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

GravityMark v1.87 GPU Benchmark

No.3 for OpenGL, default settings @ 1440p on the GravityMark leaderboard.
GM_1 - Copy.JPG
 
So far Vulkan seems to be the most efficient API. And the 6600XT dominates efficiency rankings across the board. I'm quite surprised that the more powerful GPUs aren't more efficient in higher resolutions. AMD really nailed perf/watt with Navi 23. Intel's A380 looks more efficient than Ampere, even in 4K, and Apple's M1 is in a league of its own.

And something I discovered by chance. You don't actually need a display with a matching native resolution to run the hi-res benchmarks. Give it a try!
 
Last edited:
Last edited:
hi could you correct the ram report to be 1.5 TB, I think it may not see across two nodes. THX
Fixed, cool result :)

PCIe Gen4.0 would provide a better score. Unfortunately, NVLink is not an option anymore.
 
Fixed, cool result :)

PCIe Gen4.0 would provide a better score. Unfortunately, NVLink is not an option anymore.
thank you sir, and you're right - but I'm seeing what I can get done with the auto diff sli, marginal... bit of a lie.. painful painful process.

I tried one of the cards in a pci 4.0 mobo, but it was only 1-2% faster, still, that is something
 
Hate to ask , I did some optimizations and reran the bench at 2x and 4K, it created some duplicate entries on the same hardware due to a typo, could you remove the duplicate entries? Thank you
 
Hate to ask , I did some optimizations and reran the bench at 2x and 4K, it created some duplicate entries on the same hardware due to a typo, could you remove the duplicate entries? Thank you
Merged.

We returned the original temporal anti-aliasing into the benchmark because the latest AA from the engine has a negative impact on benchmark performance.
This version will be available after testing. Maybe it helps to get 100k at 4k.
 
Merged.

We returned the original temporal anti-aliasing into the benchmark because the latest AA from the engine has a negative impact on benchmark performance.
This version will be available after testing. Maybe it helps to get 100k at 4k.
Sweet! Look forward to it and thank you!
 
Sweet! Look forward to it and thank you!
Version 1.76 is ready. Temporal AA is the same as in the first benchmark release. So the performance should be restored.
The secret command line parameter "-nosensors 1" completely disables HW sensors which should improve the overall score by a little fraction.
 
hello sir, I'm getting the following error whenever I try to run it with the commands/switches, even when VK is selected:

Untitled.jpg


however when browser is used it works fine,
 
1.76 run from me

2022-11-10 14.45.46 gravitymark.tellusim.com 9c468fd2d8ae.jpg

one question though why does the max shown not equal the 200,000 Asteroids selected I quite often see over 300K as shown
 
Last edited:
I uploaded mine and spelled my name wrong lol :laugh:
 
version 173 runs with the script
version 176 gives me the error


Untitled.jpg


this is the script I use for 173:

GravityMark.exe -temporal 1 -fullscreen 1 -screen 0 -sensors 0 -fps 0 -info 0 -benchmark 1 -alternate 1 -cuda 1 -vk -mode 4k

in 176, the script generates this the error listed above
running it with the -nosensors 1 does not solve the issue.

I did increase my score on the 173 edition, so close to 100k on the 4k bench, so close....

here are the shots (the mem is only half seen due top nodes)

at 2k and then at 4k on 173, please note how low my temps remain on the air-cooled system as well as how dense it is. 46'C on the 2k run and 49'C on the 4k run, resting temps for both 4096 is about 26'C

Please not how stable temps, frequency, and utilization remains through out the 4k test on both cards

This is in an SFF case (Themaltake P1)

Pics and build are available at:


2k.jpg

4k.jpg
 
Last edited:
use this command instead -nosensors 1
Thank you, but I did, I get the same error when I use:

GravityMark.exe -temporal 1 -fullscreen 1 -screen 0 -nosensors 1 -benchmark 1 -alternate 1 -cuda 1 -vk -mode 4k
 
Weird error. We cannot reproduce it on our side for now. Digging...
thank you, I may have done something different on my end, but I can't find it yet.
If I use the browser it works fine. I'll mess with it more over the weekend.
on ver.173: 99k.... so close.
 
thank you, I may have done something different on my end, but I can't find it yet.
If I use the browser it works fine. I'll mess with it more over the weekend.
on ver.173: 99k.... so close.
The symbol is present in the installed library:
21BE ?getDevice@D3D12Context@Tellusim@@QEBAPEAUID3D12Device@@XZ

Could you please try to reinstall the benchmark?
 
The symbol is present in the installed library:
21BE ?getDevice@D3D12Context@Tellusim@@QEBAPEAUID3D12Device@@XZ

Could you please try to reinstall the benchmark?
will do sir
 
Back
Top