Benchmark Result Discussion

[Pepito123]

Full test with a simple PBO adjustment.

Before:

After:

The difference is minimal and where the improvement stands the most in the Baseline 3 test.

 

[mostvanvege]

Full test with a simple PBO adjustment.

Before:

View attachment 303496

After:

View attachment 303497

The difference is minimal and where the improvement stands the most in the Baseline 3 test.

How much power for gpu if I may ask? My 6800XT does the exact same numbers, I am curious the generational uplift means something in efficiency at least. For mine it takes somewhere 230W for this.

 

[T1 gumayusi]

7500F+4070Ti

 

[keiseyku]

View attachment 76202
Got curious after I posted the VR one and saw everyones desktop benchmarks. Actually confused how non-vr is worse lol. Interesting......
View attachment 76207

I'm still curious how you got above 140 fps when I regularly see 4090's stuck at 70 fps in VR. Those are crazy numbers-- and it wasn't even in OpenVR mode.

 

[Pepito123]

How much power for gpu if I may ask? My 6800XT does the exact same numbers, I am curious the generational uplift means something in efficiency at least. For mine it takes somewhere 230W for this.

270W

It's amazing that you get the same results as I did in a previous generation.

What is your configuration? Can you share your benchmark?

 

[chataignesy]

enabling frames generation

Could you explain how the frames generation work?

 

[mostvanvege]

270W

It's amazing that you get the same results as I did in a previous generation.

What is your configuration? Can you share your benchmark?

It is here. Did some tweaking on cpu and ram in bios to get some more juice, but to be honest it isnt night and day with the best possible tuning. Just get your ram timing on order.. VGA consumption tops at 250W.
1080p:

1080p with tesselation set to 8x in adrenalin:

1440p:

1080p with tesselation 8x and 5800X3D:

I switched the 6 core 5600x to an 5800X3D couple months ago and while I have seen results on baseline 3 with below 3 ms I could not achieve that even with vanilla vam. Perhaps my aged win10 is the problem. Anyway I think it was a handy upgrade because of the am4 socket. The v-cache is definitely helping vam.

 

[Jimmichoochoo]

Needless to say, I need to upgrade some parts..

Would there be any sense in buying something like a RTX 3060, or will even that be bottlenecked by my CPU? If I can get the Simpler Physics scene up to an average of 30 fps then that would probably be fine for me.

My PC is 96% Office work and the last 4% occasional gaming. VAM is quite interesting to play around with, so it would be nice to run scenes without having everything at Ultra Low, and without having to spend much more than $300.

 

[starmoon]

Needless to say, I need to upgrade some parts..

Would there be any sense in buying something like a RTX 3060, or will even that be bottlenecked by my CPU? If I can get the Simpler Physics scene up to an average of 30 fps then that would probably be fine for me.

My PC is 96% Office work and the last 4% occasional gaming. VAM is quite interesting to play around with, so it would be nice to run scenes without having everything at Ultra Low, and without having to spend much more than $300.

A 3060 should do 30 fps at 1440 simpler. I'm getting 55 fps at 1440 60hz simpler.
When I had a ryzen 2600 with 3060ti base3 fps were 36 physics 20ms @1080, at 300ms that is suffering from something.

 

[Jimmichoochoo]

Needless to say, I need to upgrade some parts..

Would there be any sense in buying something like a RTX 3060, or will even that be bottlenecked by my CPU? If I can get the Simpler Physics scene up to an average of 30 fps then that would probably be fine for me.

My PC is 96% Office work and the last 4% occasional gaming. VAM is quite interesting to play around with, so it would be nice to run scenes without having everything at Ultra Low, and without having to spend much more than $300.

View attachment 309297

After upgrading to a new RTX 3060, all other components stayed the same. For Baseline 3, performance is 10x better.

I didn't run a full benchmark, but for other scenes frames more or less doubled, sometimes tripled. I'm satisfied.

 

[Solmak]

Thanks for detailed explanation
I read articles that vram and Memory Bus Width is important for VR gaming and 4K,
but on the other hand some articles saying radeon is not suitable for vr gaming because of optimization in VR?

my concern is performance especially in VR mode ,sorry for my insufficient word
4080 is better and 4090 would be best one ,i know well, but i have 2K 144hz monitor and 4070 ti is ideal
so 4080 is excessive for non-vr gaming even like cyberpunk or hogwarts legacy
thats why im reluctant to buy 4080

Considering balance between performance and cost ,
i think its better to buy 4070ti and in one or two yeas to sell it and buy 50xx series

othewise should i wait for 4070ti super? it is just a rumor

I know I am super, super late to this, but.

Do you need CUDA cores? I don't mean for RTX Raytracing, you can do that with Reshade if you really want to but frankly it's kinda overblown right now. I mean do you need CUDA cores for rendering, for machine learning, for anything that REALLY needs them?

If you don't, get an AMD GPU. If you do, you already know you don't really have a choice.

