GravityMark v1.87 GPU Benchmark

[Tellusim]

new high score

68,245

not sure why the app only sees half my ram, but I suppose that doesn't affect the score

Cool config Yep, the amount of detected memory is not affecting the score.

 

[venturi]

slight improvement
70,437

 

[mrthanhnguyen]

So Vulkan or dx12 does not matter coz I see your leaderboard is mixed with Vulkan and DX12.

 

[Hugis]

wow my old girl aint doing so well

 

[Tellusim]

So Vulkan or dx12 does not matter coz I see your leaderboard is mixed with Vulkan and DX12.

Yes, Vulkan and D3D12 are working with the same performance on the latest GPUs.
Previous Nvidia generation is faster with D3D11 because of the driver issue.
Nobody cares about OpenGL performance anymore.

Metal API with AMD GPU is 3 times slower than Vulkan.

wow my old girl aint doing so well

Direct3D11 should provide a performance boost depending on the CPU.
GTX 1060 is 20% faster under D3D11 than Vulkan, even with an outdated CPU.

 

[Hugis]

Yes, Vulkan and D3D12 are working with the same performance on the latest GPUs.
Previous Nvidia generation is faster with D3D11 because of the driver issue.
Nobody cares about OpenGL performance anymore.

Metal API with AMD GPU is 3 times slower than Vulkan.


Direct3D11 should provide a performance boost depending on the CPU.
GTX 1060 is 20% faster under D3D11 than Vulkan, even with an outdated CPU.

with d3d11 i get this

 

[Tellusim]

with d3d11 i get this
View attachment 210515

Batch launchers from C:\Program Files\GravityMark\ directory should resolve this problem. What about windowed mode?

 

[Hugis]

can confirm batch commands work for d3d11 and it made the leader board in 3rd

 

[venturi]

So Vulkan or dx12 does not matter coz I see your leaderboard is mixed with Vulkan and DX12.

I found dx12 to be slower than Vulkan. My Vulkan score is 70,000 +
On equal settings
It is a SFF sized PC after all so I haven't done any overclocking or water-cooling, just optimizations.

I reran it recently and got even higher than the top score (on Vulkan) but forgot to upload it, for the current post I forgot to turn of several apps and a VM in the background. I'll have to do a more of a pure test next time.

new score is 70,664

thx
J

side note/ observations:
*while asteroids count is set to 200,000 for the benchmark, the app reported 371,000+/- visible during the run.
*app only sees half my system ram (1.5TB), I think it only penetrates 1 cpu node/numa vs both cpus.
*app only sees individual ram from 1 graphics card (24GB) not both (48GB).
* I avoided posting the run under every conceivable screen resolution, and just picked the more common one on the leaderboard (2560x1440) even though everything I do (app/games) is usually 3840x2160, 10bit, max eye candy. It seemed bushwah to post it at every resolution.
*On my configuration Dx12 performs 7%-13% worse than Vulkan (memory optimizations/load balancing?)
*I have no overclocking, - just common sense hardware, bios, and OS optimizations.

 

[Tellusim]

side note/ observations:
*while asteroids count is set to 200,000 for the benchmark, the app reported 371,000+/- visible during the run.
*app only sees half my system ram (1.5TB), I think it only penetrates 1 cpu node/numa vs both cpus.
*app only sees individual ram from 1 graphics card (24GB) not both (48GB).
* I avoided posting the run under every conceivable screen resolution, and just picked the more common one on the leaderboard (2560x1440) even though everything I do (app/games) is usually 3840x2160, 10bit, max eye candy. It seemed bushwah to post it at every resolution.
*On my configuration Dx12 performs 7%-13% worse than Vulkan (memory optimizations/load balancing?)
*I have no overclocking, - just common sense hardware, bios, and OS optimizations.

1. The number of asteroids is doubling because of shadow map rendering, especially when the sun rays are parallel to the plane of the asteroid.
2. We haven't check multi-CPU configurations. Unfortunately, only half of your system is detected by the benchmark, but we updated it manually
3. Multi-GPU configurations can not double the amount of video memory because there are no shared resources between GPUs. NVLink is not allowing to share resources by using graphical API.
4. The profit of multi-GPU configuration is reducing with resolution increasing. Direct3D12 or Vulkan is not using NVLink, and shared texture is copied through the system memory.
5. A single API call is rendering all asteroids. The performance of this function depends on API, GPU generation, and asteroid count.

