GameTechBench GPU benchmark is already out!

[Launcestonian]

I have only set -30 in curve optimiser, nothing else.

But do you have a positive boost on the cpu clock & if you do by how much?

 

[Kminek]

But do you have a positive boost on the cpu clock & if you do by how much?

Tried set +100 on the cpu clock and result was even worse for couple %.

 

[miguel1900]

Interesting to see CPU can have an impact too, as this is mainly a GPU benchmark! Thank you both @Launcestonian @Kminek

 

[AVATARAT]

9700x iGPU at 1080p
(and 2 screenshots during the bench)

 

[miguel1900]

Thank you very much @AVATARAT ,

It's a pity you hadn't your GPU anymore! I'm in fact impresed that integrated one was able to even run it . I have taken an idea from your screenshots for Raster mode, which doesn't allow light bounces, adding now a 'fill' lighting to help with dark areas.

For the rest, FYI, he was testing a new version (v0.99), because of his ultrawide screen, with some changes I wanted to check before publishing, :

• Options adjustments in main menu, preparing for an upcoming new map and simplifying the options.
  
• Now, aspect ratio of the camera is 2.25:1, due to a mathematical challenge (below), so it should perform a little worse, as some more pixels will be rendered (previously was 2.35:1).
• In addition, I have implemented rendering at the actual pixel count of common resolutions such as 1080p, etc, which are traditionally based on a 16:9 aspect ratio. To achieve this, the resolution will automatically update to match the total number of pixels of these standardized resolutions while taking into account the actual aspect ratio of the rendered image.
  
  For example, on a 1920x1080 screen (around 2.5 million pixels), an image with a 2.25:1 aspect ratio previously rendered approximately 1.6 million pixels. Now, the resolution will be auto calculated to compensate that difference, ensuring the same total number of pixels as Full HD, in that certain example. This way, the benchmark will render the same number of pixels at standardized resolutions, making the results more universally comparable. I think this way is better now, even if sacrifying some performance; you can run a 4k test to know how this kind of graphics would run in 4k, actually. (Yes, the most easy way would have been to just use a 16:9 aspect ratio, but I wanted to keep the more cinematic appeal).
• Added infinite loops option.
• Added a final Score punctuation in the final summary.
  
  
• Only one sequence keept.
• New scene added during the sequence.
  
  

Until 15th October, the previous version and this new one will live together, so you can still use the prev one to compare with prev scores, as this new one will invalidate all prev results.

v0.99: https://drive.google.com/file/d/19Dagdnib5E-6E4XYaT0zFWeJu-gqxfUK/view?usp=drive_link

 

[Launcestonian]

Is there other requirements to run version 0.99 of this game like .net or something? because this version I can't get to run, none of the executable applications do anything. I double click it or highlight & press enter... nothing!

 

[miguel1900]

Is there other requirements to run version 0.99 of this game like .net or something? because this version I can't get to run, none of the executable applications do anything. I double click it or highlight & press enter... nothing!

Hi @Launcestonian ,

Nothing touched in that sense. In fact, I got surprised when you were able to execute it but weren't before, as I didn't touch anything related. Maybe this version of Unreal has a bug when packaging, but I only read you about this issue. You may try to install the Unreal prerequisites by executing the installer found in ...\Engine\Extras\Redist\en-us

Anyway, if you want you can wait a little as today or tomorrow I will upload a new update with some cool mainly cosmetic changes.

Hi all!

New update v.995 available!

Featuring the chart of performance displayed during all the process, but also at the end.

Featuring too the menu, with some options added or changed.

Other minor changes added, like a more high poly version of every mesh (up to double the polycount), at a zero performance cost.

Remember: from v0.99 performance will not be comparable to previous versions.

What's pending now? Integration of the new map + migration to Unreal 5.5 during this month (performance will change again, for sure).

Thank you!

 

[Zerion]

Hi @miguel1900

Is the new version (v0.995) heavier? It seems my avg fps went from around 60 to 47.3.

In addition, I cannot finish the path tracing test as it gave me a fatal error near the end.

 

[miguel1900]

Hi @Zerion !

You are right. It's not heavier, but it renders more pixels. As explained some posts above, let's talk about 1080p 16:9, for example: as the camera has black bars on top and bottom, it was rendering 1.6 million pixels, instead of the 2 millions of "full" 16:9 1080p, so in last version I increased the resolution to compensate it and to really render the same pixels as a traditional (16:9) 1080p picture (2 million pixels). So, in 1080p 16:9 you are now rendering around 25% more pixels so you can expect the same FPS loss.

