News New FurMark Version 2.0 Arrives After 16 Years

[Admin]

Infamous GPU stress testing app update is in final beta stage, and version 2.0 will hit general release shortly.

New FurMark Version 2.0 Arrives After 16 Years : Read more

 

[brandonjclark]

I didn't even realize the 1.x was still relevant.


*edit:SP

 

[SyCoREAPER]

Can't wait for my new GPU to catch fire! lol

The original FurMark was punishing as hell, can't even imagine 2.0

 

[bit_user]

I didn't even realize the 1.x was still relevant.

The thing about Furmark is that it seems both highly-parallel and 100% compute-bound. So, it's basically guaranteed to find a GPU's peak temperature, since there's no bottlenecking on memory, occupancy, or anything else to throttle it.

Make	Model	GFLOPS​	GB/s​	FLO/B​
AMD	RX 6950 XT	38707​	576​	67.2​
AMD	RX 7900 XTX	46690​	960​	48.6​
Intel	UHD 770	825​	90​	9.2​
Intel	A770	17203​	560​	30.7​
Nvidia	RTX 3090 Ti	33500​	1008​	33.2​
Nvidia	RTX 4090	73100​	1008​	72.5​

I used base clocks, since we're talking about an all-core workload. Some GPUs could well throttle even below base. The one exception was the UHD 770, where I believe it will run at peak if there's not also a heavy CPU load.

Note how bandwidth-limited GPUs are! I computed fp32 ops per byte, but a single fp32 is actually 4 bytes! This is probably why increasing L2/L3 has helped AMD and Nvidia so much, over the past couple generations. You can probably tell that if you alleviate any bandwidth limitations and have enough threads, these GPUs could easily burn a lot more power than on typical games & apps.

It also should help explain why datacenter GPUs have like 3x the memory bandwidth of high-end desktop models, but not a lot more raw compute, as in HPC type workloads you're more often doing things like reading 2 values and outputting a 3rd.

 

[bit_user]

The original FurMark was punishing as hell, can't even imagine 2.0

I'll bet the main difference is Vulkan support. Depending on how well they implemented the OpenGL version, there might not be much performance difference, between it and the Vulkan path. If there is, then you're right that Vulkan could pose a more serious fire risk!

 

[Amdlova]

I didn't even realize the 1.x was still relevant.


*edit:SP

People use to fry the graphics card! My first two days with the 6700XT have used the furmark to burn in the card and loose some coil whine in process be cause you can set the resolution change the fps behavior bur the power are Aways in max My graphics giver about little more than 175w when gaming in furmark way beyond the limit!

FurMark2 Score

FurMark benchmark scores database - OpenGL / Vulkan

www.gpumagick.com

Oc score

FurMark Scores

FurMark benchmark scores database - OpenGL / Vulkan

www.gpumagick.com

 

[aindriu80]

the old version was 32bit, kind of ancient now, is there Linux support?

 

[artk2219]

I didn't even realize the 1.x was still relevant.


*edit:SP

Most definitely, it still heavily punishes a graphics card, more than once ive found a defective one in a build with it. The card would work fine on Unigine Superposition pretty much indefinitely, but as soon as it ran furmark it would crap its pants, letting me know that its not 100% stable.

 

[fireaza]

...How is it that a GPU stress-test from 2007 can still be relevant anyway? If it was designed to stress GPUs from 2007, then surely it's a cakewalk for 2023 GPUs? Or if it's still stressful for 2023 GPUs, then how on Earth did 2007 GPUs not instantly catch fire while rendering at 1 frame per hour?

 

[bit_user]

...How is it that a GPU stress-test from 2007 can still be relevant anyway?

It just does a lot of shader arithmetic. No matter how fast a GPU is, a benchmark like that can saturate it - although, you might get absurd framerates like 900 fps.

If it was designed to stress GPUs from 2007, then surely it's a cakewalk for 2023 GPUs?

You could also say that about cryptomining. Newer GPUs do it faster, but a GPU can never do it too fast.

if it's still stressful for 2023 GPUs, then how on Earth did 2007 GPUs not instantly catch fire while rendering at 1 frame per hour?

Because GPUs typically ramp up their fans and then throttle their clock speeds, as their temperature increases or if their power limit is exceeded.