New GameTech GPU benchmark. Share your results! (STEAM page live now)

[stahlhart]

 

[miguel1900]

Thank you all!

@QuietBob , will see what I can do!
@leonavis Thank you for confirming! Blowing my mind, haha
@rusty caterpillar Sure, still not released!
@stahlhart still something I don't like (not sure if it's just underperforming, or maybe overperforming in the opposite areas. I see two strange jumps, in sec 15 and around sec 100. Before and after those timestamps looks to perform uniformly. We would need another 3070ti results to compare). Look at my laptop results, in QHD this time:

 

[stahlhart]

Might have been some sort of driver issue -- two things going on here: there was an Nvidia SDK installed (which I think may have been left over from an installation of that new Nvidia app), which I uninstalled, and I also rolled back the drivers to the same version as your laptop. So either the SDK was causing trouble, or Ampere doesn't appear to like the latest drivers for running your bench. Still investigating.

 

[miguel1900]

Might have been some sort of driver issue -- two things going on here: there was an Nvidia SDK installed (which I think may have been left over from an installation of that new Nvidia app), which I uninstalled, and I also rolled back the drivers to the same version as your laptop. So either the SDK was causing trouble, or Ampere doesn't appear to like the latest drivers for running your bench. Still investigating.

Interesting! Also glad to see the benchmark may be useful to notice this things.

Maybe you could make a couple of tests: reinstall previous drivers -> confirm the problem comes back -> If confirmed, fully cleanup the drivers (https://developer.nvidia.com/cleanup-tool) and reinstall again the same drivers -> confirm if problem persists, so it would be a 100% drivers version issue with Unreal Engine.

I will try your conflictive drivers version in my laptop too.

Thank you @stahlhart

 

[stahlhart]

It looks like it's the driver -- you should probably test your benchmark with a more recent one like this for Nvidia, or at least identify what difference could possibly be causing the hangs, especially in the first minute of the run. But again, 8Gb of VRAM fully consumed probably isn't helping either. Is there any way you can more gracefully throttle cards that hit limits. or at least put guard rails or some sort of warning in the settings, if for example an RTX 3070Ti isn't capable of running in ray tracing mode at 1440p? Because it seems like it has no problem with 1080p.

Another minor complaint -- the music playing in the garage abruptly stopping and starting between scenes breaks immersion.

 

[miguel1900]

Thank you @stahlhart ,

I have tested in my laptop and the result is the same as before:

The thing is your card didn't exhaust the memory. GPUs can reach 100% and even a very little more during a second (due to some 'cache' still not cleared and/or shared memory, maybe) without falling down. Unfortunately some Nvidia drivers bring problems to Unreal (or to other games), but that's only a Nvidia thing, so it's also not possible to warn if a GPU will be able to finely run the benchmark, besides when setting up the minimum requirements, which, in this case, for QHD, would be a 8GB VRAM GPU, like yours and mine. There can be a lot of related hardware/software problems I can't predict until the user run the benchmark (or any piece of software) and notice them locally.

Here I can only say it's a strange situation unfortunately, but your GPU should be definitely able to run it succesfully, as my weaker laptop setup is. There must be something "behind".

About the music I have thought about it some times already! It happens because it's a spatialized sound from the car radio, so you can't hear it when far away. But yeah, I have thought if it would be better to just keep the music as background, without spatialization.

Thanks!

 

[stahlhart]

How are you defining (or planning to define) running the benchmark "successfully"?

Regarding the sound, the spatialization is really only faulty in one spot, where the camera is behind the exercising man -- how can there be no music heard at all there, when it is at its loudest in the scene immediately before, when the camera is in front of him? It otherwise seems to be fine, apart from this. About the only place in the garage that you probably wouldn't hear the music would be in front of the running vehicle, where the engine noise would mostly or completely drown it out. Yes, I am being nitpicky here, and I apologize.

 

[rusty caterpillar]

Regarding the sound, the spatialization is really only faulty in one spot, where the camera is behind the exercising man -- how can there be no music heard at all there, when it is at its loudest in the scene immediately before, when the camera is in front of him? It otherwise seems to be fine, apart from this.

Trying to help.
You got on your MOBO ALC 4080 codec did you disable it? Is the driver wiped off completely?
Was reported disturbances on that codec because of the silly design, USB voltages running to close to it.
Try to run https://www.resplendence.com/latencymon along with an intensive game , 1hr or so, see if is picking up issues.
For example can be Lan driver messing up audio latency, I've seen it before, or Nvidia driver messing up the Lan. Is all about latency in drivers and how fast they respond to the hardware and vice -versa.
Hope it helps

The sound is cutting from one speaker to the other in like 5-6 places along the bench. But is not an an abrupt interruption, is like sound is simply moving form one speaker to the other muting out 1 speaker in bringing alive the other.
Hope it helps

Thank you all!

