[Kitguru]Nvidia`s big Pascal GP100 have taped out - Q1 2016 release

[Headfoot]

So back to the argument that 'code name' means anything. It doesn't.

Is this a joke?

 

[tviceman]

So back to the argument that 'code name' means anything. It doesn't.

Using your logic, if Nvidia were to release a GP107 (same numbers as a 750/750Ti) that was 4mm2 and performed twice as fast as a 980 Ti, it would be a ripoff at $500 right?

Because the 107 means it's a low end card that should be $100-$150, right?

How about no.

I've been saying the same thing. Just like when AMD went from 5870 being high end, to the 6870 being not the high end, Nvidia did the same thing with their mid dies starting with GK104. And, to that end, I've also said many times I don't care what the product's name, code name, die size, memory type, or bus width is. All I care about is it's performance at the price I am paying.

I can see Russian's point from the perspective that history dictates Nvidia comes out with very big dies and, more recently, it's second biggest die has only temporarily filled the top spot of it's lineup insofar as performance is concerned. However, Russian, while I agree with some of your arguments, you always skip over the biggest fact that at the time when GTX 980 came out, it was a compelling product in and of itself at it's price. It was the fastest single GPU and the price premium over r9 290x was less than what the 780 TI's price premium was (when it was released), all while bringing better performance, lower power draw, and new features.

 

[Headfoot]

I've been saying the same thing. Just like when AMD went from 5870 being high end, to the 6870 being not the high end

Those aren't code names. Everyone here should expect marketing names to be total unreliable junk. Code names are less likely to be pure marketing smoke and mirrors because they are the names the actual engineers and internal business folk use to refer to the product. The code names of those two products were Cypress and Barts.

 

[tviceman]

Those aren't code names. Everyone here should expect marketing names to be total unreliable junk. Code names are less likely to be pure marketing smoke and mirrors because they are the names the actual engineers and internal business folk use to refer to the product. The code names of those two products were Cypress and Barts.

You missed my point, which was that the x80 nomenclature is no longer Nvidia's flagship name, just like when AMD downgraded the x870 to non-flagship status.

Regardless, my long bolded statement in my previous post remains. I couldn't care less about product names, code names, die sizes, etc.

 

[xthetenth]

It's not about scoring points for our favorite sports team I mean video card brand, it's establishing what products are getting sold and where they fit into a generation.

 

[Headfoot]

You missed my point, which was that the x80 nomenclature is no longer Nvidia's flagship name, just like when AMD downgraded the x870 to non-flagship status.

We agree. The conclusion then is if the first-released x80 product is not the flagship, then it shouldn't be priced like a flagship.

 

[tviceman]

We agree. The conclusion then is if the first-released x80 product is not the flagship, then it shouldn't be priced like a flagship.

We do agree, but
1) You didn't finish reading my bolded statement. The last sentence, specifically. If you need me to elaborate more, I can.
2) If you think $500-550 is flagship pricing now, LOL! Both AMD and Nvidia moved that curve up long ago (AMD more recently). In fact, Nvidia spoiled AMD from raising their price ceiling more. You best be wait'in for them thare price drops in tha future!

 

[Sabrewings]

We do agree, but
1) You didn't finish reading my bolded statement. The last sentence, specifically. If you need me to elaborate more, I can.
2) If you think $500-550 is flagship pricing now, LOL! Both AMD and Nvidia moved that curve up long ago (AMD more recently). In fact, Nvidia spoiled AMD from raising their price ceiling more. You best be wait'in for them thare price drops in tha future!

I don't mind. If it's more powerful than the previous, $650 for a flagship doesn't bother me. I work closely enough to hardware engineers that I understand every successive drop in nodes and increase in performance takes a lot more R&D. This silicon cow isn't giving up the goods as easily so it will cost to continue progressing.

