New World, now officially launched, still damaging cards.

[Falkentyne]

all cards that blew up had asynchronous Powerstage/Phase counts.
like buildzoid said that current balancing is very bad/almost impossible when you have an odd number of phases and powerstages.

Furmark gets detected by most GPUs as a powervirus and the GPUs are imediately throttling by up to 500-600 Mhz.
New World seems to be as heavy (transients) as furmark but the GPU boosts as high as it can.

i'd say it's still NVidias fault to allow this kind of VRM in the first place and 2nd making 350W+ Cards.

I think they all had AOZ (Alpha and Omega) power stages too.

 

[Divide Overflow]

My 6800 XT plays this just fine. Guess this is just exposing poor power management design by nVidia's flagship products.

 

[Vayra86]

Ok, I don't fully agree with you, but I clicked Like because I appreciate your reasoned response.

You're right, I don't have reviews or studies to show how cards die or last for a shorter time if run like this. However, it's still a fact that the harder you push something, then the sooner it wears out, eg a bearing fan, especially a sleeved one, regardless of specific factors. It just is. Also, that noise indicates that all the other components are stressed too and greater heat and current does have specific wear characteristics on electronic components. Think about electromigration in CPUs when overclocked hard and overvolted which eventually kills them. It still happens at stock settings, but much less and takes longer.

Put all this together with the irritating noise of coil whine and it seems sensible not to push a card to this point, especially not for any length of time. Don't forget that these things are built to a price, which will add weaknesses. Do you really wanna pay over a grand for a graphics card? Oh hang on, we already do...

Thinking about it, remember how NVIDIA put out a warning regarding FurMark causing physical damage to their cards? I can't remember offhand if it was due to the heat, current, or various factors, but it was all over the tech press at the time. NVIDIA added a FurMark throttle to their drivers to avoid that damage and users bypass it at their peril. I've run it for a few minutes on every recent graphics card I bought, because they've got those big, efficient and silent coolers on them and yet the throttling is still quite marked on those as the card is kept to within its power limit. Even then, I only run it for a few minutes to test the effectiveness of the cooler for heat and noise, plus to stress it a little to see if it fails due to a hardware fault and then likely never again on that card.

Very much appreciate these arguments too. Good points

 

[Dredi]

i think this is almost comparable to a game acting like furmark and breaking a card

Even in furmark’s case, the SW does not break anything. The card is just built bad if it breaks. If there were some things that the cards can do, but the manufacturer does not want you to perform, it should be written in some SW implementation notes somewhere, but that is not the case. If a card breaks doing some calculations that the manufacturer has stated that it supports, the failure is purely the fault of the manufacturer.

 

[pavle]

I think it's nvidia's fault only because of their historically high "Temp Target", "Max Temp Target", slow fan curve (response wise) and quite high power targets and mediocre electronic components with little to no reserve left used in their reference designs. Of course most partners follow the reference, it's the easiest/cheapest thing to do.
I have always balanced all my cards through BIOS (I call them my low_power BIOS) or it can be done in OS with various programs to simply make the cards last long/er.

 

[delshay]

Borderlands 2 title menu. Vsync off. I've seen north of 1000 fps there.

Counterstrike. In case anybody thought stellar FPS is a novel thing...

UT'99, Quake 3... any older shooter really.

This is a old screenshot of "Close To The Sun"

GFX Card: R9 Nano

Settings: Massive Underclock/Undervolt with no other changes to the Radeon Software.
Game Settings: Everything turned up to the max @1080p ..NOTE: Vsync is "enabled"

You can clearly see card is hitting 100% load. If the card was running at default settings it would hit max temperature at a much faster rate.. Temperature in screenshot is incorrect, it gets hotter than this with underclock/undervolt & if I remember correctly the card still hits maximum temperature. Someone out there should have this game as I believe EPIC gave this game away free some time ago. I need someone to verify if their are getting the same 100% load at the game menu. Also note this game is DRM free, just go to the game folder.

