GameTechBench GPU benchmark is already out!

[miguel1900]

Thank you very much @adilazimdegilx and @AVATARAT !!

@adilazimdegilx In fact that's something I need to avoid! hahaha. As this is an "experimental/broken" version of Unreal, packaging was problematic and it seems it let those settings exposed. Thank you for reporting too


PS: Please all! Any of you could record your run and post a thread in Reddit -> Nvidia -> Benchmarks? https://www.reddit.com/r/nvidia/?f=flair_name:"Benchmarks"

You could publish instead my own video, too:

I would be super glad! (I can giveaway a free key for the game, for example) And any visibility boost would be beneficial for all of us. More users will make this more useful!


Thank you so much!

 

[yzonker]

@miguel1900 This final version works fine on my ultrawide. Must have just been something with the previous version.

FYI, looks like they deleted your reddit post.

Just for reference, this is what the previous version did. Image is offset on the screen. But seems to be fixed now.

 

[miguel1900]

@miguel1900 This final version works fine on my ultrawide. Must have just been something with the previous version.

FYI, looks like they deleted your reddit post.

View attachment 380408

Just for reference, this is what the previous version did. Image is offset on the screen. But seems to be fixed now.

Thank you for the confirmation @yzonker !

I changed how the screen resolution was managed (due to a bug in the current Unreal version, in fact) and it seems to have solved it too.

PS: I was imagining the would delete it due to self-advertising... You can post your results overthere, if you want, even just as an image!

Thanks again!

 

[yzonker]

About 60% difference between the 3090 and 4090 in this. Thought the difference might be bigger. Interesting to see what the 5090 can do.

 

[miguel1900]

About 60% difference between the 3090 and 4090 in this. Thought the difference might be bigger. Interesting to see what the 5090 can do.

View attachment 380492

Yeah, noticing this huge difference from time ago. And maybe this was one of the reasons took me here, after lot of confusing performance comparisons. Congrats for your new GPU!

I can't wait to see the 5090 too!

 

[wNotyarD]

Did some runs today, sharing now:

Spoiler: Raster 1440p

Windows

Garuda Linux - Proton 9.0-4

Garuda Linux - Proton Experimental

Spoiler: RT 1440p

Windows

Garuda Linux - Proton 9.0-4

Garuda Linux - Proton Experimental

Some conclusions of mine:
1st - RADV must still come a long way forward, regarding RT;
2nd - There's something wrong with my Windows install, these dips can't be normal.

 

[SoppingClam]

Does it or will it be able to upload the results onto a website with others? To make it easy to compare and share with others?

If its any good, it could become popular as people want to test our their fancy new GPUs to see how well they are running with a variety of different bells and whistles. Apart from 3DMark there hasn't been any mainstream new benchmarks since 2017 with Superposition.

 

[miguel1900]

Quite interesting @wNotyarD thank you for sharing ! Does it run natively in Linux or is it some kind of emulation? (Sorry, not familiar with Linux). About you Windows results, I only see some spikes, but not too much. Anyway is interesting the smoothness in Linux.

Does it or will it be able to upload the results onto a website with others? To make it easy to compare and share with others?

@SoppingClam absolutely yes!! You may have missed this post . Comming very very soon!:

 

[wNotyarD]

Quite interesting @wNotyarD thank you for sharing ! Does it run natively in Linux or is it some kind of emulation? (Sorry, not familiar with Linux). About you Windows results, I only see some spikes, but not too much. Anyway is interesting the smoothness in Linux.

Running under Proton in Linux. Technically speaking it ain't emulation but a Wine-based compatibility layer (WINE Is Not an Emulator, after all) which translates Windows calls into something Linux can run.
You can tell Steam to try and run anything it has installed with Proton, even software which has native Linux builds.

 

[miguel1900]

Running under Proton in Linux. Technically speaking it ain't emulation but a Wine-based compatibility layer (WINE Is Not an Emulator, after all) which translates Windows calls into something Linux can run.
You can tell Steam to try and run anything it has installed with Proton, even software which has native Linux builds.

Thank you for the explanation @wNotyarD ! I will check my detection tool, which detects it as Windows Unknown. Even if I'm not sure if it will be possible to get is as Linux.

