News Amazon: Don't Blame New World for GPU Deaths, Blame Card Makers

[-Fran-]

nowhere do i say that it is strictly a software issue. it is both a hardware and software issue and instead of pointing the finger solely at AIBs, amazon needs to own up to its role as the software developer and start working with AIB's to correct the issue. that is where my issue is with this whole situation.

A bit of a history lesson: when K7 was released, so great fanfare and enthusiasm, it had one fatal flaw that was killing the CPUs. The thermal sensors were crap and they were going way over their thermal limit and the built in protections were failing. Software that would run just fine with Intel (of the time) was basically burning Athlons. No software developer at the time had to do anything about their software, because it was understood that it was AMD's (and the motherboard vendors) fault. Well, this may not be 100% accurate, but it was like this.

And this whole kerfuffle is, more or less, the same exact situation.

You can code malicious software or even when developing personal projects that could may as well behave like this particular game (or Furmark). The expectation today is the hardware won't die due to crappy coding. That is the modern understanding of how things work. If they do die, then it's a hardware fault. No second reads. No second guessing the software. Sure, the software exposed a problem, but the software itself IS NOT THE PROBLEM.

Regards.

 

[De_La_BunBun]

I'm running a 2070 not even a 3000 card and my gpu has reached up to 90c with 100% fan spin, but the worst part is that new world straight up isn't giving a single sizzle about their settings. I've ran everything including resolution on the lowest settings and supposedly fps capped at 60 and it still hit those temperatures. Whilst when I've ran the fps cap that's built into msi afterburner to 60 fps max the temps drops basically immediately and the game isn't even reaching those temps when running on highest settings, only sitting at around 80-85c. It's definitely atleast partly new worlds fault for gpus dying.

 

[JamesJones44]

they do. called they listen to the rules of their control program.

look at jayz video to ez example.
his gpu was NOT able to go over 102% power even with a hacked bios...it refused.

yet New World had it over 111%.

what part of the game ignores the gpu's limits don't you understand?

issue is with it taking much more power than ti can handle.

your analogy on cpu is apples to oranges comparison. similar but not same.
a cpu is MUCH better equipped to deal with sudden surges and peaks.

you cant "limit" power in sense you want to as its out of spec and no "failsafe" can stop sudden power spikes. (same way if u have a storm surge your powerlines and you arent usign a proper surge protector.)

Have you ever written software in your life? Do you not understand that you can't just say, hey hardware... work 20% faster.

Again, if hardware makers were this dumb, virus writers would be burning out hardware every 10 seconds. It doesn't work that way.

 

[TJ Hooker]

Game shouldn't be able to run in 4k in windows with DSR off. Yes, the power figures get all the attention but game shouldn't let you run 4k on a monitor that doesn't support it with the features that are designed to let you run , turned off.

The 4k weirdness is tied to the power usage spikes from what I saw in that video of Jay. I don't know what it is asking GPU to do in that situation. Probably running as if monitor is 4k.

Rendering at a higher resolution and then downsampling to the display resolution is just SSAA (supersampling anti aliasing), which has existed and been used in games since long before DSR/VSR came out. So there's nothing concerning in particular about SSAA being used even if DSR/VSR isn't enabled. But I haven't watched the video, maybe the game's own AA settings aren't working properly.

 

[TerryLaze]

I'm running a 2070 not even a 3000 card and my gpu has reached up to 90c with 100% fan spin, but the worst part is that new world straight up isn't giving a single sizzle about their settings. I've ran everything including resolution on the lowest settings and supposedly fps capped at 60 and it still hit those temperatures. Whilst when I've ran the fps cap that's built into msi afterburner to 60 fps max the temps drops basically immediately and the game isn't even reaching those temps when running on highest settings, only sitting at around 80-85c. It's definitely atleast partly new worlds fault for gpus dying.

This is a super common practice that a lot of games use, they render the game as fast as possible no matter what FPS limit you use in-game to not create input lag and to always be able to give you the most recent frame at the frame rate you chose.

 

[hotaru.hino]

It's definitely atleast partly new worlds fault for gpus dying.

So the next time some game blows up my GPU, it's the game's fault?

