Building a gaming PC

[eskiMoe]

Yeah. I can still see some lawsuits coming their way though. Plenty of people opted for a GT 970 SLI setup to run games @ 4K and not being able to properly access that last 500mb portion is an issue to them.

Well, even with "only" 3,5GB of usable VRAM there isn't really nothing that compares to the 970 in terms of performance per dollar/euro/ with the same efficiency. I guess I still keep these.

But really, would be nice to get some free games at least or something.. right?

 

[Tar-Aellin]

How do you connect your GPU to your monitor? It could be a cable/connection/monitor issue.

Sorry for late return. It's connected by DVI. I'll seek closely that cable issue because if i'm not mistaken it did that once when my foot kind of hit cables around. It was very recent, yesterday or the day before that. I hope it's just that.

 

[Gilrond-i-Virdan]

Is there any practical lesson from this debacle? May be one should wait for a while before buying new hardware to get more bug reports to analyze?

 

[eskiMoe]

Is there any practical lesson from this debacle?

Well not to trust Nvidia obviously.

 

[Gilrond-i-Virdan]

Well not to trust Nvidia obviously.

You mean they knew it all along?

 

[eskiMoe]

You mean they knew it all along?

I don't see how that's relevant? Whether by accident or on purpose they fucked up. And are unwilling to admit that.

So yeah, the trust is gone.

 

[volsung]

I still think Nvidia will have to address this issue somehow, but at the moment they are probably panicking and figuring out how much it's going to cost.

What can happen if an entire series of cards has manufacturing errors? They either fix, refund or replace, if they want to keep their customers.

 

[sidspyker]

May be one should wait for a while before buying new hardware to get more bug reports to analyze?

The card has been out for 5 months now and nobody reported any such issues. How long would you want people to wait? A year? Two?

 

[Gilrond-i-Virdan]

5 months and until now no one performed a thorough benchmark which includes such a basic test like filling up all the VRAM? I guess then it's another lesson - someone should come up with a more comprehensive list of benchmarks to run on any new GPU hardware.

 

[GuyNwah]

That's actually nVidia QA's job. Their QA department got caught with their pants down on this one. Products are not supposed to reach GA with performance-destroying bugs like this one, even when they are leading-edge.

nVidia will make the case that their cards are sold to run applications, not to run artificial benchmarks. They will be right in this. Nia's "benchmark" as a measure of whether their card is merchantable is meaningless. It is the ability to perform in games, Photoshop, and the like that matters.

But what they have done is still the equivalent of making an airplane that is specified to fly at 40,000 feet, and not only fails to do so but does not warn you when you pass 35,000 feet and causes you to lose control. It wouldn't be so bad if it enforced a 3.5 GB limit (or a 35,000 foot ceiling), but the real fault is that it crosses that limit and then fails to perform.

 

[tahirahmed]

I think the problem is also associated with the hype made by critics and reviewers. Maxwell was and is still being hyped to extreme levels, it's a good architecture and evolution of GPU technology no doubt but nothing should be hyped so much.

 

[MkTama]

Kinda sad to say that but waiting is always the best strategy.. New software and hardware at launch is more and more often a huge bet

 

[sidspyker]

It's been 5 months since launch though, nobody noticed the issue until recently. How long could one possibly wait?

There have been reports of older Nvidia GPUs having similar issues as well. Like GTX Titan and GTX 770 4GB.

^Also this, people with cards ranging from the 500 series down to the 700(and ofc 900) showed similar results

 

[eskiMoe]

Kinda sad to say that but waiting is always the best strategy.. New software and hardware at launch is more and more often a huge bet

There have been reports of older Nvidia GPUs having similar issues as well. Like GTX Titan and GTX 770 4GB.

 

[MkTama]

There have been reports of older Nvidia GPUs having similar issues as well. Like GTX Titan and GTX 770 4GB.

Of course with "new" I was not only referring to Maxwell, was more a general things involving the PC world (and not only). I tend to purchase 6 months/one year behind also with games, it's really great from many points of view, IF you can manage the fact that you don't play at release (which I don't know if I can do also with TW3 cause I'm so f****** hyped)

@sidspyker you're perfectly right.. but still the issue is there and cannot be easily resolved if not by waiting.. and hoping

 

[eskiMoe]

Better just play Monopoly and avoid any hardware or software issues altogether.

 

[GuyNwah]

There have been reports of older Nvidia GPUs having similar issues as well. Like GTX Titan and GTX 770 4GB.

If those are the reports claiming impairments on "Nia's benchmark", the more plausible conclusion to draw would be that the "benchmark" is invalid.

Performance with real applications and benchmarks devised to simulate the behavior of real applications is the only thing the card is specified and sold for. Contrived "tests" that do nothing but encumber memory are bullshit.

There's still a real problem with the 970. But the evidence for it is other than the "Nia's benchmark" claim.

 

[MkTama]

Do you think that the problem with 970s could be driver-solvable or not?

 

[GuyNwah]

Do you think that the problem with 970s could be driver-solvable or not?

I suspect it will be, and what I think it will do is lock out part of the VRAM.

The best thoughts on the nature of the bug that I have read are that it is related to the disabling of 3 shader modules (SMs) on the 970. The GM204 chip has 16 SMs, and all 16 are used on the 980; but on the 970 3 are disabled, leaving 13. This is done for the legitimate reason of yield management.

if the architecture is such that access to the VRAM is through the SMs, it is plausible (but nobody has confirmed) that a path to some of the memory is through the disabled SMs, making it slow or unusable for some purposes.

Preventing attempts to allocate memory in the inaccessible ranges is something that should be within the capabilities of the driver. This would reduce a card promoted as 4GB to 3.25GB or so, but it would be less catastrophic than having games or, worse, professional applications take a performance hit when they demand memory that is present but can't be used efficiently.

 

[MkTama]

I suspect it will be, and what I think it will do is lock out part of the VRAM.

The best thoughts on the nature of the bug that I have read are that it is related to the disabling of 3 shader modules (SMs) on the 970. The GM204 chip has 16 SMs, and all 16 are used on the 980; but on the 970 3 are disabled, leaving 13. This is done for the legitimate reason of yield management.

if the architecture is such that access to the VRAM is through the SMs, it is plausible (but nobody has confirmed) that a path to some of the memory is through the disabled SMs, making it slow or unusable for some purposes.

Preventing attempts to allocate memory in the inaccessible ranges is something that should be within the capabilities of the driver. This would reduce a card promoted as 4GB to 3.25GB or so, but it would be less catastrophic than having games or, worse, professional applications take a performance hit when they demand memory that is present but can't be used efficiently.

So in the end, if they do not correct the problem at the source (which I doubt they intend to do) 970s will be "castrated" regarding VRAM quantity.. Not a great deal unfortunately