 

[Vayra86]

This is a old screenshot of "Close To The Sun"

GFX Card: R9 Nano

Settings: Massive Underclock/Undervolt with no other changes to the Radeon Software.
Game Settings: Everything turned up to the max @1080p ..NOTE: Vsync is "enabled"

You can clearly see card is hitting 100% load. If the card was running at default settings it would hit max temperature at a much faster rate.. Temperature in screenshot is incorrect, it gets hotter than this with underclock/undervolt & if I remember correctly the card still hits maximum temperature. Someone out there should have this game as I believe EPIC gave this game away free some time ago. I need someone to verify if their are getting the same 100% load at the game menu. Also note this game is DRM free, just go to the game folder.

I don't have the game but how is this even remotely comparable? Not only is the R9 Fury an oddball card with different memory and overclocking was never a thing on it, but the menu isn't even running north of the Vsync cap in FPS.

 

[delshay]

I don't have the game but how is this even remotely comparable? Not only is the R9 Fury an oddball card with different memory and overclocking was never a thing on it, but the menu isn't even running north of the Vsync cap in FPS.

That's what I'm saying. Regardless if the FPS is high or low, overclocked/overvolt or underclocked/undervolt, it still hits 100% load & gets very hot.
This is why I use this game menu to setup my card. It will show strange artefacts around the lights if pushed too far regardless of which direction I push, if I push too hard in any direction.

 

[Vayra86]

That's what I'm saying. Regardless if the FPS is high or low, overclocked/overvolt or underclocked/undervolt, it still hits 100% load & gets very hot.
This is why I use this game menu to setup my card. It will show strange artefacts around the lights if pushed too far regardless of which direction I push, if I push too hard in any direction.

Okay, well... to me that's the same futile exercise as running Furmark which also doesn't relate to a real world use case at all, only to a temperature limit and how to get there. You're not getting any useful info on clocking under regular scenarios. You're literally testing the card's workings in and for that specific game menu, which is an odd one out and does not allow the card to work as it is engineered.

 

[Falkentyne]

It has to do with shader load, not FPS.
FPS just makes the problem worse.

Path of Exile with Global Illumination Shadows=Ultra will probably kill these bad cards even faster than New World (if shaders are to blame), but it's possible other parts of the card may not be used the same way as New World does.

 

[InVasMani]

I suspects it's tied to the transients along with 0.1% frame rates essentially a perfect storm cocktail of clock skew and boost with power delivery at low latency. Basically you start with PLL clock skew of both CPU/GPU which both boost clock speeds. Also you throw in a 2% helping from SSC clock skew not much yet GPU's are packing more and more shaders and performance at higher clock speeds. That 2% SSC is effectively a overclock above the expected 100MHz clock rate. That 2% of 100MHz adjusted for 0.1% frame rates with all the clock boosting going on and on a higher end GPU is a much bigger deal than a old janky Kelpler GPU. RTX 3080 pushes 489W at 1ms on a FE no less.

Now just imagine the RTX 3090/RTX 3090Ti not to mention the AIB card models pushing things further on those and the RTX 3080 in regard to reference models with higher power limits and clock speeds. From what I heard cards are dying even with frame rate caps in New World funny how that works here we thought it was the 9000+ FPS or whatever that was reported which is it!? Now what happens when you cap the frame rate!? The GPU runs cooler and GPU can boost higher or for longer. It'll also finish more frame and precede to idle for a bit while it waits on the CPU. Look what happens at <1ms to wattage though. Now sure the PSU can handle a lot if built well and has a lot of protection built in as well. What about the GPU power delivery though? Perhaps in a menu screen churning a high amount of 0.1% frame rates for a sustained period!? Isn't power limits, boosts, and various intricacies of clock skew all part of the problem? Is it fair to blame the developers because of issues like this or is this something hardware manufacturers needs to address better. If you look at the RTX 3080 graph from 10ms down to 1ms you shave off over 100w that's more than the TDP of a lot of CPU's.