FurMark 2.0 is coming for more GPU stress testing and benchmarking

[Jimmy2x]

Something to look forward to: FurMark has been pushing the limits of graphics cards since the 2000s. A go-to tool for overclocking enthusiasts, last month developer Geeks3D announced a new beta release of the software's successor. As always, FurMark 2.0 is expected to be lightweight, furry, and brutally effective.

FurMark developer Geeks3D will release the latest version of the popular benchmarking software later this month. The public release should follow the latest beta milestone, v2.0.10, which dropped via the Geeks3D Discord channel on August 27.

Geeks3D released the first beta of FurMark 2.0 in December 2022. Since then, the developer has addressed several reported bugs, expanded GPU support, expanded resolution support, and enhanced data reporting outputs. Version 2.0.10's features and bug fixes include:

• Fixed Radeon RX 6850M XT name (XT was missing).
• Added support of AMD Radeon PRO W7900, PRO W7800, PRO W7600 and PRO W7500.
• Added support of AMD Radeon RX 7900 GRE.
• Added support of NVIDIA GeForce RTX 4060 Ti 16GB
• Added support of Intel Arc A570M, Arc A530M, Arc Pro A60M and A30M.
• (Windows) updated with GPU Shark2 2.0.12
• Updated with GeeXLab 0.53.0 libs

FurMark 2.0 still provides users with a familiar, simple user interface to initiate predefined or custom test runs. Users can select the parameters of their test by choosing the OpenGL or Vulkan API, desired resolution, fullscreen or windowed mode, and whether the tool should display the on-screen interface. Users can also launch GPU-Z and GPU Shark monitoring tools directly from the application's launch window.

Since no benchmark software is complete without bragging rights, FurMark provides users with a leaderboard for submitting and comparing results with other users and hardware configurations.

The current rankings feature a wide range of scores for various GPUs under the software's presets, which may indicate further scoring refinement is required before the feature goes live in the upcoming public release. If you want a head start, jump on to Geeks3D's Discord server and navigate to the FurMark 2 thread. You can find links to download version 2.0.10 here.

Permalink to story.

https://www.techspot.com/news/100175-geeks3d-release-furmark-20-gpu-benchmark-later-month.html

 

[MaestroIT]

Will Furmark v2.0 melt the cpu core and desolder it from the main board? /s

 

[Lew Zealand]

Public Service Announcement

Don't use Furmark to test your GPU as it presents an unrealistic load and some GPUs actively clock themselves down to avoid this load. Superposition is free and uses a reasonable, heavy DX12 load and parts of 3DMark are free. Valley is an OK DX11 test for older/lower end GPUs.

Just avoid Furmark.

 

[Vanderlinde]

Public Service Announcement

Don't use Furmark to test your GPU as it presents an unrealistic load and some GPUs actively clock themselves down to avoid this load. Superposition is free and uses a reasonable, heavy DX12 load and parts of 3DMark are free. Valley is an OK DX11 test for older/lower end GPUs.

Just avoid Furmark.

Furmark is perfect for testing cooling in worst conditions.

Only thing it does it put a high and often unrealistic load onto your GPU - but at least guarantees that the thing will be working in the event of a OC.

I used it quite some times. Esp when I was ramping over 350W through a RX580 over a single 8 pin wire. Not a sweat.

All these cards are tested using furmark before they even leave the factory.

 

[Lew Zealand]

Furmark is perfect for testing cooling in worst conditions.

Only thing it does it put a high and often unrealistic load onto your GPU - but at least guarantees that the thing will be working in the event of a OC.

I used it quite some times. Esp when I was ramping over 350W through a RX580 over a single 8 pin wire. Not a sweat.

All these cards are tested using furmark before they even leave the factory.

If you look at the Furmark results that TechPowerUp shows when compared to a gaming load, you will see Furmark voltage spikes on many cards which are notably higher than their power limit and those spikes are much smaller or absent in the gaming load.

While that should guarantee that stability in Furmark will mean stability in games, it also means that you could get more performance with a higher overclock or power limit setting and still be stable in games compared to what Furmark testing indicates.

 

[guillejp]

Furry you say?

 

[hahahanoobs]

Furmark is perfect for testing cooling in worst conditions.

Only thing it does it put a high and often unrealistic load onto your GPU - but at least guarantees that the thing will be working in the event of a OC.

All these cards are tested using furmark before they even leave the factory.

There is no one app that can guarantee stability. You could easily run Furmark overnight and still crash when watching a youtube video for example.

Can I get a source on factories using Furmark for QA?

 

[Vanderlinde]

If you look at the Furmark results that TechPowerUp shows when compared to a gaming load, you will see Furmark voltage spikes on many cards which are notably higher than their power limit and those spikes are much smaller or absent in the gaming load.

