Benchmark Result Discussion

[TToby]

@TToby
Don't come back until you got nearly the same results Freeze it!

Sorry for coming back without having nailed it down.
With 4 days of testing, I found some results for myself:
- AMD BIOS settings are much more difficult than Intel. No "click and ready". Saying this, my last CPU was a 4790k, things might have evolved since that days.
- As we all had mentioned long before, for VaM the single core speed is really important, not the multicore speed.
- So for overclocking AMD CPUs, you want to use Precision Boost Overdrive instead of setting a common muliplier like many OC tools do.
- Quality setting in the GPU driver tool will have an big impact. Use them wisely.
- For VaM it is not only pure speed, there is something else I still doesn't have found...
EDIT: it was only an automatically lowered custom tesselation GPU driver setting!

In the meantime, I have slightly OCed my PC to some very good values:
I have a 19000 score in Timespy and 650/6500 in CPU-z.
My 5800x is clocking up to > 4700MHz, my 6900xt is on 2500 MHz, VRAM on +140 MHz...
My custom loop watercooling is running well and keeps everything well below 70 C.
But I still got roughly the same values like in my last picture.

The strange thing is, 24h ago, I had found some settings giving me roughly the same results like Endolu had!!!!
But then I idiotically changed some BIOS settings I can't remember and it was again back to "normal".
In the meantime I have watched several youtube videos of how to set up an AMD cpu and I know more about OC than I would ever have like to... but still, there is some mysterious setting that is obviously important for VaM, but I simply can't find it.
EDIT: It was only a wrongly set custom tesselation setting in the GPU driver tool that was giving me those extra fps! I unknowingly switched it back to the correct "use application settings", and it was giving me the shock of aprox a minus of 30 fps in VaM on the next try.

P.S.: This is NO "my cock is bigger than yours" posting. Till some months ago, I have played VaM on a very old PC, too. It was OK for me. Then I had invested much more money for an all-new rig than I had originally planned.
Now I have realized that something stupid is pulling the brake for more than 30 fps, and it is driving me insane.

 

[keycode]

I was loading the benchmark in desktop mode with the intention to check if my system could match MacGruber results: so I had to reduce my display res. to 1920x1080 in a second session. My gpu is the first release of 3070 (old VRAM modules, *not* X) and I prefere to leave it at standard clock as well as my CPU, a normal i7 10700 that I was setting with a "quiet" max clock 4700 Ghz... this is the max speed I can get if I set my cpu bios option to the mentioned value(so I penalize/waste 100Ghz less for the max clock... but with no real difference in performance and with a lot more favorable max W. absorption or max temp.).

 

[keycode]

this is the benchmark at native display res.

 

[TToby]

OMG
This is embarrassing. It was so simple, it literally hurts. For me it was simply the tesselation!
It "only" took me 4,5 days to find it out.
I am still not completely sure about that, cause I tried a lot of BIOS settings in the same time, but it seems pretty obvious:

If you frenetically click around in the Adrenalin AMD Driver settings, and click on the "e-sports" preset,
the driver will automatically switch to a manual tesselation setting with the standard value of 8x. It is well hidden behind that "advanced options" button.

VaM obviously is using a tesselation of ca 16x, wich will be applied if you change the tesselation settings to "use application settings". Mind that you have to restart VaM to see the difference.
With VaM only have to render half of the hair, it is no wonder why I've got some pretty impressing benchmark values. Those values exactly matches my previusly measured dream-benchmark results.

Tesselation is the setting that can dramatically change your performance, much more than overclocking your hardware or something like that. For my PC, switching between 8x and 16x is worth more than 30 fps in average, without seeing a big difference in that darkish lighting.

After all I have learned: Hair is one of the biggest performance killer in VaM. That was pretty obvious, because we all know that lowering the hair physics values has a visible effect on performance.
But I never have thought that it would be this much!

 

[MacGruber]

@HolySchmidt It was the driver quality settings like tesselation, antialising, etc.

Some random @TToby quote from Friday

 

[TToby]

Some random @TToby quote from Friday

Waaahh... this is so embrassing.
Whereas some of those quality settings gave me a big MINUS of fps on Friday,
the unknowingly changed tesselation settings gave me a big PLUS on Saturday.
If something is better than before, I first of all don't think about having done something wrong.

