RTX 4090?

[evernessince]

Not even 0.1% of cards in circulation, my point stands tbh. Considered that the 4090 actually sold better than the 4070 (all variants) and 4080 (all variants) + the entire RX 7900 series combined

No idea where you got that 0.1% figure but it's another thing you are pulling out of your rear. No one has the exact number of RMA'd cards attributable to burned up connectors, even board partners only see a part of the picture. That said we can extrapolate from the information we do have that it is significant in terms of total RMA volume and was noted to the point where boards partners had to contract it out to additional outside repair shops.

I will never understand why people defend the new connector, any amount of additional failures is a net negative regardless of how insignificant you personally think it is. People convince themselves that's it's not important enough to care and that's a customer attitude that allows companies to make a lot of design compromises that add up over time. If you are paying $1,700+ for a video card you should demand a decent connector, not somethings that's worse than a 17 year old standard.

 

[Dr. Dro]

No idea where you got that 0.1% figure but it's another thing you are pulling out of your rear. No one has the exact number of RMA'd cards attributable to burned up connectors, even board partners only see a part of the picture. That said we can extrapolate from the information we do have that it is significant in terms of total RMA volume and was noted to the point where boards partners had to contract it out to additional outside repair shops.

I will never understand why people defend the new connector, any amount of additional failures is a net negative regardless of how insignificant you personally think it is. People convince themselves that's it's not important enough to care and that's a customer attitude that allows companies to make a lot of design compromises that add up over time. If you are paying $1,700+ for a video card you should demand a decent connector, not somethings that's worse than a 17 year old standard.

It'd only be a shame if the connector not only didn't present a problem, but had already been revised by the PCI-SIG and fixed. Besides, all newer production cards already use the revised 2x6 connector, which will be safe even if you deliberately mishandle it and intentionally pull the connector under max load, as you can see here:

Rest in Peace 12VHPWR Connector - Welcome 12V-2×6 Connector, important modifications and PCIe Base 6 | Exclusive | igor´sLAB

The king is dead, long live the king. Or the other way around: The findings of the last few months are finally taken seriously and the connector is modified, work is done on a new standardization and…

www.igorslab.de

Not only the failure rate is statistically insignificant, but all companies involved have stepped up to make things right and the benefits from the new connector far outweigh any cons, especially considered it uses less material to manufacture and you don't need to run three bulky cables off your power supply to feed your graphics card. In particular, if you have any NVIDIA GPU manufactured in the past 7 months, you're guaranteed to have a revised connector installed. Not to mention the original didn't exactly burn unless it was used exclusively with the 4090 (at 400W+ load) and inserted incorrectly.

 

[evernessince]

It'd only be a shame if the connector not only didn't present a problem, but had already been revised by the PCI-SIG and fixed. Besides, all newer production cards already use the revised 2x6 connector, which will be safe even if you deliberately mishandle it and intentionally pull the connector under max load, as you can see here:

Think about what you just wrote. Your logic is essentially "there is no issue but if there was an issue it's already been addressed". Somehow you are ignoring the obvious fact that the existence of the revision is an admission there there are issues.

Specifically that the connector could still draw power when not fully seated. The revision does nothing to address the low safety margin, the 30cm of straight cable out of the connector requirement, or when the cable shears horizontally from the connector it puts pressure on the pins thus increasing resistance. These are all known facts.

Not only the failure rate is statistically insignificant, but all companies involved have stepped up to make things right and the benefits from the new connector far outweigh any cons, especially considered it uses less material to manufacture and you don't need to run three bulky cables off your power supply to feed your graphics card. In particular, if you have any NVIDIA GPU manufactured in the past 7 months, you're guaranteed to have a revised connector installed. Not to mention the original didn't exactly burn unless it was used exclusively with the 4090 (at 400W+ load) and inserted incorrectly.

Again, the mere fact that companies needed to step up to make things right tells us that it was in fact not statistically insignificant. You don't seem to be able to connect the dots that company's took action because it was indeed an issue.

Benefits? You mean benefit, it's more tidy / less cables. Using less material and having thinner wires isn't a good thing when it comes to power delivery. That just increases the probability of issues. Don't see people listing less bulky / thinner wires as a benefit to low end power supplies but maybe PSU companies should hire you to advertise their low end models. Could sell home fire insurance while you're at it.