Then we should start absolving hardware companies for past issues. Spectre/Meltdown got you in the dumps? Just don't run software that exploits it! ezpz lemon squeezy.

 

[TJ Hooker]

Very timely and this guy is one of the few on the internet you can 100% trust in what he says in regards to hardware:

View: https://www.youtube.com/watch?v=w-zSgiMEzkU


Regards.*

Buildzoid is commendable for being one of the few people in the tech press who actually does deep dives into hardware, particularly VRMs. But he's far from infallible. Even his methodology in that video seems a bit off. I.e. when he's probing each probing all the VRM input filtering caps to try to identify the particular phase/power stage that failed. All those caps are between 12V and ground. So the all caps tied to the PCIe slot 12V are in parallel, and all the ones tied to the PSU power connector 12V are in parallel (I believe these ones may also be separated into two separate groups of parallel components, depending if the power connector inputs are shorted together on the card, which is optional per the PCIe spec IIRC). So when he measures a short across one of the caps, it doesn't mean that exact power stage/phase is shorted, it just means that at least one of the power stages/phases that's connected to the same 12V source (PCIe slot or power connector) is shorted. But at first, he seems to be saying that measuring a short on a particular cap means that particular power stage is short. He only seems to realize that's not the case further on in the video, when he's also seeing shorts on caps associated with power stages that are for other GPU power rails.

Which is one of the issues I've had with the handful of videos I've watched of his over the years, he ends up figuring things out as he goes. Which isn't bad in itself, but he presents everything as 100% correct along the way, and doesn't seem to really go back and address the mistakes he's made. The fact that doesn't have a real background in electronics also means he often doesn't have the understanding to be able to put things in proper context. I.e. he will refer to some components/designs as garbage just because they have worse ratings in a few metrics he pulled out of a datasheet compared to some other components he's looked at previously. But he'll ignore the possibility that both designs are more than sufficient for their intended purpose, or that there could be other reasons the engineers chose the components they did (other than just cost). Or he used to do this at least.

Edit: I'm probably being too hard on the guy, I really do like what he's done for hardware reviews, particularly motherboards. Actually diving into how many "real" phases a board has, or whether they're using doublers or "fat phases", etc. Compared to a lot of reviews that would just regurgitate what the mobo's product page says, or at best just count the number of inductors on the board and list that as number of phases.

 

[-Fran-]

Buildzoid is commendable for being one of the few people in the tech press who actually does deep dives into hardware, particularly VRMs. But he's far from infallible. Even his methodology in that video seems a bit off. I.e. when he's probing each probing all the VRM input filtering caps to try to identify the particular phase/power stage that failed. All those caps are between 12V and ground. So the all caps tied to the PCIe slot 12V are in parallel, and all the ones tied to the PSU power connector 12V are in parallel (I believe these ones may also be separated into two separate groups of parallel components, depending if the power connector inputs are shorted together on the card, which is optional per the PCIe spec IIRC). So when he measures a short across one of the caps, it doesn't mean that exact power stage/phase is shorted, it just means that at least one of the power stages/phases that's connected to the same 12V source (PCIe slot or power connector) is shorted. But at first, he seems to be saying that measuring a short on a particular cap means that particular power stage is short. He only seems to realize that's not the case further on in the video, when he's also seeing shorts on caps associated with power stages that are for other GPU power rails.

Which is one of the issues I've had with the handful of videos I've watched of his over the years, he ends up figuring things out as he goes. Which isn't bad in itself, but he presents everything as 100% correct along the way, and doesn't seem to really go back and address the mistakes he's made. The fact that doesn't have a real background in electronics also means he often doesn't have the understanding to be able to put things in proper context. I.e. he will refer to some components/designs as garbage just because they have worse ratings in a few metrics he pulled out of a datasheet compared to some other components he's looked at previously. But he'll ignore the possibility that both designs are more than sufficient for their intended purpose, or that there could be other reasons the engineers chose the components they did (other than just cost). Or he used to do this at least.

Edit: I'm probably being too hard on the guy, I really do like what he's done for hardware reviews, particularly motherboards. Actually diving into how many "real" phases a board has, or whether they're using doublers or "fat phases", etc. Compared to a lot of reviews that would just regurgitate what the mobo's product page says, or at best just count the number of inductors on the board and list that as number of phases.