 

[keycode]

Honestly, after those modest (poor) results with the native res. of my monitor, I was expecting tragic results with reverb and all those crapware setting lists in steamVR.

 

[keycode]

Waaahh... this is so embrassing.
Whereas some of those quality settings gave me a big MINUS of fps on Friday,
the unknowingly changed tesselation settings gave me a big PLUS on Saturday.
If something is better than before, I first of all don't think about having done something wrong.

Ok! You merit a heretic report from vam inquisitors team for not buying nvidia/intel hardware. And because I have no fucking tesselation option in nvidia control panel !! (sh..t!)

 

[keycode]

Waaahh... this is so embrassing.
Whereas some of those quality settings gave me a big MINUS of fps on Friday,
the unknowingly changed tesselation settings gave me a big PLUS on Saturday.
If something is better than before, I first of all don't think about having done something wrong.

question to the smart german old amd client: do you think my value for resolution scale (1x) is ok? 4376x2140 should be (more or less) the native res. of my reverb, isn't it? I was reading your tips about setting headset res. .... but when I must consider lowering the so called "internal resolution" for getting more FPS I can't ... I can't do it.

 

[TToby]

question to the smart german old amd client: do you think my value for resolution scale (1x) is ok? 4376x2140 should be (more or less) the native res. of my reverb, isn't it? I was reading your tips about setting headset res. .... but when I must consider lowering the so called "internal resolution" for getting more FPS I can't ... I can't do it.

Hi, don't fool yourself. Test it at the same resolution you are usually playing at. This is not about "who has the most fps" but it is for yourself to see if you can optimize your PC and to maybe compare your values with other Reverb users. I would suggest 100%, so it is more reasonable. There are so many things that have to be taken into account, like different resolution, different supersampling, open VR or Oculus, direct or streamed/compressed, aso. Maybe that old Unity engine isn't working good with AMD, or one CPU has a better singlecore speed, but lower multicore.... we might see. This is about having an approximate idea how well your rig is performing, if it is reasonable to buy a new GPU for only some more fps, to see how much supersampling will cost you or undervolting your CPU, or to even have an idea what is subjetively "fast enough" in the eyes of one user and "too slow" for another one.

 

[keycode]

Hi, don't fool yourself. Test it at the same resolution you are usually playing at. This is not about "who has the most fps" but it is for yourself to see if you can optimize your PC and to maybe compare your values with other Reverb users. I would suggest 100%, so it is more reasonable.

so... after comparing my desktop mode benchmark with the Gruber I can stop putting my fingers in something that goes beyond my knowledge, and of course my little kidding about AMD was nothing more than just kidding... even considering that processor equivalence with my intel cpu performance. Thanks for answering.

 

[Hedgepig]

Does adding a bigger memory card make any difference? I have no idea about such things, but, if so, this isn't like Stage 1/2/3 tuning an engine, in which the payoff is more power but a shorter engine life due to the higher stress on the engine, is it ? At stages 2 and 3 you need all kinds of mods to stop the engine exploding.

 

[keycode]

Does adding a bigger memory card make any difference? I have no idea about such things, but, if so, this isn't like Stage 1/2/3 tuning an engine, in which the payoff is more power but a shorter engine life due to the higher stress on the engine, is it ? At stages 2 and 3 you need all kinds of mods to stop the engine exploding.

the engine risks less (to implode) than a bank account

 

[WanderingWomble]

Does adding a bigger memory card make any difference? I have no idea about such things, but, if so, this isn't like Stage 1/2/3 tuning an engine, in which the payoff is more power but a shorter engine life due to the higher stress on the engine, is it ? At stages 2 and 3 you need all kinds of mods to stop the engine exploding.

OK, imagine if someone is working on your car, they have a workshop full of all the tools they could ever need, but they can only have a few out at a time. Your memory is how many tools they can have close to hand. More memory means more tools to quickly grab. If you run out of memory it means they needed one more tool than they could get close to hand, so they have to stop work, walk over to their tool draws and pull out what they need while putting away the tool they used the longest ago.

So, if you're low on memory it's like they are constantly running back and forth across their workshop messing about with which tools they have out. If a computer gets into that state the performance will be horrendous. Not just in game, but everything will be crawling.