 

[Dr. Dro]

Arguably, even one card catching fire is a serious enough issue for manufacturers to step in; however it clearly wasn't anywhere near as bad as to require a global recall. And in relative terms of all units sold, it's clearly statistically insignificant. I think the industry learned with its mistake and NVIDIA took care of all affected parties. No use holding this against them.

 

[evernessince]

Arguably, even one card catching fire is a serious enough issue for manufacturers to step in; however it clearly wasn't anywhere near as bad as to require a global recall. And in relative terms of all units sold, it's clearly statistically insignificant. I think the industry learned with its mistake and NVIDIA took care of all affected parties. No use holding this against them.

Many people thought there should have been a recall. Your argument in this comment relies on the assumption that Nvidia would recall it's products in good faith if not forced to by a regulator.

Let's challenge that assumption. We have in fact a revised connector but Nvidia did not offer any of it's customers the ability to return their card to get a replacement with the new revision. You already said in a past comment that the revision fixed the issue, the same comment you paradoxically said the issue didn't exist but was also somehow "fixed":

It'd only be a shame if the connector not only didn't present a problem, but had already been revised by the PCI-SIG and fixed

So you've admitted that at the very least that there was an issue that the revision fixed. By Nvidia not issuing a recall to owners of cards with first revision connectors, the issue that existed by your own admission is still present on all those early cards. Therefore, just based off your own provided statements we can see that in this case Nvidia has not acted in good faith as early 4000 series owners are still straddled with a connector to which you yourself admit has an issue in which the revision fixes.

This is before we even consider the information provided by other outlets in regards to this issue and Nvidia's notrious anti-consumer history. I'd argue that it's foolish to believe any company would do something in good faith but to believe Nvidia of all companyes would requires a certain high level of mental abstraction. Nvidia explicitly failed to issue a recall for 1st gen connector equipped cards when in your own words PCI-SIG "fixed" the issue with it's revision. I say fixed because you have yet to address the fact that it only fixed one out of four issues with the connector. More or less your argument at this point is a rotating set of goal posts as you continue to change your argument.

 

[Chrispy_]

According to HWU's industry sources and repair shops that have contracted with board partners to repair cards with burnt connectors, the incident rate of burnt connectors on 4090s was significantly higher than any other RMA reason to the point where it significantly increased overall RMA rates. The revised adapters reduced those rates but one contracted shop is on record as still receiving 200+ burnt 4090s / month. Not terrible but still notably higher than other causes for RMAs. That's the revised adapter, I'd expect that number was significantly higher on the first design.

The 7900 XTX defective vapor chamber literally impacted a single batch of coolers, that's it. A single bad batch is not going to be comparable to a design flaw that impacts every 4090 and in particular cards with the 1st gen connector (a design flaw of which still isn't fully remedied). You are trying to weight them as if equal but they are anything but.



Now on to the OP's question as to whether they should upgrade to the 4090. IMO it's only "worth it" if you have money to blow or plan to use it heavily for work. Otherwise at 1440p it's 15% faster for 1.75x to 2.1x the money.

On top of that we are already more than half-way through this generation. You in essence want to upgrade to the card that is going to loose $700 - $900 in value in less than a year. That's considering that you are coming from a 4080, which was itself the poorest retainer of value from the entire 4000 series lineup since launch. Again the only frame of mind that makes sense to me in is either work or you have money to burn. If you have that kind of money to blow you might as well pickup a P5800X 800 GB as well for a cool 2K. After all it is by far the best storage for consumer workloads.

When you're dropping $2000 on hardware that fails within the first year, the warranty has you covered - even for third-tier brands like Palit.
For the love of all that is right, do not infer trends from Reddit!

I have no love for Palit; they always seem to be the cheapest, minimum-effort product in my regions, but when GPUs have failed, their RMA process was no worse than the Tier-one vendors like Asus/Gigabyte/MSI.

 

[evernessince]

When you're dropping $2000 on hardware that fails within the first year, the warranty has you covered - even for third-tier brands like Palit.
For the love of all that is right, do not infer trends from Reddit!

