GravityMark v1.87 GPU Benchmark

[Tellusim]

I uploaded mine and spelled my name wrong lol

Fixed:

GravityMark Report

GravityMark GPU Benchmark

gravitymark.tellusim.com

will do sir

The "Entry Point Not Found" error is caused by the library conflict.
Somehow the new GravityMark.exe (1.76) uses the Tellusim_x64.dll from the previous (1.73) installation.

 

[venturi]

Fixed:

GravityMark Report

GravityMark GPU Benchmark

gravitymark.tellusim.com

The "Entry Point Not Found" error is caused by the library conflict.
Somehow the new GravityMark.exe (1.76) uses the Tellusim_x64.dll from the previous (1.73) installation.

View attachment 269362

Interesting, should I manually remove it or it a fixed asset ?
I have uninstalled a reinstalled, behavior unchanged

 

[Tellusim]

1.76 run from me

one question though why does the max shown not equal the 200,000 Asteroids selected I quite often see over 300K as shown

Default rasterization and compute rasterization modes use shadow maps. That requires an additional geometry pass and increases the overall number of instances.
Raytracing rendering mode doesn't require it and the geometry count is always constant.

Interesting, should I manually remove it or it a fixed asset ?
I have uninstalled a reinstalled, behavior unchanged

The only possible scenario is if there is an old Tellusim_x64.dll library with higher priority.

 

[Athlonite]

here's a VK CS run that makes my GPU use alot more power

It actually got as high as 213W which made the RX6800 sweat a bit at 72~75c upto 81c near the end

that puts me in 17th right under a 3080Ti

Actually surprised that kaisersolo's 5800X3D didn't spank my system

 

[venturi]

The only possible scenario is if there is an old Tellusim_x64.dll library with higher priority.

UPDATE, I believe it is solved

 

[HD64G]

My recently aquired Sapphire Pulse RX6750XT was tested and below are the Vulkan API's results in 4 different configs/profiles mainly trying to optimize for efficiency and to have a low power tune for not demanding games for my 1080P monitor.

34,3K@Stock (2,7GHz@1,2V with 173W average die power draw)

34,4@default UV setting from the driver suite (2,7GHz@1,175V with 168W average die power draw)

33,4K@manual UV setting with fast timings for the VRAM (2,6GHz@1,175V with 138W average die power draw)

27,5K@manual low power setting (2GHz@1V with 91W average die power draw)

This last one resulted also in 1810 efficiency points
https://gravitymark.tellusim.com/report/?id=431d829625c3fc3f05425dabfde55e92e9d209a9

 

[QuietBob]

This last one resulted also in 1810 efficiency points

Well done! Time to tweak my 6600XT

 

[HD64G]

Another undervolting tune got almost 1800 efficiency points @1080P and more than 1500 @1440P while maintaining 90% of the stock performance. I set clocks at 2300MHz and voltage at 1,125V that in real ended at just 0,97V with max 110W of power draw and 30K@1080P and ~27K@1440P points in the benchmark.

 

[3x0]

5600X with PBO +200MHz, 32GB 3600CL14 Dual Rank RAM, 6600XT OCed to 2833/2933 Core, 2250 VRAM (18Gbps effective), 175W PPT, FullHD Vulkan: https://gravitymark.tellusim.com/report/?id=f267c313a0b205d91aae309d50d6bcf487dac646

Tried pushing the GPU core to 2950/3050 which I know isn't stable 24/7, got this score from the attachment (didn't want to submit the result since I already got the first place for 6600XT's )

 

[HD64G]

A small improvement today because of cold weather (18C inside the house) using the auto-UV setting in the driver suite.

 

[Thimblewad]

Don't know why it won't use more than 200W, when clearly it goes up to 255W in most cases when not CPU limited but oh well. Easily running 2.55 GHz on stock power limit and 1.1 V.

 

[HD64G]

The wattage shown there is the same AMD driver shows which is not the total board power draw. Usually you have to add 20-40W (depending to the GPU) to get the total figure. And the figure the screenshot shows isn't the maximum of this benchmark. Mine reached 210W at one instance but in the end it shows just 187W.

