• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

GravityMark v1.87 GPU Benchmark

venturi said:
thank you ;)
Click to expand...
A new 1.51 version has better NVLink performance.
RTX 3090 x2 AFR / RTX 2080 Ti x2 AFR CU = 71,424 / 55,086 = 1.3
It's very likely that it will be possible to go over 100,000 scores at lower resolutions.
 
Hi I just tried it, how do I use the shortcut with the switches and include the person name info?

also, forgive the question, how do I get Vulkan into "performance mode"

Thanks
 
venturi said:
Hi I just tried it, how do I use the shortcut with the switches and include the person name info?

also, forgive the question, how do I get Vulkan into "performance mode"

Thanks
Click to expand...

-f 1 # show fps
-i 1 # show info
-s 1 # show sensors
-b 1 # run benchmark
-fs 1 # fullscreen
-mode 2k # video mode (hd, hd+, fhd, 2k, 4k, 5k, 8k)
-n $USER # name

set NAME=username
GravityMark.exe -f 1 -i 1 -s 1 -n %NAME% -b 1 -alternate 1 -vk -cuda 1 -fs 1 -mode 2k
GravityMark.exe -f 1 -i 1 -s 1 -n %NAME% -b 1 -alternate 1 -vk -cuda 1 -fs 1 -mode 4k
GravityMark.exe -f 1 -i 1 -s 1 -n %NAME% -b 1 -alternate 1 -vk -cuda 1 -fs 1 -mode 5k
GravityMark.exe -f 1 -i 1 -s 1 -n %NAME% -b 1 -alternate 1 -vk -cuda 1 -fs 1 -mode 8k

Only this configuration is routing Cuda traffic through NVLink on Windows:

sli.png
 

Attachments

Last edited:
Tellusim said:
-f 1 # show fps
-i 1 # show info
-s 1 # show sensors
-b 1 # run benchmark
-fs 1 # fullscreen
-mode 2k # video mode (hd, hd+, fhd, 2k, 4k, 5k, 8k)
-n $USER # name

set NAME=username
GravityMark.exe -f 1 -i 1 -s 1 -n %NAME% -b 1 -alternate 1 -vk -cuda 1 -fs 1 -mode 2k
GravityMark.exe -f 1 -i 1 -s 1 -n %NAME% -b 1 -alternate 1 -vk -cuda 1 -fs 1 -mode 4k
GravityMark.exe -f 1 -i 1 -s 1 -n %NAME% -b 1 -alternate 1 -vk -cuda 1 -fs 1 -mode 5k
GravityMark.exe -f 1 -i 1 -s 1 -n %NAME% -b 1 -alternate 1 -vk -cuda 1 -fs 1 -mode 8k

Only this configuration is routing Cuda traffic through NVLink on Windows:

View attachment 239604
Click to expand...
Quite awesome and THANK YOU
 
thx,
Tellusim said:
Thank you for your report.
NVLink is not providing much improvement in the 2K mode. But it can double performance at 4K and higher resolutions.
Click to expand...
I did a few benches on 4k and with ray tracing, posted, could you correct the incorrect ram report, somehow it still doesn't see the ram past the 1st processor, should be 1.5TB of ram
oddly, twice it did see all of it. Maybe its numa or qpi related...

by the way yes, you're right (logically) better nvlink performance at the higher resolution


and the new test are using dual 3090 TI FE RTX in nvlink.....
;)


ps, is the hierarchy tab on site aggregated to include the dual and or the 3090 ti?
 
Last edited:
venturi said:
thx,

I did a few benches on 4k and with ray tracing, posted, could you correct the incorrect ram report, somehow it still doesn't see the ram past the 1st processor, should be 1.5TB of ram
oddly, twice it did see all of it. Maybe its numa or qpi related...

by the way yes, you're right (logically) better nvlink performance at the higher resolution


and the new test are using dual 3090 TI FE RTX in nvlink.....
;)


ps, is the hierarchy tab on site aggregated to include the dual and or the 3090 ti?
Click to expand...
Congratulations on obtaining two 3090 TI.

The amount of ram has been updated (we don't have a multi-CPU system to test this issue).

We will add multi-GPU configuration into the Hierarchy.

Thank you!
 
@Tellusim I just downloaded the GravityMark benchmark but I'm not able to configure it properly! Opening the Browser.exe gives me this:

IPOntk3.jpg


How do I configure the benchmark properly? I am running 2x 3090 Ti in SLI (NVLink).
 
ThirtyIR said:
@Tellusim I just downloaded the GravityMark benchmark but I'm not able to configure it properly! Opening the Browser.exe gives me this:



How do I configure the benchmark properly? I am running 2x 3090 Ti in SLI (NVLink).
Click to expand...

This is something new. We will investigate it. Thanks.

You can run the benchmark from the command line.
There are many batch files with various parameters inside the installation directory.
run_fullscreen_vk_rt_cu.bat is for NVLink.
 
Hello Thirty

leave that part running and use the web interface to configure the session

select AFR
And
Let it run for the full three minutes
 
Tellusim said:
This is something new. We will investigate it. Thanks.

You can run the benchmark from the command line.
There are many batch files with various parameters inside the installation directory.
run_fullscreen_vk_rt_cu.bat is for NVLink.
Click to expand...

Okay, will try that. Thank you.

venturi said:
Hello Thirty

leave that part running and use the web interface to configure the session

select AFR
And
Let it run for the full three minutes
Click to expand...

Hi Jay,

I can't get the web interface to show up - i.e. clicking on the 'Browser.exe' file shows that error I posted before. I've tried both Chrome & Brave browsers to no avail.
 
ThirtyIR,

The run_fullscreen_vk_rt_cu.bat file is located inside C:\Program Files\GravityMark\ directory
 
ad in the browser exe shortcut this:

D:\Gravity\bin\Browser.exe -root browser/ ../browser.zip

where D is the drive where it was installed
 
@Tellusim @venturi thank you guys! Was finally able to run the benchmark!

How do I enter my 'name' to submit to the benchmark? I hit 'Send' at the end but not sure if it came through and was submitted correctly(?).

Anyway, here are my results for two runs - one without RT and one with RT:

QPLZMdy.jpg


rqT4UTp.jpg
 
ThirtyIR said:
@Tellusim @venturi thank you guys! Was finally able to run the benchmark!

How do I enter my 'name' to submit to the benchmark? I hit 'Send' at the end but not sure if it came through and was submitted correctly(?).

Anyway, here are my results for two runs - one without RT and one with RT:
Click to expand...

Your welcome! Thanks for the cool results.
 
The new beta version with compute shader rasterization, mesh shader rendering, and LOD bias option is ready.

Compute shader rasterization is twice faster than hardware rasterization on high polygonal modes (LOD Bias).
There are some visual artifacts on AMD GPUs, but we are working on this issue right now.
CS rasterization scales much better on AMD.

GravityMark Leaderboard

GravityMark GPU Benchmark
gravitymark.tellusim.com gravitymark.tellusim.com
 
The benchmark becomes better and better :)
I have one question, why the read temperature from the benchmark for Radeon 6000 is not the GPU Junk temp (hot spot)?
Because it is the most important :)

1658569722909.png
 
AVATARAT said:
The benchmark becomes better and better :)
I have one question, why the read temperature from the benchmark for Radeon 6000 is not the GPU Junk temp (hot spot)?
Because it is the most important :)
Click to expand...

Thanks! The AMD temperature sensor has been updated to the hot spot.
 
Finally, the visual artifacts on AMD GPUs in CS and RT modes have been fixed in the 1.67b version.
 
Last edited:
You must log in or register to reply here.
Back
Top