 

[shady28]

This is the only place I hear people saying the 980 isn't or wasn't high end. For a card to remain the performance king for nearly a year and a half is unparalleled.

http://www.anandtech.com/show/8526/nvidia-geforce-gtx-980-review

"At the very high end the GTX 980 will be unrivaled."

http://www.maximumpc.com/nvidia-geforce-gtx-980-review2014/

"Due to circumstances largely out of its control, Nvidia has had to go ahead with a 28nm high-end Maxwell part instead, dubbed GM204. "

http://www.eteknix.com/nvidia-geforce-gtx-980-4gb-maxwell-28nm-graphics-card-review/

"Since Nvidia gave us a taste of Maxwell we’ve been eagerly awaiting that high-end variant and that’s exactly what we get today with the GTX 980’s launch. "

http://gadgets.ndtv.com/laptops/rev...80-review-the-fastest-gpu-on-the-block-635274

"It has taken a while for Maxwell to make the leap into the high-end space, but now we have the GeForce GTX 980 and 970 which replace the Kepler-based top-end models of the previous GTX 7xx series. "

http://www.techspot.com/community/t...ew-first-high-end-maxwell-gpus-arrive.205403/

The title says it all :

"Nvidia GeForce GTX 980 & GTX 970 Review: First High-End Maxwell GPUs Arrive"

 

[Sabrewings]

For a card to remain the performance king for nearly a year and a half is unparalleled.

It was only for six months. September 2014 to March 2015.

This was the right amount of time for Nvidia to finalize GM200 (Titan X). Wait a few more months for Nvidia to launch a pre-emptive strike on the Fury X in the form of the 980 Ti.

 

[RussianSensation]

I don't mind. If it's more powerful than the previous, $650 for a flagship doesn't bother me. I work closely enough to hardware engineers that I understand every successive drop in nodes and increase in performance takes a lot more R&D. This silicon cow isn't giving up the goods as easily so it will cost to continue progressing.

:thumbsup:

Exactly, the side-effect to NV bifurcating a generation and worse competition from AMD is PC enthusiasts now either have to upgrade twice in a generation or wait longer for the real flagship. Your other point also has a lot of merit - $650 for a true flagship is actually a very reasonable price given inflation, etc. but $500-550 for a mid-range card that gets leveled by 40%+ in the same generation is a crazy rip-off. I guess for gamers who upgrade every generation from x04 chip (680->980->GP204), they should be satisfied with the performance increase but that means they are still paying $500+ for what essentially is just a next gen mid-range chip/videocard that is artificially marketed as flagship because NV planned the new generation launch specifically that way. If NV released the large die 16nm GP100 first and it gives us 60-100% more performance than 980Ti in 2016, I will be pleasantly surprised but I don't see that happening to be honest since why would they when they can milk their next gen x04 part for another $500-550 for 9+ months? It doesn't look like AMD is much of a threat for NV at this point as they've used up the HBM advantage but NV will have new architecture + new node + HBM2. On paper NV has a huge edge going into the Pascal generation.

This is the only place I hear people saying the 980 isn't or wasn't high end.

It's very unfortunate that so many professional sites did such a disservice to their readers and didn't have the balls to call out the facts. Your post is now trying to re-write the entire history of NV for the last 20 years but you still want to argue it.

In the most popular games today which are GPU demanding, where does the 980 stand, closer to the 290X (clearly a mid-range product) or closer to the Fury X, Titan X, 980Ti cards? It can't be more obvious.

If it takes 2 980s to just match a single 980Ti in The Witcher 3, there is no way a 980 is a high-end videocard of the Maxwell generation. That's not even logical.

However, Russian, while I agree with some of your arguments, you always skip over the biggest fact that at the time when GTX 980 came out, it was a compelling product in and of itself at it's price.

This in itself is highly debatable in the objective PC gamers' viewpoints. You are just capturing a very small time-frame of where your statement could even remotely be argued as true but once the dust settles at the end of this entire generation, the 980 will go down as a waste of $ because for the majority of this generation this is how it went down:

1) 970 SLI > 980 for just $60 more.
2) R9 295X2 > 980 for a similar price or often just $50 more.
3) 290X offering 85-90% of the performance of 980 at high resolution for nearly half the price.
4) 980Ti OC destroying 980 OC by 30-40% in games, while providing 50% more VRAM.