PS: some real data already arriving in the last 10 hours! Please all, could you make anew run to include your GPU? Like @yzonker with your 4090! Only Ray Tracing benchmark at every resolution will count. (Scores with OS as 'Unknown' will be discarded, sorry):

PS2: We already have a 5090!! But GPUs under embargo will remain hidden. If you have one, rest assured!

(Graph still not available publicly, but very very soon!) @SoppingClam

Thank you very much!

 

[miguel1900]

The comparison charts are ready now!!

And the 5090 is awaiting that countdown!

 

[miguel1900]

Some ("impressive"?) 5090 results!!

 

[QuietBob]

First off, thank you @miguel1900 for your continued efforts to improve the benchmark. I just ran the Steam version and here is how the RT ranked test performs on my hardware:

Some observations and questions:

• The background for the main menu still does not render, displaying a solid white screen instead.
• Upon completing the benchmark the graphs in the main menu also do not show, and the benchmark result isn't saved when re-opening the app. The graphs show correctly on first launching the app.
• Perhaps changing the bar colors to represent the GPU vendor would make them easier to read. As in, green for Nvidia, red for AMD and blue for Intel.
• What does the "Warming up duration" option do and how does it affect performance? Is this time spent building shaders?

EDIT:
With more testing, I noticed the following on my end:

• The Steam client can't be in offline mode for the graphs to show. Still, in online mode occasionally they won't display.
• 4320p tests crash in progress every other time, hard locking the PC.
• The PT test with "Low VRAM PT" set to off invariably crashes the app upon completion. It shows the result, returns to the main menu, and then hangs. It will automatically quit to desktop after some time. Curiously, this doesn't happen with the "Low VRAM PT" set to on.
• After finishing any test, the bar graphs show the average results for the monitor's default resolution, rather than the one just tested.
• The average results show rasterized (software Lumen) performance, even after selecting an RT (hardware Lumen) test, or completing one.
• There are some inconsistencies between the bar graphs presented and the selected resolution and test mode, like here:

Please don't take my comments as criticism. I absolutely love this benchmark and admire all your hard work. I'd love your app to reach the same popularity as 3DMark, and become a reference point for evaluating new GPU architectures

 

[wNotyarD]

• What does the "Warming up duration" option do and how does it affect performance? Is this time spent building shaders?

If I were to interpret it, it's literally preheating the card so you don't have "cold test" variance.

 

[miguel1900]

• The background for the main menu still does not render, displaying a solid white screen instead.
• Upon completing the benchmark the graphs in the main menu also do not show, and the benchmark result isn't saved when re-opening the app. The graphs show correctly on first launching the app.
• Perhaps changing the bar colors to represent the GPU vendor would make them easier to read. As in, green for Nvidia, red for AMD and blue for Intel.
• What does the "Warming up duration" option do and how does it affect performance? Is this time spent building shaders?

Thank you @QuietBob !

1- The problem is that I don't know why a few of you see the background as white I have not found the common pattern in you.
2- They should be shown, as for me, 99% of times. Very punctually, however, any of the 6 leaderboards may fail when retrieving from Steam (a Steam server's issue maybe?), but this should be quite rare. It tries to retrieve the leaderborads iterativelly, during 10 seconds, but I will take a look, if I notice anything incorrectly programmed.
Are you sure your results are not saved? They should be integrated, however, into the average score, keeping your best one.
3- Can be better for filtering, with so many cards, yeah! Will be included
4- What @wNotyarD said!

 

[QuietBob]

@miguel1900 Please see edits to my previous post

No, the results aren't saved after testing a different resolution or quitting the app. It would be great if you could see your last score across all resolutions on the comparison chart every time you run the app.

 

[miguel1900]

With more testing, I noticed the following on my end:

• The Steam client can't be in offline mode for the graphs to show. Still, in online mode occasionally they won't display.
• 4320p tests crash in progress every other time, hard locking the PC.
• The PT test with "Low VRAM PT" set to off invariably crashes the app upon completion. It shows the result, returns to the main menu, and then hangs. It will automatically quit to desktop after some time. Curiously, this doesn't happen with the "Low VRAM PT" set to on.
• After finishing any test, the bar graphs show the average results for the monitor's default resolution, rather than the one just tested.
• The average results show rasterized (software Lumen) performance, even after selecting an RT (hardware Lumen) test, or completing one.
• There are some inconsistencies between the bar graphs presented and the selected resolution and test mode, like here:

View attachment 381501
Please don't take my comments as criticism. I absolutely love this benchmark and admire all your hard work. I'd love your app to reach the same popularity as 3DMark, and become a reference point for evaluating new GPU architectures

Don't worry! Thank you for the feedback.