So, open Task Manager (press Ctrl+Shift+Escape), click on the Performance tab, then see if your memory is right near max usage and if your drive is going crazy as that is what has to pick up the slack.

Your GPU also has its own memory and a similar issue, if the GPU memory is maxed out then it can't cope with the size of the textures and other tools it needs. You may have noticed in other games where messing with the graphics settings will tell you how much VRAM those settings will use. They are actually calculating how much memory the graphics card will need to use to run with those settings. I don't think VaM has that, but if you have a game that does do that then have a tinker to get a feel for how things change. It should help you get an idea on what tweaks you can make to get a game to fit on your graphics card

As for upgrade stages, it's a yes and no. Yes if you buy power hungry parts that your power supply can't feed then it's not gonna work. No it's not likely to break anything, just crash a lot. The danger comes from overclocking, think ramping up the boost on a turbo past the manufacturers rating for the engine. So that aside you can swap bits in and out as long as you take care that you have enough power for it and connecters. to plug it into. If you are running out of memory then that is a very safe upgrade to do, it's not going to drink a ton of power. It can hurt performance if the memory is all running at different speeds though, but that's peak performance, so it would be fine to daily drive like that but you might not get the fastest lap time if you don't take care to match the memory properly.

 

[HolySchmidt]

Thank you MacGruber for useful benchmark
I can't try my oculus right now but here is my result in desktop mode 1080p.
5800x is running in ecomode so it may add a few more fps I guess.

View attachment 71565

I just wanted to know, and for everyone who wants to compare:

Thats i7-9700K & 3080 VS. 5800X & 6900XT at same resolution, so can't compare better.
Result: 140FPS VS. 179FPS. (+27%)

I'm still interested in how your VR results look like

 

[keiseyku]

So this is kinda surprising. My primary computer (i5-10400 / RTX 3070) pulled some pretty decent numbers. I noticed that my ASW was still on and when I tried to run the test it was 20 fps max but after disabling ASW I got this (CPU load undocumented):

VR mode:

Desktop mode:

I'll try bringing my 3070 to my secondary computer to run a i7-10700k with it. Also did everyone have ASW/MS off this entire time and I was just out of the loop?

 

[Murasaki]

Here's what I got. Not much but the 1070 is still kicking and doing it's best! It's a real trooper. <3

I'm running Windows 11 [ version: 10.0.22000 Build 22000 ] and I have some overclock in place:

CPU: Running at 4625MHz 1.325V

GPU: +128MHz on the core and +300MHz for the memory.

 

[Hedgepig]

OK, imagine if someone is working on your car, they have a workshop full of all the tools they could ever need, but they can only have a few out at a time. Your memory is how many tools they can have close to hand. More memory means more tools to quickly grab. If you run out of memory it means they needed one more tool than they could get close to hand, so they have to stop work, walk over to their tool draws and pull out what they need while putting away the tool they used the longest ago.

So, if you're low on memory it's like they are constantly running back and forth across their workshop messing about with which tools they have out. If a computer gets into that state the performance will be horrendous. Not just in game, but everything will be crawling.

So, open Task Manager (press Ctrl+Shift+Escape), click on the Performance tab, then see if your memory is right near max usage and if your drive is going crazy as that is what has to pick up the slack.

Your GPU also has its own memory and a similar issue, if the GPU memory is maxed out then it can't cope with the size of the textures and other tools it needs. You may have noticed in other games where messing with the graphics settings will tell you how much VRAM those settings will use. They are actually calculating how much memory the graphics card will need to use to run with those settings. I don't think VaM has that, but if you have a game that does do that then have a tinker to get a feel for how things change. It should help you get an idea on what tweaks you can make to get a game to fit on your graphics card

As for upgrade stages, it's a yes and no. Yes if you buy power hungry parts that your power supply can't feed then it's not gonna work. No it's not likely to break anything, just crash a lot. The danger comes from overclocking, think ramping up the boost on a turbo past the manufacturers rating for the engine. So that aside you can swap bits in and out as long as you take care that you have enough power for it and connecters. to plug it into. If you are running out of memory then that is a very safe upgrade to do, it's not going to drink a ton of power. It can hurt performance if the memory is all running at different speeds though, but that's peak performance, so it would be fine to daily drive like that but you might not get the fastest lap time if you don't take care to match the memory properly.