While that should guarantee that stability in Furmark will mean stability in games, it also means that you could get more performance with a higher overclock or power limit setting and still be stable in games compared to what Furmark testing indicates.

Yes, because the graphical load applied is a dummy load, and once all brakes are lose on the GPU it will run pretty much to it's limit. That's why it's designed in the first place. You can do 20 boring runs of your favourite bench application or do one quick testing within minutes to know if your card is stable or not stable.

I've used it on Polaris cards to find the max limit the card can handle or take. Well that was 350W pulled over the 8 Pin connector and hey, it survived because it was build with quality components. Exactly as it should. And it tested my cooling intentionally knowing that it can cool the GPU at 300W avg below 60 degree. Something I could not reproduce while using another benchmark tool. And because of that in gaming I could dial fan speeds in lower.

Same as with Linpack - very unreal CPU stress tool which can easily raise temps with another 20 degree. Perfect for testing cooling. Not perfect for running on chips that are pushed beyond as degradation is real and with tools like this, you can degrade your chip in days. Used it extensively on a FX clocked at 5Ghz to iron out any imperfections in my back then, OC. If it can run linpack for at least 5 minutes your OC is as close as solid.

There is no one app that can guarantee stability. You could easily run Furmark overnight and still crash when watching a youtube video for example.

Can I get a source on factories using Furmark for QA?

Sure.

GPU Factory tour by Derbauer. Thats furmark for you. As part of the quality testing before cards leave the factory. So in a way cards are already burned in to lower the amount of DOA's. These 10% of cards are tested 24 hours on a extreme burn test.

I mean - many people forget how overbuild VRM's on motherboards or graphics cards are. You can easily push these 2.5x higher then the avg board power. They do that to provide a solid voltage in any possible situation you can think of. They also bench or run cards intentionally in a 60 degree oven while running furmark.

When you play around with furmark, the actual hardcore setting is listed in settings, it's 2 options you tap and you got a power hog to your benefit. Just mind you that, you can burn in or make VRM's explode with the high currents GPU's can pull at that point. So always make sure the cooling is EXCELLENT before you even try.

Cards can easily run into 110 degree territory with that hidden setting uncorked in Furmark.

And yes - some OC's do work in furmark, some OC's do not work in gaming or streaming. Those things are just different aspects from the GPU your taxing - and once running beyond spec you need to verify and test. A good OC is one that works under all conditions. Not half half.

Ive done various OC's in my time, even a 100% GPU overclock coming from a 2600Pro Radeon over 1300Mhz (stock 600) - replaced all caps, added voltage boosters, ran with a compressor to keep it below -40 degree etc. Was a fun time. But hardware now is pretty much pushed to it's limits already. They have figured out ways to maximize silicon on a automated level - you get percentages now and no longer jumps.

 

[hahahanoobs]

Yes, because the graphical load applied is a dummy load, and once all brakes are lose on the GPU it will run pretty much to it's limit. That's why it's designed in the first place. You can do 20 boring runs of your favourite bench application or do one quick testing within minutes to know if your card is stable or not stable.

I've used it on Polaris cards to find the max limit the card can handle or take. Well that was 350W pulled over the 8 Pin connector and hey, it survived because it was build with quality components. Exactly as it should. And it tested my cooling intentionally knowing that it can cool the GPU at 300W avg below 60 degree. Something I could not reproduce while using another benchmark tool. And because of that in gaming I could dial fan speeds in lower.

Same as with Linpack - very unreal CPU stress tool which can easily raise temps with another 20 degree. Perfect for testing cooling. Not perfect for running on chips that are pushed beyond as degradation is real and with tools like this, you can degrade your chip in days. Used it extensively on a FX clocked at 5Ghz to iron out any imperfections in my back then, OC. If it can run linpack for at least 5 minutes your OC is as close as solid.



Sure.

GPU Factory tour by Derbauer. Thats furmark for you. As part of the quality testing before cards leave the factory. So in a way cards are already burned in to lower the amount of DOA's. These 10% of cards are tested 24 hours on a extreme burn test.

I mean - many people forget how overbuild VRM's on motherboards or graphics cards are. You can easily push these 2.5x higher then the avg board power. They do that to provide a solid voltage in any possible situation you can think of. They also bench or run cards intentionally in a 60 degree oven while running furmark.

When you play around with furmark, the actual hardcore setting is listed in settings, it's 2 options you tap and you got a power hog to your benefit. Just mind you that, you can burn in or make VRM's explode with the high currents GPU's can pull at that point. So always make sure the cooling is EXCELLENT before you even try.

Cards can easily run into 110 degree territory with that hidden setting uncorked in Furmark.

And yes - some OC's do work in furmark, some OC's do not work in gaming or streaming. Those things are just different aspects from the GPU your taxing - and once running beyond spec you need to verify and test. A good OC is one that works under all conditions. Not half half.