I haven't been following him for long, but all of the videos seem pretty correct to me. He rambles a lot though and I can see where you're coming from, but giving how he's also exposing what he is doing you can actually drive your own conclusions in most videos he's put out.

This being said, you're not wrong on your overall impression. He's not infallible, like pretty much anyone on earth (with or without a degree), but at least he has very good experience on the table in order to diagnose a card like that properly, which is enough in this particular case, no?

EDIT: And also the article was updated with the followup video in which he shows the card fixed. Good timing, again

Regards.

 

[RemmRun]

So who do they send the refund request too?

 

[$h0nuff]

Traditionally, it would be the responsibility of both the software maker and the hardware maker to get together, identify and resolve issues. It's in the best interest of both parties to address issues and avoid bad publicity. Not much traditional happening these days in the gaming industry segment.

EVGA came clean about and known issue on first-run cards and then Amazon adopted that response as the default PR go-to. Lazy and/or something slightly more nefarious at play. Ya know..nothing shady ever happens in the business world..

Hardware power limitations are set in software (RISC) coding. Jay pointed out some crappy compatibility issues between Amazon's software and multiple pieces of disparate vendor hardware. Compatibility issues which should not be occurring and are the responsibility of Amazon to fix.

How power levels are spiking beyond defined limits on the card components needs to be addressed by the hardware vendors who, yikes, also use software!!

My guess is it has something to do with the GPU clock boosting algorithm and how the VRM addresses power scaling in relation to clock scaling. I would bet my paycheck multiple card vendors have reached out to Amazon for testing assistance and all of them are still awaiting a response. READ: Jeff's unethical method for picking a favorite vendor to purchase. There, I said it..

 

[Brian28]

There is a video of proof that the game ignores actual gpu settigns and will upscale the game to a resolution you dont even have the settign enabled for by default.
as well as ignoring power limits on gpu's settigns.

If a game is able to force the GPU beyond the limits, then that is the driver's fault, not the game. The game doesn't do anything without going through a driver.

 

[Blacksad999]

This game is hitting cards harder than a benchmark does, except that a benchmark runs for a very short time frame, while people are playing this game for hours and hours. There's 100% something wrong with how this game is coded, as it shouldn't be doing that obviously. The game isn't even all that demanding, and we've seen no issues from games that are significantly more demanding than this. I call BS on Amazon's end.

 

[-Fran-]

This game is hitting cards harder than a benchmark does, except that a benchmark runs for a very short time frame, while people are playing this game for hours and hours. There's 100% something wrong with how this game is coded, as it shouldn't be doing that obviously. The game isn't even all that demanding, and we've seen no issues from games that are significantly more demanding than this. I call BS on Amazon's end.

Or just use an AMD card which doesn't have any issues with the game?

I mean, you're basically advocating for nVidia's wet dream here; zero accountability demanded from them or the AIBs. Passing all the blame to Amazon and, more importantly, to the developers of the game that just use cards to develop and test. Who knows, maybe Amazon did tell nVidia about it and they didn't do anything? AMD doesn't seem to have an issue with the game, why would Amazon need to do anything?

I call BS on everyone not realizing this is nVidia's and the AIB's problem and not Amazon.

Regards.

 

[Deleted member 14196]

We already Figured out it was hardware maker’s fault in the last thread

 

[$h0nuff]

We already Figured out it was hardware maker’s fault in the last thread

We also figured out the earth is round and time is linear (for the most part). sheesh..

 

[Deleted member 14196]

No you’re wrong time is a man-made construct. This thread is absolutely useless and a rehash of what already been established. Don’t bother responding you will be ignored

 

[$h0nuff]

Or just use an AMD card which doesn't have any issues with the game?

That's fine for users who have unlimited disposable income but is far from addressing the immediate issue. And most of us are on a budget. Odd response..

I mean, you're basically advocating for nVidia's wet dream here; zero accountability demanded from them or the AIBs. Passing all the blame to Amazon and, more importantly, to the developers of the game that just use cards to develop and test. Who knows, maybe Amazon did tell nVidia about it and they didn't do anything? AMD doesn't seem to have an issue with the game, why would Amazon need to do anything?