Hmm, this sounds tricky, I image that the my 3060's memory card is suited to the task. I currently have 16 GB of RAM. Apparently you can ramp it up to 64.

Seller:

"The Computer has 4 Slots to install memory, already with 16GB standard memory installed. For best OMEN 25L GT12-0064 Computer performance use the maximum amount of 64GB, fill all the slots with the max allowed memory per slot for your Computer."


But that might just be the seller trying to sell memory cards? "Yeah man, so you just bolt this turbo onto your bike and set the boost at 50psi." Doesn't mention it will shred your standard back tire when you wring the throttle.


What's your opinion on the seller's claim?

 

[MacGruber]

I currently have 16 GB of RAM. Apparently you can ramp it up to 64.
(...)
What's your opinion on doing this?

At the moment, 16GB should be enough for VaM. If it doesn't crash, its enough. You could argue that if you switch between complex scenes all the time, maybe 32GB could give you a tiny bit. I would say games will not to make use of more than 32GB for quite a while. The only reason to have more memory is if you do some professional applications that actually make use of that. In times of PCIe 4.0 SSDs even a RAM drive doesn't make that much sense anymore. My PC in the office has got 128GB, but there I do heavy load stuff all the time, like running our game server plus multiple game clients on the same machine for testing purposes. Normal people won't ever need that.

However, what does make sense is to have a set of either 2 or 4 identical (= bought as ONE set!) RAM modules and putting them in the correct slots as indicated in your mainboard manual. That way the modules can run in Dual- or Quad-Channel mode. Note that even some high-end boards do not support Quad-Channel.

The other thing to think about is CAS Latency. These are the "CL" numbers. Common RAM modules are CL16 these days. In simple terms that means the CPU needs to wait 16 cycles before requested data is actually available. Hence a CL14 module is theoretically 16/14 = 14.3% faster. A module of DDR4-3200 with CL14 is still slightly faster than a DDR4-3600 CL16 module. Of course that does not directly translate into FPS in VaM or any other game, but still, CPUs waste a lot of time just waiting on data from RAM, so faster RAM gives you certainly a few noticable %. Even more important if you went for Intel CPUs which tend to have less L3 cache and therefore need faster RAM. General advice, if you spend big money on CPU and GPU, don't save money by buying slow RAM

 

[Hedgepig]

At the moment, 16GB should be enough for VaM. If it doesn't crash, its enough. You could argue that if you switch between complex scenes all the time, maybe 32GB could give you a tiny bit. I would say games will not to make use of more than 32GB for quite a while. The only reason to have more memory is if you do some professional applications that actually make use of that. In times of PCIe 4.0 SSDs even a RAM drive doesn't make that much sense anymore. My PC in the office has got 128GB, but there I do heavy load stuff all the time, like running our game server plus multiple game clients on the same machine for testing purposes. Normal people won't ever need that.

However, what does make sense is to have a set of either 2 or 4 identical (= bought as ONE set!) RAM modules and putting them in the correct slots as indicated in your mainboard manual. That way the modules can run in Dual- or Quad-Channel mode. Note that even some high-end boards do not support Quad-Channel.

The other thing to think about is CAS Latency. These are the "CL" numbers. Common RAM modules are CL16 these days. In simple terms that means the CPU needs to wait 16 cycles before requested data is actually available. Hence a CL14 module is theoretically 16/14 = 14.3% faster. A module of DDR4-3200 with CL14 is still slightly faster than a DDR4-3600 CL16 module. Of course that does not directly translate into FPS in VaM or any other game, but still, CPUs waste a lot of time just waiting on data from RAM, so faster RAM gives you certainly a few noticable %. Even more important if you went for Intel CPUs which tend to have less L3 cache and therefore need faster RAM. General advice, if you spend big money on CPU and GPU, don't save money by buying slow RAM

Thank you. I might buy 16GB and see what happens.

 

[TToby]

@Hedgepig It is exactly like MacGruber said. And he allready said all I wanted to answer to your question.
But I liked the car workshop example of WanderingWoumble:
64GB RAM is like driving a pickup while only needing space for one crate of beer per week. 128GB is like buying a truck for the same purpose. 32 GB is like a station wagon limousine, 16GB is like a normal car, 8GB is like a two seats SMART these days. Some games are greedy and wants to put as much beer in the luggage trunk as may fit in, while they are maybe only able to drink one crate per week. Lets say VaM is a bit overweening and drinks two crates of beer. But if you make some room in your normal car, those will still fit.