When 980Ti came out for $650 and trounces the 980 so badly but the 980 can barely beat a 290X and is trading blows with an R9 390X (which is just an after-market 290X more or less), it's hard to say anything positive about the 980. It's essentially just a 560Ti more or less priced at $550 and deep down you know it.

Getting back to the original discussion of Pascal, if NV calls the first chip GP100 and the 2nd larger chip GP200, we would still use the conventional basis of comparison which is performance. If GP100 is just 25% faster than 980Ti, then no, it's not the high-end Pascal card even if the code name is GP100. I don't believe NV would intentionally start playing games with code names though and start misleading to this extent but anything is possible I suppose.

Go back to any NV generation in the past and you will see that any flagship card more or less remained close to flagship in performance for the entire duration of that generation. GeForce 3 - check, GeForce 4600 - check, GeForce 5900U - check, 6800U - check, etc.

Claiming that 680 or 980 were high-end chips is akin to saying that 7900GT/7950GT was a high-end card because it sure wiped the floor with 6800U in games so according to some here suddenly if we go back in time, then NV should have priced 7900GT at $550? :biggrin:

As I said since this strategy has worked so well, and it clearly has since people now even get offended and defend that a x04 chip is suddenly flagship, NV has even more incentive to milk a generation into 2 halves and personally I would do the same in their shoes. Heck, maybe even split the generation into 3 parts:

1) Mid-range priced as flagship (GP204)
2) Cut-down flagship (aka 780/980Ti)
3) Full flagship with highest clocks (aka 780Ti/Titan Black / Titan X).

Historically speaking, no professional review site would have called a next generation mid-range card high-end/flagship even if it was 15-20% faster than the last generation's flagship but today, it's good PR/marketing $$ for the reviewer and I am sure if they called the chips out for what they are, NV would be royally pissed.

 

[Jaydip]

I think the performance of the card is what matters if we can call a card a true flagship or not but RS has a valid point here.I remember when 780Ti launched NV advertised it as a successor to 580 with 80% more performance so yeah I think internally NV thinks like that as well

 

[tviceman]

Well I do entirely agree that, given the 970's intro price, the 980 has a steep premium. Given 28nm's wacky and inflated pricing, I was happy to see a new fastest GPU come out with more vram and at a lower price than the previous fastest GPU. Being on the same node and getting 65% more performance with more vram and way better efficiency than the first 28nm high end card (HD7970) is progress.

But anyways, going forward I won't upgrade until I can get >85% more performance than a 980 at a similar TDP for $500.

 

[Headfoot]

I think the performance of the card is what matters if we can call a card a true flagship or not but RS has a valid point here.I remember when 780Ti launched NV advertised it as a successor to 580 with 80% more performance so yeah I think internally NV thinks like that as well

It's plain as day to anyone who is pro-consumer and not pro-brand. You're exactly right. They compared to their old, real flagships. $650 is probably reasonable given the die size for the true flagship 980 Ti but $550 is ridiculous for a chip that takes the same slot in the line up as the 460 1GB did (980). I find is strange how people accept the wool being pulled over their eyes, and in fact embrace it. Emotions about past purchases have no place in a reasoned discussion like this.

 

[Innokentij]

Ofcourse the codename matters but it is not set in stone and moves about for whatever nvidia wants to do themself. What i and u think about it they couldnt care less about, it is just a codename after all.

