NVidia CEO and President on GTX 970

[Trik'Stari]

They have to cripple the last half gig because when they cut down the core, they lose three SMs and a segment of L2 cache. That segment of L2 cache and those SMs are attached to a memory controller. now three SMs and one segment of L2 has to handle two memory controllers, but it can't do so at it's absolute capacity. I'm pretty sure that's how it's described.

-snip

Why not just....not...do that?

[TheSLSAMG]

It IS a new feature of Maxwell - it would be a 3GB card if it wasn't for this feature.

What feature though? The GTX 980 is a 4GB card regardless. Since when has cutting down a core become a feature?

 

Not sure, not trying to troll, just want an answer.

[Whorax]

He didnt have much to say unfortunately.

Kinda what I was thinking.

[79wjd]

Why not just....not...do that?

Because then it would be a 980.

970s are made out of the chips that physically can't be 980s due to manufacturing defects -- i.e. take a 980 where those few components don't work properly and then disable them and market it as a 970.

[Trik'Stari]

Because then it would be a 980.

970s are made out of the chips that physically can't be 980s due to manufacturing defects.

Meaning: It's market segmentation, not a feature.

[79wjd]

Meaning: It's market segmentation, not a feature.

It's the same reason why i3s, i5s, and i7s exist.

An i5 is just an i7 with hyperthreading disabled as the chip wasn't capable of running at i7 specs.

It's a feature that allows the card to access more than 3gb of VRAM; A 980 has the feature working at 100%, while the 970 does not.

[Speedbird]

What feature though? The GTX 980 is a 4GB card regardless. Since when has cutting down a core become a feature?

 

Not sure, not trying to troll, just want an answer.

The GTX 980 is a full GM204 with all the SMXs, cache and ROPs. The 970 has a core cut down because that is how things work. Need a lower-end card? Turn off a few SMXs. By doing that, they lose a bit of L2 cache and some ROPs, which prevents the last memory controller from running at it's full capacity. If this was Kepler or older, they would have to turn off the whole memory controller and make it a 3GB card. But because Maxwell is more modular, they gave it 3.5GB of full VRAM and 512MB of slower VRAM.

[TheSLSAMG]

Meaning: It's market segmentation, not a feature.

It's market segmentation for pretty much any card out there.

 

R9 290s are 290Xes that have defective portions of the core, as are R9 280s R9 280Xes. GTX 760s were defective GTX 770s, R7 265s are defective R9 270s/R9 270Xes and so on. Standard market practice, except for how they're cut down. If they didn't do this, they wouldn't make very much money.

[TheSLSAMG]

The GTX 980 is a full GM204 with all the SMXs, cache and ROPs. The 970 has a core cut down because that is how things work. Need a lower-end card? Turn off a few SMXs. By doing that, they lose a bit of L2 cache and some ROPs, which prevents the last memory controller from running at it's full capacity. If this was Kepler or older, they would have to turn off the whole memory controller and make it a 3GB card. But because Maxwell is more modular, they gave it 3.5GB of full VRAM and 512MB of slower VRAM.

Oh yeah, I forgot about this.

 

Realistically, while most would shit on the idea, the architecture enables this when previous models couldn't. This makes it a feature at least in my mind.

[razor767]

nice to hear some kind of apology, I just wish they would finally stop calling this a "new feature of maxwell". Its a shot in the foot, not a feature.

I wouldn't call it an apology myself, just more PR nonsense.

[Epitome]

Does it even effect performance in any major way?

 

(Seriously asking) 

[TheSLSAMG]

I wouldn't call it an apology myself, just more PR nonsense.

We won’t let this happen again. We’ll do a better job next time. (The last two sentences of the statement)

I'd call it a start. Usally apologies start with "I'm sorry." But if this is as close as we will get, I'm okay with it for now.

[ivan134]

He wait until a lawsuit to say something? This arrogance is staggering.

[TheSLSAMG]

Does it even effect performance in any major way?

After the first 3.5GBs are accessed and it has to add the last bit in, you can notice frame stuttering and a significant performance dip.

 

However, the card still performs as well as it did on Day 1, about as well as the 290X at 1080p and 1440p (trading blows) while getting a bit behind at 4K (which isn't exactly viable yet IMO considering the cost of a good IPS panel along with the GPUs to run it at 50FPS+/Ultra.)

[Guest Kloaked]

He wait until a lawsuit to say something? This arrogance is staggering.

No, he just figured most of his his audience and customers were smart enough to realize what they're doing and didn't have to release a statement. He was dead wrong obviously.

[Vrbas]

He wait until a lawsuit to say something? This arrogance is staggering.

Even with the lawsuit he could of kept quiet. Lawsuit had nothing to do with it.

[ThomasD]

A terribly dishonest statement. 

 

They flat out claimed it had the same memory architecture as the 980, now they want to say they invented something "new" and only failed to tell the world about their wonderful new invention.

 

Manure.

 

This only makes me feel worse about Nvidia.  Had they just shut up it wouldn't have been this bad.

[nj4ck]

It IS a new feature of Maxwell - it would be a 3GB card if it wasn't for this feature.

Board partners have been adding memory for ages and the 980 doesn't appear to have any problems with 4GB of functional memory.

[ivan134]

No, he just figured most of his his audience and customers were smart enough to realize what they're doing and didn't have to release a statement. He was dead wrong obviously.

Very subtle. Apparently anyone who recognized this BS for what it is, is "not smart enough". I'll give it to NVIDIA though. They sure do know how to work their audience and know they would let them get away with murder.

[79wjd]

A terribly dishonest statement. 

 

They flat out claimed it had the same memory architecture as the 980, now they want to say they invented something "new" and only failed to tell the world about their wonderful new invention.

 

Manure.

 

This only makes me feel worse about Nvidia.  Had they just shut up it wouldn't have been this bad.

It does have the EXACT same memory architecture as the 980. Parts are just disabled due to them not functioning properly -- otherwise the card would be branded as a 980.

[ivan134]

Even with the lawsuit he could of kept quiet. Lawsuit had nothing to do with it.

Yea, he just decided to release a statement on a whim even though this has been going on for a month. The recent lawsuit was just a coincidence. I have a few bridges to sell you.

[Speedbird]

Board partners have been adding memory for ages and the 980 doesn't appear to have any problems with 4GB of functional memory.

The 980 doesn't have problems because it's a full GM204... please read post 24 on the previous page.

[cesrai]

It does have the EXACT same memory architecture as the 980. Parts are just disabled due to them not functioning properly -- otherwise the card would be branded as a 980.

But it isn't.

[ThomasD]

It does have the EXACT same memory architecture as the 980. Parts are just disabled due to them not functioning properly -- otherwise the card would be branded as a 980.

Something that was made abundantly clear when they marketed it.

 

Or, maybe not.

 

Like a Corvette only bored and tapped for six spark plugs... 

 

But it still has eight cylinders!

 

Never mind that this press release says it is different and that they knew it was different from the start.  So try a different type of handwaving.

[Trik'Stari]

It does have the EXACT same memory architecture as the 980. Parts are just disabled due to them not functioning properly -- otherwise the card would be branded as a 980.

I think it's more a case of "well we can't have a cheaper card compete with the 980...turn these things off", not that they don't function properly.