 

[Jimmichoochoo]

Recently having upgraded from a GTX 970 to an RTX 3060, I was initially happy with the result, but playing around in VR, it did lack in performance unless scenes were kept very simple. It blasts through everything in Low Quality setting, but what's the fun in that.

For the fun of it, here's a benchmark with the Quest 2 and my old 2700x processor. I'm getting a 5700x on monday and then I'll redo the benchmark. While I don't expect much better performance, a 30% increase would help make simple scenes a bit more fluent, as they're usually hanging around 35 fps in VR.

Hopefully someone out there will appreciate these budget upgrade results.

 

[wanderer]

Here's my oculus diag settings

 

[Jimmichoochoo]

Christmas came early, the 5700X arrived a couple days earlier.

Result-wise, jumping from a 2700X to a 5700X gave me a 50% increase in 1440p.

But why have my FPS dropped in VR mode? That makes no sense? My 2700X ran 20fps in VR but the 5700X just at 18fps???

VR Results:

 

[turtlebackgoofy]

Did some profiling, reverse engineering and patching with dnspyex on the VAM engine and got a FREE 13% performance increase in the physics engine (See Baseline 3). More physics objects in a scene yield better performance boosts using my patch.

All I did was add batching and multithreading in code hotspots in GPUCollidersManager::ComputeColliders. No artifacts or bugs or any other side-effects possible. Just simple first grader optimization of engine code.

The profiler also showed me that the physics simulation is 1% waiting for GPU to finish collision simulation, 9% CPU calculations and 90% plain CPU time waste in managed wrapper calls to the mono runtime. This explains why Ryzens with X3D are so fast in VAM, their abnormous cache eats all those useless wrapper calls.

With some further optimization a 100% increase in physics speed in feasible. Outloading the physics calculations to a native library and calling the UnityPlayer.dll directly could feasible increase physics speed by 1000%.

Overall Unity and VAM code are a shitshow.

Original Assembly-CSharp.dll

Patched Assembly-CSharp.dll

 

[nomannoname]

So, I ran a bunch of desktop and VR benchmarks today on my R5 5600x, RTX 3060ti, 16gb ram at 3200mhz, running from nvme ssd, apart from enabling XMP nothing is overclocked and everything is stock.
The desktop resolutions came out a bit off in the results, might've had something to do with windowed mode I'm not sure, but I assume it would be too small of a difference to have any visible impact..
For the headset I used a Quest 3, both via wired oculus link, and wireless virtual desktop, when it says "Oculus Rift CV1" that's just the Oculus pc app reporting it as a CV1, because meta is lazy I guess.

Desktop: 1920x1080
107.08 fps avg

Desktop: 2560x1440
82.32 fps avg

Desktop: 3840x2160 ("4k")
51.84 fps avg

Oculus Link wired cable, oculus runtime: 3744x2016 (1X, 90hz)
38.51 fps avg

Oculus Link wired cable, oculus runtime: 4128x2240 (1.15X, 72hz) I threw this one in because I saw some people using 72hz mode
37.13 fps avg

Oculus Link wired cable, oculus runtime: 4512x2448 (1.2X, 90hz)
28.44 fps avg

Wireless VR over Virtual Desktop, Oculus Runtime, 160mbps, HEVC, SSW disabled: Low settings (3456x1824, 90hz)
48.09 fps avg

Wireless VR over Virtual Desktop, Oculus Runtime, 160mbps, HEVC, SSW disabled: Medium settings (4032x2112, 90hz)
40.76 fps avg

Wireless VR over Virtual Desktop, Oculus Runtime, 160mbps, HEVC, SSW disabled: High settings (4992x2592, 90hz)
32.67 fps avg

Another benchmark that I chose not to include was that you need on average 2 barfbuckets per vr benchmark

 

[donkeykong69]

5900X PBO & 3060 Ti 104% power limit +80 Core +1300 Mem

With 1 CCD via Bios. Pretty nice fps boost.

 

[nomannoname]

5900X & 3060 Ti ...

Interesting, so I'm getting a slightly higher avg fps on my stock 3060 ti than your stock 3060 ti, in spite of you having a better cpu and double the amount of ram.. I mean the fps is negligible but still.. I am really curious what your results would be on 4k and/or vr. Either way this probably indicates that at 1080p the 3060 ti is being maxed out, and probably the bottlenecking factor

Edit: So I ran it again in full screen mode to compare, and I figured I would run hwmonitor and msi afterburner to see if there would be a difference. It looks to me that the 3060ti is the bottleneck in VAM even for a 5600x, even on 1080p.. Which might be handy to know since I was considering upgrading.
Quick question, you have a 5900X, isn't CCDs boost only for the X3D?