This entire paragraph is speculation. Perhaps the rest of us (without unlimited disposable income) are trying to find facts versus speculate on what might be the case..

I call BS on everyone not realizing this is nVidia's and the AIB's problem and not Amazon.

Raised.. Do you work for Amazon? Very partisan, pro-amazon responses on a forum which was, 20 years ago, pretty bipartisan.. Everything is for sale these days but the salesmanship is a bit lacking here..

 

[$h0nuff]

No you’re wrong time is a man-made construct. This thread is absolutely useless and a rehash of what already been established. Don’t bother responding you will be ignored

The passage of things which happened (stay with me), takes place in a linear fashion. Time, as you refer to it, was broken into measurable chunks to facilitate reflection against linear passage. It's, like, good for science stuff bro! Linear movement is independent of time.

How will I get through the day considering your admonishment? lol

 

[COLGeek]

Folks, let us maintain our decorum, please. Thank you.

 

[A-Zilis]

what part of the game ignores the gpu's limits don't you understand?

A game being able to ignore power limits is clearly a problem with the hardware or its drivers. Do you really expect the hardware-maker to just throw up its hands and say "the software is ignoring my limits . . . nothing I can do"?

A power limit is not just a guideline to coders to choose to obey or disobey. If that's something you genuinely don't understand, that's alright, but stop trying to talk like an expert on the subject when you clearly have little understanding of it.

 

[Tac 25]

a software that destroys gpu's? If true, then this is just sad when graphics cards are so rare or cost a bomb.

 

[Phaaze88]

Hardware faults are more consistent. Software faults are more random. [They're probably interchangeable...]
How these cards off themselves appears to be consistent. Black screen and ded.
What's doing it doesn't appear to be. New World isn't the only reported offender, and it doesn't appear to be happening to every RTX 3080, 3080Ti, 3090, etc... at least not yet. Maybe they need more time than others...

 

[Mattzun]

Nvidia fans blamed power supply manufacturers with tight OCP for failures for 3090 and 3080 problems long before New World came out.

It's pretty clear that the 3xxx NVidia cards blow way past the power limits in MANY cases.
In the past, these have been fairly short transients, so power supplies that didn't allow short spikes way past power limits were blamed for the problem instead of the card (this could actually be a power supply problem if short (<10ms) spikes are normal)

Apparently New World keeps the cards at unsafe levels for significantly longer periods.

If I was Amazon, I would have publicly offered to send NVidia, EVGA etc the menu code (prepatch) that easily recreated the problem and asked them to help with a workaround for the bad NVidia drivers/firmware until NVidia (or the OEM) had a real fix.

If I was at Nvidia/EVGA QA, I would have ASKED Amazon for the menu code that could trigger a hidden flaw in my product and trying to reproduce
I'd also be supporting them in avoiding the flaw in my product (NVidia supporting optimizing a major game is pretty normal)

As a QA Engineer, getting a reproducible scenario is 90 percent of the problem in getting a fix.

If Amazon was doing anything that bypassed the APIs that should enforce the power limits, I'm sure we would have heard about it by now instead of having EVGA acknowledge manufacturing flaws

 

[$h0nuff]

a software that destroys gpu's? If true, then this is just sad when graphics cards are so rare or cost a bomb.

This.. Choosing to sabotage a competitor is as old as business practices. Many seem to forget this is Amazon we're talking about. Misappropriated billions to launch rockets, firmly in the monopoly category but somehow avoiding scrutiny, dismissive of their number one resource (employees), etc.

The issue has been addressed and fixes are on the way. Truth be told, this was a dry-run for Intel's upcoming product release. Why Nvidia and not AMD. Logic..

 

[Falkentyne]

So far, all of the cards that have died seem to have Alpha and Omega (AOZ) power stages (50A?). The eVGA (revision 0.1 PCB) and many Gigabyte cards have these power stages.
The new revision 1.0 PCB eVGA Cards seem to have Onsemi power stages and none of them have been reported as blowing up.

The Founder's Edition (3080 and 3090) have some smart power stages but I don't remember what they're called but they're not AOZ, and they're 70 amp stages per phase.