR game I have along with horizons just fine. If I run elite in non vr mode it runs just fine and get decent frames.

 

[Morbad]

I know my CPU isnt great but that isnt the reason at all.

A faster CPU would improve the frame rates you see in VR.

The game is still terribly optimized for VR.

Yes.

 

[Mogen Teras]

I know my CPU isnt great but that isnt the reason at all. The game is still terribly optimized for VR. I can literally run every other VR game I have along with horizons just fine. If I run elite in non vr mode it runs just fine and get decent frames.

He isn't saying your CPU is bad. He's saying that having a very good GPU in relation to a semi-old CPU is going to put objectively more load on the CPU, since the GPU can easily chew through all the draw calls the CPU is sending it and will spend a lot of time simply waiting for more draw calls from the CPU. And since Odyssey seems to have posh-all for CPU optimization, the hardware disparity is going to become much worse.

 

[ebbrell]

Frustrating update 11 is worse than 10 for me. Only a few frames down while in a station sitting in a ship... but enough for me to notice.

I can sort orlf understand why settlements are struggling smoke dust lighting etc.

But something which was in game from way back in like sitting in a station in a ship.... horizons performed while odyssey doesn't now.

 

[Unknown Zombie]

I just played for the first time since December and oh man, the LOD and texture pop-in is so bad. It's gross.

Even in the main menu, as simple of a scene as it is, you don't get to see your ship's highest resolution skin texture. Why???

 

[gorankar]

3090, 5900x, 64Gb of ram. As of today, solid 60fps everywhere but the settlements and occasionally station interiors, where I get 25-45. So, it's fine, as long as I avoid settlements, especially ones with active CZ, or other players present. Even then, it's playable, just not optimal.