1080p stock settings

1080p overclocked, 103% power limit, +100 core clock, +1000 memory clock

4k stock

4k overclocked, 103% power limit, +100 core clock, +1000 memory clock

 

[donkeykong69]

Interesting, so I'm getting a slightly higher avg fps on my stock 3060 ti than your stock 3060 ti, in spite of you having a better cpu and double the amount of ram.. I mean the fps is negligible but still.. I am really curious what your results would be on 4k and/or vr. Either way this probably indicates that at 1080p the 3060 ti is being maxed out, and probably the bottlenecking factor

Edit: So I ran it again in full screen mode to compare, and I figured I would run hwmonitor and msi afterburner to see if there would be a difference. It looks to me that the 3060ti is the bottleneck in VAM even for a 5600x, even on 1080p.. Which might be handy to know since I was considering upgrading.
Quick question, you have a 5900X, isn't CCDs boost only for the X3D?

Hey. I'm actually running my 3060 Ti overclocked at 104% power limit +80 Core +1300 Mem. My 5900X also has PBO and is undervolted. Though I do notice VAM crashes sometimes so maybe I have to dial back my GPU OC.

I disabled a CCD because I wanted to make sure that VAM runs on the best cores. Plus, I wanted to reduce any chance of inter-CCD latency. That's what I heard anyway about old games. I'm sure newer games make better use of multi-core processing so there would be no reason to disable a CCD or set core affinity.

 

[nomannoname]

Hey. I'm actually running my 3060 Ti overclocked at 104% power limit +80 Core +1300 Mem. My 5900X also has PBO and is undervolted. Though I do notice VAM crashes sometimes so maybe I have to dial back my GPU OC.

I disabled a CCD because I wanted to make sure that VAM runs on the best cores. Plus, I wanted to reduce any chance of inter-CCD latency. That's what I heard anyway about old games. I'm sure newer games make better use of multi-core processing so there would be no reason to disable a CCD or set core affinity.

I noticed myself that going over 1000 mem clock crashed vam for me also, 1000 was the max for me to make it through the benchmark without crashing.
but I didn't know that was possible on the 5900x, I thought that was reserved for the newer x3d models, in that case I should have a look at that, maybe it's possible on mine also then.
The weird thing I found though was that for me it seemed like the cpu was not bottlenecking the gpu, and the 3060ti was the limiting factor, but you clearly got a big uplift by cpu settings alone right?

 

[donkeykong69]

I noticed myself that going over 1000 mem clock crashed vam for me also, 1000 was the max for me to make it through the benchmark without crashing.
but I didn't know that was possible on the 5900x, I thought that was reserved for the newer x3d models, in that case I should have a look at that, maybe it's possible on mine also then.
The weird thing I found though was that for me it seemed like the cpu was not bottlenecking the gpu, and the 3060ti was the limiting factor, but you clearly got a big uplift by cpu settings alone right?

5600X has a single die so you can't disable another one. 5900X is a pretty bad cpu is for VAM since the game relies more on single core performance. Disabling a CCD effectively makes my cpu a higher clocked 5600X.

I believe at 1080P, the biggest bottleneck is the CPU since I could squeeze more fps with the 3060 Ti. I could see that my physics time improved by a lot by optimizing my CPU.

 

[jpegxl]

Base System

3600X -> 5800X3D

2060 SUPER -> 4070 SUPER

 

[Jimmichoochoo]

@LsRp Very interesting results.

To directly compare with my 5700X and 3060, against your 5800X3D and 2060S, the 1080p results are nearly identical for all scenes. The 4070 does a massive difference, wow! I do wonder if I shouldnt' have spent a few hundreds more on a better GFX than the 3060 but oh well now too late.

 

[mokumoku12]

5800X3D
RTX 4080
I set resolution FHD but result was 2560 x 1417 and FHD fullscreen tested

 

[w21403]

在 VAM 引擎上使用 dnspyex 进行了一些分析、逆向工程和修补，并在物理引擎中获得了 13% 的免费性能提升（参见基线 3）。使用我的补丁，场景中更多的物理对象会产生更好的性能提升。

我所做的只是在 GPUCollidersManager::ComputeColliders 的代码热点中添加批处理和多线程。不可能出现任何伪影、错误或任何其他副作用。只是简单的一年级学生对引擎代码的优化。

分析器还向我显示，物理模拟有 1% 等待 GPU 完成碰撞模拟，9% 是 CPU 计算，90% 是在对单运行时的托管包装器调用中浪费了普通 CPU 时间。这解释了为什么配备 X3D 的 Ryzen 在 VAM 中速度如此之快，其异常的缓存会吞噬所有那些无用的包装器调用。

通过进一步优化，物理速度提高 100% 是可行的。将物理计算导出到本机库并直接调用 UnityPlayer.dll 可以将物理速度提高 1000%。

总体而言，Unity 和 VAM 代码简直就是一坨屎。

原始程序集-CSharp.dll
View attachment 319356
修补程序集-CSharp.dll
View attachment 319357

Can you please release your patch for fixing? Thank you.