But I'm thinking about reverting it, if you, hardcore users, don't think it's important to exactly render the exat number of pixels of traditional resolutions based on traditional 16:9, as the world is very diverse. (And because the upcoming forest scene is too heavy).

Said that, maybe higher scores and a more smooth experience is preferable, isn't it? What do you all think?

About the Path Tracing mode error, I noticed that too in some builds of the project (solved rebuilding, without any other change), but wasn't able to reproduce. I think the current UE version may have a bug when packaging, as other related issues happens. But I will update to UE5.5 as soon as it's available, and it can bring some very cool improvements!

 

[Zerion]

Hi @Zerion !

You are right. It's not heavier, but it renders more pixels. As explained some posts above, let's talk about 1080p 16:9, for example: as the camera has black bars on top and bottom, it was rendering 1.6 million pixels, instead of the 2 millions of "full" 16:9 1080p, so in last version I increased the resolution to compensate it and to really render the same pixels as a traditional (16:9) 1080p picture (2 million pixels). So, in 1080p 16:9 you are now rendering around 25% more pixels so you can expect the same FPS loss.

But I'm thinking about reverting it, if you, hardcore users, don't think it's important to exactly render the exat number of pixels of traditional resolutions based on traditional 16:9, as the world is very diverse. (And because the upcoming forest scene is too heavy).

Said that, maybe higher scores and a more smooth experience is preferable, isn't it? What do you all think?

About the Path Tracing mode error, I noticed that too in some builds of the project (solved rebuilding, without any other change), but wasn't able to reproduce. I think the current UE version may have a bug when packaging, as other related issues happens. But I will update to UE5.5 as soon as it's available, and it can bring some very cool improvements!

I see, I thought it would be easier to just remove the black bars.

It's alright for me since it is processing the actual number of pixels.

 

[miguel1900]

I see, I thought it would be easier to just remove the black bars.

It's alright for me since it is processing the actual number of pixels.

Haha, in fact it is!! But I prefer to preserve the cinematic aspect ratio to give it more realism

 

[Bet0n]

For science

Btw the CPU is oc'd to 4.6 GHz, not that it matters...

 

[miguel1900]

For science


Btw the CPU is oc'd to 4.6 GHz, not that it matters...

I love science and data!

Would you want to test it again? It should now get less than 6GB Vram for 1440p (but in your case, select 1080p or even 720p), which was tanking your performance (additionally).

@Zerion , would you mind redonwloading again too, specifically to check the Path Tracing error?

PS: v0.996 available. Some internal changes to optimize data for very looooooong tests (dozens of loops), which could cause some punctual stuttering due to huge amount of FPS data stored (an array with half a million values, for example).

 

[Bet0n]

I love science and data!

Would you want to test it again? It should now get less than 6GB Vram for 1440p (but in your case, select 1080p or even 720p), which was tanking your performance (additionally).

I ran it in 1080p, it's on the results screenshot.
I re-downloaded the benchmark from your site, both versions and I got this error message from both:
"Internet connection lost or expiration date reached.
Please, check if a new version is available for download.
Apologies for the inconvenience. Closing automatically."

 

[miguel1900]

I ran it in 1080p, it's on the results screenshot.
I re-downloaded the benchmark from your site, both versions and I got this error message from both:
"Internet connection lost or expiration date reached.
Please, check if a new version is available for download.
Apologies for the inconvenience. Closing automatically."

Sorry, my fault! The previous expiration date was yesterday, but I introduced the new one in the wrong place, so it was still yesterday. (Fixed now)

Apologizes

 

[RUSerious]

Sorry, my fault! I introduced the new expiration date in the wrong place. (Fixed now)

Apologizes

Oh great! I went to try it a couple of days ago and got the same error. I'll give it a try later on.

 

[miguel1900]

Waiting for it, @RUSerious !

I have added a new, more dynamic shot (another 15 secs) during the sequence. Will be added in the next update:

 

[Athlonite]

Raster vs Ray Traced
Raster

Ray Traced

 

[stealth83]

 

[RUSerious]

Here you go @miguel1900

 

[mbeeston]

 

[miguel1900]

Thank you all guys!

Good machine @stealth83 ! However, did you notice any stuttering or something? As your performance chart looks a little (too much) noisy

 

[Kodehawa]

 

[stealth83]

Thank you all guys!

Good machine @stealth83 ! However, did you notice any stuttering or something? As your performance chart looks a little (too much) noisy

Didn't notice any, all seemed fine

 

[miguel1900]

Didn't notice any, all seemed fine

Thank you @stealth83 ! Maybe it's more sensible than the eye, which is good too.

Updated to v0.997! The main change is the new shot at the final of the sequence and some minor changes in the graph.