GF110 = 560ti/448core/570/580 <- High-end
2× GF110 = GeForce GTX 590 <- Enthusiast

GK104-400-A2 = GeForce GTX 680 <- High-end
GK104-355-A2 = GeForce GTX 690 <- Enthusiast

NVF0 (GK110) = 780/Titan <- High-end
NVF1 (GK110B) = 780TI <- Enthusiast

GM204 = 970/980 <- High-end
GM200 = 980TI/TitanX <- Enthusiast

Notice how the 500 series all was mashed into high end? on a full X10 codename? Nvidia does what it want. #Dealwithit

 

[Headfoot]

Notice how the 500 series all was mashed into high end? on a full X10 codename? Nvidia does what it want. #Dealwithit

You're the one that put them all in the high end category dude... that proves exactly nothing. Unless you've got some source to show nVidia considered 560 Ti high end I dont buy it.

 

[Innokentij]

You're the one that put them all in the high end category dude... that proves exactly nothing. Unless you've got some source to show nVidia considered 560 Ti high end I dont buy it.

Google is ur friend, as u can see all those 3 cards are on same codename.

https://en.wikipedia.org/wiki/GeForce_500_series

 

[Innokentij]

Is this a joke?

No he is quite right, the codename and what card performance will be is not always the same as history has show for nvidia. Just look at the 500 series, 3 cards on the same codename.

GF110 = 560ti/448core/570/580

 

[nvgpu]

http://www.kitguru.net/components/g...f-chips-using-16nm-finfet-process-technology/

Taiwan Semiconductor Manufacturing Co. announced on Thursday that it had started volume production of chips using its 16nm FinFET manufacturing technology in the second quarter of 2015. The world&#8217;s largest contract maker of semiconductors is delivering the first batch of products made at 16nm node to its customers right now, according to chief executive of TSMC. The high-volume ramp of 16nm FinFET tech begins this quarter, on-track with expectations.

 

[tviceman]

Is this ff+ or just plain ff?

Edit: article doesn't say.

 

[shady28]

http://www.kitguru.net/components/g...f-chips-using-16nm-finfet-process-technology/

I'm going to make a wild guess that they are pumping out the new Apple A9.

This is from April :

http://appleinsider.com/articles/15...tsmc-for-30-of-a9-chip-orders-for-next-iphone

Well-connected analyst Ming-Chi Kuo of KGI Securities issued a note to investors on Wednesday, a copy of which was obtained by AppleInsider, revealing that Apple has apparently made what he called a "last-minute decision to recruit TSMC." Apple is said to have called an audible after partner GlobalFoundries continued to experience poor yield rates on production of the next-generation CPU.

Specifically, GlobalFoundries' "A9" chip yield rate is said to currently be at about 30 percent yield rate, which is well below what Kuo said is a mass-production "basic requirement" of 50 percent.

 

[Sabrewings]

Is this ff+ or just plain ff?

Edit: article doesn't say.

They mention all three 16nm FF lines:

TSMC’s 16nm FinFET (CLN16FF), 16nm FinFET+ (CLN16FF+) and 16nm FinFET compact (CLN16FFC) fabrication...

 

[Samwell]

Is this ff+ or just plain ff?

Edit: article doesn't say.

It's FF+, they were always talking about it when they talk about 16nm Finfet process more recently. 16FF would have been production ready few month ago, but nobody wants to use the process because the performance characteristics aren't good enough.
Anyway TSMC is not the limiting factor for Pascal. The process will be ready in few month, but Nvidia needs to wait for HBM2.

 

[jpiniero]

Thought it a bit more and still stick by my prediction of:

-$1099 for the GP100 Titan (1/2 or 1/3 DP, 30-35% faster than Titan X)
-$899 for the cut GP100 (bad DP, 10-25% faster than Titan X)
- Maxwell 2 rebrands for the rest of the lineup, except for a new Maxwell 2 low end

The performance difference is going to be entirely based upon any efficiency gains and clock speeds and not core counts. There's a chance the core count is actually lower than the 3072 equivalent... depends on yields of course. The increase in the transistor count will be used for the new features and DP, and that increase is why the price hike is there.

 

[dark zero]

If they are going to full HBM (hardly) they will increae the prices and ditch the lowest tier since Intel is covering that tier now.

The thing is how much they increase