But to get back to the question:
You may want to take a look at the inside of your PC. The RAM slots are normally 4 slots next to the CPU, maybe partically covered by the big CPU cooler. They are rougly 8mm wide and 70mm long and pretty obvious. If you have 16GB installed, it may be 2 x 8GB or 1 x 16GB modules. 2 x 8GB would be a bit better, because your PC can use two of them at the same time, which will give you a bit more speed. But we are talking about maybe 10 fps more at max? Maybe it has not even a visible effect.
So If you have the money and there is only one module, you may buy a second 16 GB module of the same brand and kind (!). Don't mix different modules if possible. If there are already two and you never had any issues with VaM crashing sometimes, you will most likely not see any difference from buying two new modules (always should be in pairs).
Please mind that you can seriously destroy your PC by static electricity, if you are too careless (been there, done that after more than 30 years of experience).

EDIT additional warning: If you have to remove the CPU cooler for inserting the RAM modules, please mind that there is thermal grease between cooler and CPU. This grease is more important these days as one may think. It is dried in and if you put back the cooler without removing the old grease and applying a thin layer of new grease, you can get serious heat issues.

 

[Hedgepig]

@Hedgepig It is exactly like MacGruber said. And he allready said all I wanted to answer to your question.
But I liked the car workshop example of WanderingWoumble:
64GB RAM is like driving a pickup while only needing space for one crate of beer per week. 128GB is like buying a truck for the same purpose. 32 GB is like a station wagon limousine, 16GB is like a normal car, 8GB is like a two seats SMART these days. Some games are greedy and wants to put as much beer in the luggage trunk as may fit in, while they are maybe only able to drink one crate per week. Lets say VaM is a bit overweening and drinks two crates of beer. But if you make some room in your normal car, those will still fit.

That image of the SMART car made me laugh, thank you so much. I visualized 2 large guys and a crate of beer. And, yes WanderingWoumble's example was clear, as was MG's post.

Weighing the advice, I don't think there's much point in adding another 16 GBs, at least not yet.

I hope other users make good use of this information.

 

[geo_gan]

OMG
This is embarrassing. It was so simple, it literally hurts. For me it was simply the tesselation!
It "only" took me 4,5 days to find it out.
I am still not completely sure about that, cause I tried a lot of BIOS settings in the same time, but it seems pretty obvious:

If you frenetically click around in the Adrenalin AMD Driver settings, and click on the "e-sports" preset,
the driver will automatically switch to a manual tesselation setting with the standard value of 8x. It is well hidden behind that "advanced options" button.

VaM obviously is using a tesselation of ca 16x, wich will be applied if you change the tesselation settings to "use application settings". Mind that you have to restart VaM to see the difference.
With VaM only have to render half of the hair, it is no wonder why I've got some pretty impressing benchmark values. Those values exactly matches my previusly measured dream-benchmark results.

Tesselation is the setting that can dramatically change your performance, much more than overclocking your hardware or something like that. For my PC, switching between 8x and 16x is worth more than 30 fps in average, without seeing a big difference in that darkish lighting.

After all I have learned: Hair is one of the biggest performance killer in VaM. That was pretty obvious, because we all know that lowering the hair physics values has a visible effect on performance.
But I never have thought that it would be this much!

Is there a bug in benchmark that means it is not controlling and standardizing some tessellation setting in VAM then during benchmark run?

 

[MacGruber]

Is there a bug in benchmark that means it is not controlling and standardizing some tessellation setting in VAM then during benchmark run

A normal game, let alone a VaM-plugin with all its restrictions, can't change or even just reliably detect things you do via driver-level overrides. Except for the resolution maybe, that you can read at least.

 

[TToby]

Is there a bug in benchmark that means it is not controlling and standardizing some tessellation setting in VAM then during benchmark run?

The benchmark is absolutely great and is standardizing as much as possible. Much more than I ever thought it was doable. But those tesselation settings were in the GPU's driver tools like Vsync and other driver settings, not in VaM. Lets say the benchmark is somewhat fool-proved, but with some extra efford a stupid like me can still make stupid mistakes.