@QuietBob , will see what I can do!
@leonavis Thank you for confirming! Blowing my mind, haha
@rusty caterpillar Sure, still not released!
@stahlhart still something I don't like (not sure if it's just underperforming, or maybe overperforming in the opposite areas. I see two strange jumps, in sec 15 and around sec 100. Before and after those timestamps looks to perform uniformly. We would need another 3070ti results to compare). Look at my laptop results, in QHD this time:
View attachment 373651

Thanks. Is on my wish list in Steam. For example with the sound issue, I run again your bench but, on 1440P this time on frame 41 sound is moving fast and abrupt from R > Left speaker same on 101, 117, 131 but sometimes is not muting the speaker is just a much lower volume or tries to build up the sound somewhere in the back or other surroundings. I have ALC 1220.
Hope it helps

 

[stahlhart]

Trying to help.
You got on your MOBO ALC 4080 codec did you disable it? Is the driver wiped off completely?
Was reported disturbances on that codec because of the silly design, USB voltages running to close to it.
Try to run https://www.resplendence.com/latencymon along with an intensive game, 1hr or so, see if is picking up issues.
For example, can be Lan driver messing up audio latency, I've seen it before, or Nvidia driver messing up the Lan. Is all about latency in drivers and how fast they respond to the hardware and vice -versa.
Hope it helps

The sound is cutting from one speaker to the other in like 5-6 places along the bench. But is not an abrupt interruption, is like sound is simply moving from one speaker to the other muting out 1 speaker in bringing alive the other.
Hope it helps

I am not using onboard sound on this motherboard, and I have never installed the drivers for it, since it was recently given a fresh 24H2 installation. Back when I first upgraded to Z690, I had problems from the start with the ALC4080 causing popping and stutters in audio that I was never able to resolve, in spite of multiple driver versions and various adjustment attempts, so I threw in the towel and installed sound cards in all three of the builds here that have it. Checking the codecs with msinfo32 reveals that they're just the same five that I am assuming Windows installs by default, because they look identical everywhere.

But you did catch the problem, in spite of this -- the Creative drivers had defaulted the speaker settings to 5.1. I am a dumbass for not having checked this before posting earlier, because it has happened to me before with Creative's application, I just completely forgot to look at that setting. Putting it back to plain old 2.1 fixed the spatial issue.

I don't know why Creative insists on setting to 5.1 when it doesn't bother to check to see what jacks you are using, and worse, why it will occasionally decide to put the setting back to that after you have already told it otherwise. But I appreciate you prompting me to look, thank you -- just getting too old.

The bench is still dropping into the red with high VRAM usage with ray tracing enabled on this build, spiking a few seconds into the run. Sometimes it recovers afterward, other times it is a slide show all the way through. Will keep looking into it. But this build has recorded Legendary scores in 3DMark that I have posted here in another thread, it's been successfully stress tested with OCCT, and I'm not seeing any bad behavior in any other benchmark or game that is running or set up within the hardware's limits.

 

[Ivan Bliminse]

 

[rusty caterpillar]

I am not using onboard sound on this motherboard, and I have never installed the drivers for it, since it was recently given a fresh 24H2 installation. Back when I first upgraded to Z690, I had problems from the start with the ALC4080 causing popping and stutters in audio that I was never able to resolve, in spite of multiple driver versions and various adjustment attempts, so I threw in the towel and installed sound cards in all three of the builds here that have it. Checking the codecs with msinfo32 reveals that they're just the same five that I am assuming Windows installs by default, because they look identical everywhere.

But you did catch the problem, in spite of this -- the Creative drivers had defaulted the speaker settings to 5.1. I am a dumbass for not having checked this before posting earlier, because it has happened to me before with Creative's application, I just completely forgot to look at that setting. Putting it back to plain old 2.1 fixed the spatial issue.

I don't know why Creative insists on setting to 5.1 when it doesn't bother to check to see what jacks you are using, and worse, why it will occasionally decide to put the setting back to that after you have already told it otherwise. But I appreciate you prompting me to look, thank you -- just getting too old.