Ive done various OC's in my time, even a 100% GPU overclock coming from a 2600Pro Radeon over 1300Mhz (stock 600) - replaced all caps, added voltage boosters, ran with a compressor to keep it below -40 degree etc. Was a fun time. But hardware now is pretty much pushed to it's limits already. They have figured out ways to maximize silicon on a automated level - you get percentages now and no longer jumps.

That's a hardware stability test, not an OS stability test. Like I said, there is no one test for complete stability.
Furmark is unnecessary for 99.9% of users. GPU VRM's are fine for 99.9% of users.

 

[Vanderlinde]

Most hardware has extreme overbuild VRM's. Like you got motherboards with 12 phase VRM or GPU's with up to 22 phase of VRM's. Did anyone notice ever a server board with only 4 phase vrm's powering 350W powerhogs of CPU's? And how perfectly fine that all is for 24/7 constant usage? Lol.

People are paying so much for unneeded stuff the last 10 years.

 

[neeyik]

Most hardware has extreme overbuild VRM's. Like you got motherboards with 12 phase VRM or GPU's with up to 22 phase of VRM's. Did anyone notice ever a server board with only 4 phase vrm's powering 350W powerhogs of CPU's? And how perfectly fine that all is for 24/7 constant usage? Lol.

The VRMs on server motherboards are designed to be utterly blasted by 16k rpm fans all the time, so it doesn't matter if they're being inefficiently run -- they'll never get to the point that they'll burn out. On consumer-grade boards, the amount of airflow around the VRMs is generally very poor, especially now that lots of people use AIO coolers (which provide little in the way of cooling around the CPU socket).

This is why they have an excess of VRMs -- one, to keep the load of the individual components down, so they won't overheat and two, it provides a modicum of redundancy in case one does decide to go south. They also have more than server boards because they undergo load changes more frequently, so by having a larger bank of capacitors (usually one per VRM), it helps to prevent the core voltages from dropping too low before the high-output drivers can be activated. This last point is why GPUs have so many -- they switch even more frequently than CPUs do.

Of course, this isn't to say that consumer-grade products aren't overly festooned with VRMs (they are!) but as the enthusiast market now expects expensive motherboards and graphics cards to have lots of them, it would hurt the sales of a manufacturer's product if it didn't look like it was up to the task.

 

[Vanderlinde]

The VRMs on server motherboards are designed to be utterly blasted by 16k rpm fans all the time, so it doesn't matter if they're being inefficiently run -- they'll never get to the point that they'll burn out. On consumer-grade boards, the amount of airflow around the VRMs is generally very poor, especially now that lots of people use AIO coolers (which provide little in the way of cooling around the CPU socket).

This is why they have an excess of VRMs -- one, to keep the load of the individual components down, so they won't overheat and two, it provides a modicum of redundancy in case one does decide to go south. They also have more than server boards because they undergo load changes more frequently, so by having a larger bank of capacitors (usually one per VRM), it helps to prevent the core voltages from dropping too low before the high-output drivers can be activated. This last point is why GPUs have so many -- they switch even more frequently than CPUs do.

Of course, this isn't to say that consumer-grade products aren't overly festooned with VRMs (they are!) but as the enthusiast market now expects expensive motherboards and graphics cards to have lots of them, it would hurt the sales of a manufacturer's product if it didn't look like it was up to the task.

Yep - but since AM4 all motherboard vendors are forced to build boards including the 45$ AM4 boards that are and should be capable of running a CPU to it's maximum TDP. So basicly if you buy a utterly high end 5950X 16 core chip with a 144W TDP your guaranteed that your 45$ motherboard will take it and run it perfectly fine.

Your making an assumption that all boards now must have 8 to 16 phase VRM's while the reality is you can get away with just 4. That you need cooling with a VRM is a different story - the more VRM's the less or at least more distributed heat output in terms of energy losses compared to a small 4 phase VRM.

Since motherboards have less and less things to build in, I.e the CPU has pretty much all it needs such as Soc, chipset, PCI-E and such there's less components now. So what you get is additional stuff planted on boards for a premium price. Like there's a majority buying a high end motherboard with so many VRM's that never in their lifetime will be even taxed to 50% of it's duty - unless you start to ramp up with LN2 or you know tricks to increase the VRM Switching frequency etc...

Other then that - the market has bin saturated with stuff people don't even understand. Oh wait so this has 16 VRM's so it must be good right?! It's like the AM3+ motherboards all over again - a sabertooth blowing up because it was advertised as 8, 12 or even 16 phase VRM but uncapable of running a FX-9590 because the specsheet told something completely different.

I worked with both servers and consumer hardware - servers can operate on just 4 phase VRM's and it's not even about the cooling. It's just how efficient those VRM's work. People never needed a 16 to even 24 phase VRM to run stock settings in the first place.