 

[venturi]

1. The number of asteroids is doubling because of shadow map rendering, especially when the sun rays are parallel to the plane of the asteroid.
2. We haven't check multi-CPU configurations. Unfortunately, only half of your system is detected by the benchmark, but we updated it manually
3. Multi-GPU configurations can not double the amount of video memory because there are no shared resources between GPUs. NVLink is not allowing to share resources by using graphical API.
4. The profit of multi-GPU configuration is reducing with resolution increasing. Direct3D12 or Vulkan is not using NVLink, and shared texture is copied through the system memory.
5. A single API call is rendering all asteroids. The performance of this function depends on API, GPU generation, and asteroid count.

Thank you, I appreciate the core product you provide as well as the benchmark you have developed.

While I play games on the side, the core function of my home rig is to further my doctorate/dissertation in Ai (deep learning) in diagnostic imaging, assist radiology impressions and findings. Multiple GPUs are a must for what I do. My prior rig had 4 Titan Vs and it was also struggling. I can use up all the 1.5TB of ram quite easily on this rig with a single simple algorithm for a single pass in a training data set. I use benchmarks such as you provided to look for opportunities to optimize. I run Ubuntu and Windows server - Some apps run better in linux, but most games run better in windows. Hence the dual boot.

Thank you again.

J

 

[johnspack]

Setting up this laptop for a friend... it's partially using the vega7 onboard too, not sure how to disable that yet.... here's a ryzen lappy with a mobile 3070:

 

[Tellusim]

Setting up this laptop for a friend... it's partially using the vega7 onboard too, not sure how to disable that yet.... here's a ryzen lappy with a mobile 3070:

It's most likely that GeForce is working in headless mode on all AMD+Nvidia laptops, and Radeon is constantly reading shared memory.
AMD+Radeon laptops have zero utilization of integrated GPU when discrete GPU is working (under Windows only).

 

[johnspack]

Yeah, I'm unable to disable the vega... it won't even run the bench if I do. And it's showing 20% usage on the vega while benching.
The vega graphics do heat up during the bench... but not sure if that's just the cpu or what....

 

[Tellusim]

The new macOS version is available for download - 1.5x performance boost and fully GPU-driver rendering on macOS Monterey. It is a fair test for Apple M1 Max.

 

[QuietBob]

1080p full screen with 6600XT @ stock + SAM:

 

[venturi]

Hi
just posted a new 71,424 on the default benchmark, can you fix the detail report again? -for some reason it only sees half the ram (768 out of 1.5TB), might be from not being able see past or through the UPI on 2 sockets.

thanks

JV

 

[Tellusim]

Hi
just posted a new 71,424 on the default benchmark, can you fix the detail report again? -for some reason it only sees half the ram (768 out of 1.5TB), might be from not being able see past or through the UPI on 2 sockets.

thanks

JV

Hi, it is fixed. We will solve that issue in the upcoming updates.

 

[freeagent]

This is mine, CPU @ PBO +150, mems at 3600 14-14-14-34, GPU boosting to 2100/1312

 

[venturi]

Hi, it is fixed. We will solve that issue in the upcoming updates.

Awesome and thank you

 

[Tellusim]

It's time to overcool your phones:

‎GravityMark GPU Benchmark

‎GravityMark GPU Benchmark demonstrates the capabilities of modern GPUs by rendering an enormous quantity of objects in real-time, utilizing GPU acceleration. We avoid typical CPU-based performance bottlenecks by delegating the entirety of scene management and rendering steps exclusively to the...

apps.apple.com

Android version is coming soon.

 

[johnspack]

Heh, hope it will run on my old S7. Will be a horrible score, but still fun to see none the less!

 

[Tellusim]

Heh, hope it will run on my old S7. Will be a horrible score, but still fun to see none the less!

Android version is available at https://gravitymark.tellusim.com/

Adreno 650+ with 530+ driver is required. Unfortunately, it's difficult or even impossible to update drivers on Android devices.
Samsung Galaxy S21 meets that requirements. But all Mali-based phones, including Goggle Pixel 6, can't handle it.
We will try to improve this situation, but it will take some time. So the Android version is available as is.

GravityMark Report

GravityMark GPU Benchmark

gravitymark.tellusim.com

 

[Tellusim]

We have released new updates for Windows and Linux with virtual screen resolution support.
So everybody can run the benchmark even at 8K mode.

 

[Nuckles56]

There's a big difference between DX11 vs Vulkan with the 1080ti alright, about 50% more performance with DX11, running stock settings