The bench is still dropping into the red with high VRAM usage with ray tracing enabled on this build, spiking a few seconds into the run. Sometimes it recovers afterward, other times it is a slide show all the way through. Will keep looking into it. But this build has recorded Legendary scores in 3DMark that I have posted here in another thread, it's been successfully stress tested with OCCT, and I'm not seeing any bad behavior in any other benchmark or game that is running or set up within the hardware's limits.

If you disabled 4080 in BIOS than windows can't install the drivers for it. That's the norm.
For example I disabled BT and WI FI on my MB in BIOS but, Win 10 see it and is trying to install drivers and fails. Yet it seems is allocating resources to those devices, which is kind of a vulnerability.
Reported to Asrock but, for them seems normal. Though I never seen before that in my life, disabled devices in BIOS to be seen by the OS.

I had Creative Fatality before, always had fishy drivers/"system files" which will be placed in c:\Windows\System32\ and always delete them, after each installation. Always my security flaring up at each Creative audio installation. Was supported for short time with drivers, in the end I got rid of it.
I had hard time to find a Mobo for myself with ALC 1200 or 1220 and be reliable on other parts to, since ALC 897 is too old and might go out of support while all the ALC starting with 40 numbers had issues because of the silly implementation.
So I had to chose only MB with ALC 1200 or 122o or take the chances with ALC 897 and be ready to buy a soundcard but, than what sound card with clean drivers and not a short support. Was kind of blurry.

 

[stahlhart]

[

If you disabled 4080 in BIOS than windows can't install the drivers for it. That's the norm.
For example I disabled BT and WI FI on my MB in BIOS but, Win 10 see it and is trying to install drivers and fails. Yet it seems is allocating resources to those devices, which is kind of a vulnerability.
Reported to Asrock but, for them seems normal. Though I never seen before that in my life, disabled devices in BIOS to be seen by the OS.

I had Creative Fatality before, always had fishy drivers/"system files" which will be placed in c:\Windows\System32\ and always delete them, after each installation. Always my security flaring up at each Creative audio installation. Was supported for short time with drivers, in the end I got rid of it.
I had hard time to find a Mobo for myself with ALC 1200 or 1220 and be reliable on other parts to, since ALC 897 is too old and might go out of support while all the ALC starting with 40 numbers had issues because of the silly implementation.
So I had to chose only MB with ALC 1200 or 122o or take the chances with ALC 897 and be ready to buy a soundcard but, than what sound card with clean drivers and not a short support. Was kind of blurry.

I was really hoping that I would be permanently switched over to onboard audio back at Haswell-E, but ALC 4080 was a real disappointment. Not sure what possessed Realtek to go and break something that was working perfectly okay as it was (SupremeFX, in my case). Why USB 2.0, what was the point of that?

Anyway, after more testing, the results are the same -- crash and burn at 1440p in ray tracing mode. Of note is that after the benchmark run my C: drive (a Western Digital SN850X 2Tb) has increased in temperature about 10C, which only happens with extensive writes, so the bench has run out of VRAM and is swapping in and out of storage instead.

These are the drive temps after the 1440p RT run.

 

[miguel1900]

Hi @stahlhart and @rusty caterpillar,

Thank you both for the interesting conversation and for helping each other out!

I haven’t noticed any abrupt interruptions, but I’ve observed some minor oddities in the sound, specifically on my laptop. The sound wasn’t deeply tested in this build. I’ve redesigned the audio behavior for the upcoming version (not yet available for download), but it still hasn’t undergone thorough testing.

Regarding the ray tracing performance drops, @stahlhart, I’m sorry I can’t offer more help here. That said, for a benchmark, it’s fine to be a bit sensitive and "on the edge." Other benchmarks or games might run smoothly on a similar setup (are those hitting around ~7.25GB?), but sometimes only one very specific scenario can reveal an underlying issue. Of course, this should be an exception—if it were the norm, the software would be the issue, but that doesn't seem to be the case.

Unfortunately, I haven’t seen this behavior before, and my similar yet less powerful laptop runs it without surprises. We would need to see another 3070ti run to comparei (please, anyone with one, try it at 1440p!). It’s possible there’s an issue specific to your unit or that particular model’s last 0.5GB of VRAM (for example, remembering the GTX970), which might cause it to throttle prematurely. It’s worth to keep an eye and comparing results with similar builds. I’m sorry I can’t provide a clearer answer for now, but it does seem like a strange and isolated case. Maybe if you write to the manufacturer, they would be able to test too and give you/us an answer.

By the way, have you tried enabling or disabling ReBAR?

Regards!