 

[Morgoth]

i can only run 1080p...

 

[Athlonite]

i can only run 1080p...

you could try using AMD's RSR (Radeon Super Resolution) and then run gravitymark at 1400p or 2160p even though it's still outputted in 1080p it is rendered at 1440p or 2160p first giving and approximate score for those resolutions on your system

 

[Tellusim]

The new version 1.78 is ready with:

• Additional command line parameters for batch scripts.
• Correct VK_EXT_mesh_shader bytecode generation.
• Fan Speed sensor visualization.
• OpenXR support for Linux.
• FPS chart in reports.

And finally, multi-GPU Vulkan works with the latest Nvidia drivers on Linux.

 

[mechtech]

Where's all the integrated graphics benchies??

 

[Tellusim]

Where's all the integrated graphics benchies??

Starting there:

GravityMark Leaderboard

GravityMark GPU Benchmark

gravitymark.tellusim.com

 

[venturi]

sir,
I click "send" after the bench, but it never shows on the main leaderboard.
However it does create an internet shorcut in my documents that when I click on that it shows the report, but it just won't populate the actual leaderboard

Suggestions?

 

[Tellusim]

sir,
I click "send" after the bench, but it never shows on the main leaderboard.
However it does create an internet shorcut in my documents that when I click on that it shows the report, but it just won't populate the actual leaderboard

Suggestions?

Antialiasing is not activated. Cuda interop is not required for 4090 because there is no physical bridge between GPUs. An additional Vulkan-Cuda context switch is reducing performance, but the overall bandwidth is the same.

 

[venturi]

ill resubmit with antialiasing

 

[AMF]

 

[Blaeza]

 

[venturi]

Sir
I resubmitted the score with antialiasing, if you could at 2k and 4k and the 1.5Tb ram.

One is on my wife's PC ( the prior scores ) and mine (current scores) from today.
She was the 8280L at 2.6ghz and mine is the 8280 2.7ghz, so they are different builds, but more or less clones, .technically they are both my PCs but the other is the one she uses at her desk


----before I ran into the postings today I was trying to explain an observation:

the 528 series and the 527 series nvidia drivers drops my score 15%

I reverted to the 522.30 and the score returned to "normal".

In both drivers I was using the latest vulkan runtime.

Anyhow, there is a 15% drop in score using the 528 series drivers.


Thank you

J

 

[venturi]

Sir
I resubmitted the score with antialiasing, if you could at 2k and 4k and the 1.5Tb ram.

One is on my wife's PC ( the prior scores ) and mine (current scores) from today.
She was the 8280L at 2.6ghz and mine is the 8280 2.7ghz, so they are different builds, but more or less clones, .technically they are both my PCs but the other is the one she uses at her desk


----before I ran into the postings today I was trying to explain an observation:

the 528 series and the 527 series nvidia drivers drops my score 15%

I reverted to the 522.30 and the score returned to "normal".

In both drivers I was using the latest vulkan runtime.

Anyhow, there is a 15% drop in score using the 528 series drivers.


Thank you

J

good evening sir (at tellusim)

I broke through the 4k 100,000 barrier as well as over 134,000 in 2k.

It seems all drivers after the 522.30 carry a 13-17% penalty.


Can you fix the ram reported on my latest tests, I have 1.5TB of ram, but I think it does not see the ram on the 2nd cpu (768 + 768)

Thank you


and maybe you can look into what has happened on nvidia drivers released after 522.30 (example 528) that hobble Gravitymark performance so much.
I reinstalled the 522.30 one more time just to see the performance return, but have moved back to 528. both instances are using the latest Vulkan runtime

thank you

GravityMark Leaderboard

GravityMark GPU Benchmark

gravitymark.tellusim.com

Jay

 

[Athlonite]

I have 1.5TB of ram, but I think it does not see the ram on the 2nd cpu (768 + 768)

as it seems to not even see a second CPU it's only reporting what it can see which is a single 8280 and it's accompanying ram