I have no love for Palit; they always seem to be the cheapest, minimum-effort product in my regions, but when GPUs have failed, their RMA process was no worse than the Tier-one vendors like Asus/Gigabyte/MSI.

Being covered by warranty is decent but being out of your $2,000 GPU 1-2 months plus shipping is not. It's also not nice when you get a refurb card back because they aren't required to give you something new. The vast majority of people would prefer to just not have to deal with RMAing at all, especially when you see horror stories like that from Gigabyte where customer's broken cards are returned with stickers apparently labeling a small nick or scratch that could have easily been from the factory as a reason to deny RMA. It's just introducing more BS when they could have just made the card proper from the start and gives them another opportunity to jerk around the customer. I don't think having a warranty is an excuse for product defects being acceptable. If we start accepting higher failure rates because it's under warranty, you create a mentality that it's ok to release lower quality products because it's covered for a short period of time (and I do personally believe 2-3 years is short for a video card).

That's for an in warranty product. Warranty is only going to cover 2-3 years, after that you have to contend with the fact you have an extremely expensive product with an elevated chance of failing. If I'm a company I'm designing a product to last just long enough to cover that warranty period as much as possible, it's after the warranty period you have to be concerned with as a customer. The pins on the 12VHPWR cable spread more after each cycle and depending on if the end user is following spec. There was never a broad attempt to inform users that in spec the cable requires 30 cm of straight cable and should not be horizontal offset. Failure to adhere to both increases the pressure on the pins, spreading them out. This may not be evident immediately as that kind of wear takes time. Considering we are talking about a $2,000 product, this is the kind of information I'd be interested in and it's precisely why I have my 4090 and 4080 mounted in a horizontal mount case with zero sag following the specifications to the T with the 4090 set to 85% power.

Unfortunately a lot of users are not aware of these factors as Nvidia has made no effort to educate customers almost certainly because any failures of that nature are likely to fall outside the warranty period.

 

[Chrispy_]

Yeah, having to RMA sucks, but I don't think it's actually affecting that many people.

I 100% agree that the 12VHPWR connector is flawed and melts at full load, but so few people ever actually hit full, sustained load that I don't think it's a huge deal. A "significant doubling of RMAs" is still only 2-3% of a relatively niche market. Mainstream GPUs are outselling 4090s 25:1

 

[Dr. Dro]

Yeah, having to RMA sucks, but I don't think it's actually affecting that many people.

I 100% agree that the 12VHPWR connector is flawed and melts at full load, but so few people ever actually hit full, sustained load that I don't think it's a huge deal. A "significant doubling of RMAs" is still only 2-3% of a relatively niche market. Mainstream GPUs are outselling 4090s 25:1

Interestingly enough, this doesn't seem to be case. The 4090 is the best seller this generation, it sold more than the entire AMD stack top to bottom, and outsold all flavors of 4070 and 4080 combined as well. The last data I read was from around ~3 months ago and it was around ~1.9M 4090 samples vs 803K 4080s and just shy of 493K 7900XTX's. DM T4C Fantasy for the updated stats, he'll probably share.

Gotta say though, don't expect GPU prices to go down anytime soon

 

[Outback Bronze]

Gotta say though, don't expect GPU prices to go down anytime soon

Why would they drop the price if their top tier card is selling the best? I'm surprised the 4090 has sold so well within the current economic landscape. Just goes to show there is still plenty of deep pockets.

There is now an incentive for Nvidia to put prices up as they are quite well aware people are happy spending their dough on high end tech, even if it is overpriced.

OP: Keep the 4080 matey, 50 series RTX Blackwell could be around end of year.

I actually saw a 4090 cheap and still decided to put my money towards next series.

 

[Dr. Dro]

Why would they drop the price if their top tier card is selling the best? I'm surprised the 4090 has sold so well within the current economic landscape. Just goes to show there is still plenty of deep pockets.

There is now an incentive for Nvidia to put prices up as they are quite well aware people are happy spending their dough on high end tech, even if it is overpriced.

OP: Keep the 4080 matey, 50 series RTX Blackwell could be around end of year.

I actually saw a 4090 cheap and still decided to put my money towards next series.

Either deep pockets or there's a general willingness to buy really good products, the performance of the 4090 was really on a class all its own this generation. But agreed