 

[rusty caterpillar]

[


I was really hoping that I would be permanently switched over to onboard audio back at Haswell-E, but ALC 4080 was a real disappointment. Not sure what possessed Realtek to go and break something that was working perfectly okay as it was (SupremeFX, in my case). Why USB 2.0, what was the point of that?

Anyway, after more testing, the results are the same -- crash and burn at 1440p in ray tracing mode. Of note is that after the benchmark run my C: drive (a Western Digital SN850X 2Tb) has increased in temperature about 10C, which only happens with extensive writes, so the bench has run out of VRAM and is swapping in and out of storage instead.

View attachment 373925

View attachment 373926

View attachment 373927

View attachment 373929

View attachment 373930

These are the drive temps after the 1440p RT run.

With ALC 4080 I don't incline towards Realtek fault . They made the chip but don't think so they push MOBO manufacturers to run USB trails next to it.
IMO the fault lies with the MOBO manufacturers in implementing ALC 4080 together with USB as crammed as possible. A flawed design from the start.

And they knew the electricity trails will influence negative the sound. Is an old known fact. Some of them added metal casing on top of the sound circuits to provide EMF shield yet the problem was underneath the shield already

Glad that you trying to check everything on your system, but drives temps shouldn't influence the benchmark IMO, specially when your boot drive is very fast and is a good one, is not a P3 to throttle at low temps.
I would run this benchmark x3 @1440P and check memory junction temps for the 3070, maybe the VRAM is throttled by the heat, just a wild guess, I don't know if they coded something in the BIOS of 3070 and 3080 to throttle down VRAM if is too hot. PCBs on this cards are very crammed and thin which will make the heat traverse the PCB faster towards the backplate.
On the other hand the benchmark is at the start. Is an alpha stage I guess.
Your VRAM 256 bit bandwidth should suffice for 1440 P but maybe benchmark is pushing it hard... IDK.
As you can see my 1080 Ti with 352 bit bandwidth doesn't do better on the VRAM side and is got wider bandwidth and +3GB more.

Is also possible that, maybe, you have leaky pads on your 3070, silicon grease/oil will attract moist trough dust collection points on PCB will which will enhance the possibility of unwanted small electrical discharges all over the place(PCB level) in between electrical components which will result in instability of the card, is a chain reaction.

You can see in the picture bellow how dust was piling up, around various components(which is the main issue dust is trapping moisture), on silicon oil leak patches on my 1080Ti PCB under the back plate.
Of course the humidity in the room is an important factor. The black patches where the thermal pads was sitting, the rest is silicon oil with cumulated dust.

I had bad leaky pads but, my card is EVGA. Change them all with thermal putty Upsiren UTP 8 and got 8 C less. That putty is very good and cheaper than thermal pads. It won't leak either.
Other thermal pads will leak, I've seen on Sapphire and other big brands.

Just some insights in "electronic" instability, and obviously poor stability under the stress.

 

[miguel1900]

Thank you @rusty caterpillar for the interesting hardware reports!

A new little update is available valid until the 15th. Link in itch.io's page. I hope there is a better sound experience now, between other little things.

@QuietBob , maybe you are still not seeing the animated menu background (do not know why), but maybe it's now showing a dark grey solid background, could you check it please?

@stahlhart , I have read, bu casuality (a not very deeply) that the last Windows 11 update and last Nvidia drivers are causing general issues with Unreal, at least, so it's a thing to be solved by them, apparently.

Thank you!

 

[Launcestonian]

Followed on Steam. Thanks miguel1900!
Ryzen 9 9900X - custom PBO with SMT disabled. dGPU running at stock.

 

[miguel1900]

Thank you @Launcestonian !

Usually, CPU shouldn't play a main role in this bench, mainly the GPU.

PD: I will be posting a new little update during today-tormorrow. Previous version has expired today.

 

[Prime2515102]

Where can I get the latest version of this? When I run 5.3.2.0 (the newest I can find) I get this:

Steam only has an add to wish list option. Did something change? Is it disabled until the Steam launch?

 

[miguel1900]

Hi @Prime2515102 !

Just in the post above!

The current version expired today, but I'm publishing a new update very soon.

In Steam there is only the whishlist, you are right. I'm not very familiar with it but I will try to post there the same version (as early access I suppose) too and/or the Demo, but in some days.

Thank you for commenting! Keep an eye here for the update.

 

[Launcestonian]

Thank you @Launcestonian !

Usually, CPU shouldn't play a main role in this bench, mainly the GPU.

PD: I will be posting a new little update during today-tormorrow. Previous version has expired today.