I think some of them are due to not using them as intended:
1- Yes, you need to be connected to retrieve the online database (in the next update it must be more flexible, if it fails)
2- This is the first crash report about it. It's a huge resolution, maybe you GPU can't handle it. Anything below a 3080ti, probably. Do you have a 7900 XTX right? I think @AVATARAT too and also ran that test. Maybe he an tell us if it crashed for him. A benchmak is also made to detect instabilities
3- Yeah, a few users reported it. Unfortunately it only happens when going back to menu. I think it's something internal of this Unreal version, as this rendering system is quite experimental and the last Unreal version too. I hope Epic will fix the related thing behind in the next hotfix.
4- Yes, the menu options returns to their defaults, and as leaderboards depends on the menu options (the resolution selected there is the one shown in the leaderboards), they also shown the default resolution ranking.
5- There are no Raster leaderboards, only Ray Tracing and Path Tracing ()
6- How have you got that non-correspondent resolution-leaderboards? I can't reproduce this issue. Maybe it only happens when there are issue retrieving the Steam leaderboards. Let's see in the next update.

PS: update coming in some minutes! Please, try again to discard some of the points

 

[AVATARAT]

No crashes here.

 

[miguel1900]

No crashes here.

Thank you for the confirmation @AVATARAT ! The misteries of computers, haha. But 'may be' good to have some users with crash and others without. It may indicate some kind of....... instability/weakness? I don't know. Time will tell about that misterious final crash after the PT test.

BTW, @QuietBob :

BTW, @AVATARAT , please, how much score do you get in 3dmark? You must have the world record for the 7900 XTXs. For me, it still looks like a very lot! And I'm usually asked about that super high score

 

[AVATARAT]

No crashes here.

For the @4320p crashes, if there is something getting video memory, your benchmark will crash for sure, I usually close everything when running benchmarks. If you do lower resolution runs and end up running this at 4320p, it will crash too, so better try that first. This resolution eats up all the video memory on the 7900 XTX and it plays on the edge on some MB, so be careful

 

[Paramount]

RT score

Raster

I saw differ scores compared to Nov 2024 benchmark !
Benchmark crash twice during launch
But its great tool

 

[miguel1900]

For the @4320p crashes, if there is something getting video memory, your benchmark will crash for sure, I usually close everything when running benchmarks. If you do lower resolution runs and end up running this at 4320p, it will crash too, so better try that first. This resolution eats up all the video memory on the 7900 XTX and it plays on the edge on some MB, so be careful

In terms of memory, if you (any) don't have enough memory, it's always recommended to restart the benchmark to fully clear/reset the memory load, if you want to compete in the master race. I have implemented all the clearing tricks inside Unreal, but additional MBs will be always kept ni memory after a test. BTW, it "only" got 22,5GB as max in my 3090.

I saw differ scores compared to Nov 2024 benchmark !
Benchmark crash twice during launch
But its great tool

Thank you Paramount! Sure, scores before version 1.0 should be discarded

At launch?! That's new! First time reading this, unfortunately. But this is using some experimental techs, which is even better for a benchmark, to be more "instability-sensitive".

Thank you all!

 

[yzonker]

Well since the 30th was sad banana day with only paper 5090's available, I decided to make a better run with the 4090. Looks like it picks up my Steam ID of talon95.

 

[remekra]

Need to find out the OC headroom on that card Got the Gigabyte one that can be pushed to 450W, gotta get closer to 4090

EDIT

+500 on memory and +450 on core, so far stable but needs more testing.

 

[miguel1900]

Well since the 30th was sad banana day with only paper 5090's available, I decided to make a better run with the 4090. Looks like it picks up my Steam ID of talon95.

Wow! How did you get that outstanding score? It's around 300pts over the second best 4090 user

Need to find out the OC headroom on that card Got the Gigabyte one that can be pushed to 450W, gotta get closer to 4090



EDIT

+500 on memory and +450 on core, so far stable but needs more testing.

Nice performance improvement! A pro overclocker was telling me this is a great OC tool as it's quite instability-sensitive too, when the OCs are too high