Correct, normally a CPU would not be such an indicator of outright performance, but as you know there is more than just UE5 game engine out there & devs have a track record of not optimizing games for PC normally. So as an end user, you do your bit from your end for peak performance & hope for the best. Those CPU settings I have are my daily driver gaming rig.

 

[venturi]

Please make these so that they don't have an expire date, thank you

 

[miguel1900]

Correct, normally a CPU would not be such an indicator of outright performance, but as you know there is more than just UE5 game engine out there & devs have a track record of not optimizing games for PC normally. So as an end user, you do your bit from your end for peak performance & hope for the best. Those CPU settings I have are my daily driver gaming rig.

Sure! I didn't want to mean your CPU mention had no sense. It may be useful, anyway, for stuttering, better lows 1% and 0.1%, loading times, etc! Every component is important, of course. I just wanted to clarify a little more the nauture of my bench.

Please make these so that they don't have an expire date, thank you

Haha, so easy to ask for!

I understand you, but in this stage (public beta) is quite important to have an expiration date: first of all, to be able to close the public beta after a date, but much more important because this is not a game, but a benchmark; it's really important, specially for you, the users, to ensure all of you are using the same versions, so you can compare your benchmark results with equity. Anyway, I'm working in a better 'remote' system to control it, instead of a fixed date. Coming very soon.

EDIT: new update already available!

Thank you!

 

[miguel1900]

As commented yesterday, the new update v0.999 is already available! In the road to release!! (Before new Nvidia 50 GPUs release).

I made some minor changes. Mainly, you will notice a slightly higher scores.

I'm also waiting and making tons of test with the latest version of Unreal, to see if I can update it before release, but it's still quite broken and requires a lot of work arounds so, maybe, I will stay in Unreal 5.3

 

[venturi]

Sure! I didn't want to mean your CPU mention had no sense. It may be useful, anyway, for stuttering, better lows 1% and 0.1%, loading times, etc! Every component is important, of course. I just wanted to clarify a little more the nauture of my bench.


Haha, so easy to ask for!

I understand you, but in this stage (public beta) is quite important to have an expiration date: first of all, to be able to close the public beta after a date, but much more important because this is not a game, but a benchmark; it's really important, specially for you, the users, to ensure all of you are using the same versions, so you can compare your benchmark results with equity. Anyway, I'm working in a better 'remote' system to control it, instead of a fixed date. Coming very soon.

EDIT: new update already available!

Thank you!

It actually works the other way around for some of us:

When I compare an equipment upgrade, I will do an apples top apples of what that app, game. benchmark was on the old equipment and the new. So having a static benchmark app is always a welcome comparison. If versions keep changing and versions no longer launch, then that comparison is not possible.

Or let folks download and keep the relative versions so that they can intelligently do the hardware and system comparisons on timeline.

Honestly I have never used an expiring benchmark tool. Loses the timeline and hardware comparisons.

Critical comparisons can be between an April BIOS and a December BIOS, a situation I would like to compare.

While expiring benchmark tools, make the purpose of a benchmark results NOT comparable.

Having the ability to statically compare using the same "measuring stick" is the ONLY way to make the comparisons.


A great example is gravity mark:
multi platform, multi language, multi parameter, and multi engine --> and available for download the prior versions and they don't expire.

check out gravity mark:

GravityMark GPU Benchmark

GravityMark GPU Benchmark

gravitymark.tellusim.com

check out scoring system:

GravityMark Leaderboard

GravityMark GPU Benchmark

gravitymark.tellusim.com

In my humble opinion on benchmarking

 

[miguel1900]

Thank you for your feedback @venturi !

Quite fair.

In fact, you shouldn't worry about performance discrepancies between versions, as the benchmark won't receive important updates that could affect performance after the official release (scheduled around Nvidia's CES), because reviewers may have already started making GPU reviews based on the benchmark. From that moment, performance will need to be maintained to make it comparable with future reviews as well (RTX xx60, for example). Updates after that point could be just bugfixes and/or cosmetic, such as UI updates, new options, or additional information, but they could also involve sponsorships. Therefore, it would be important to make users update the software to view the new logos, for example.

However, in this beta stage, I still don't want a lot of "unofficial" versions circulating. That's why I specified that "killing" old beta versions is specially useful now. In addition, there will be a paid version, so until both versions are perfectly defined, I must keep things under control, at the beginning, at least.

Anyway, there will also be a free/demo "permanent" version (more limited in options or information shown).

Regarding the variety of options, I may add more customizable options in the future if there are enough users to create a large database for each one. For the moment, many options would only fragment the results database, making it less useful.

I hope you like it more now.

Thank you again!