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

[YBS1]

Seems like both Nvidia and Intel just smack AMD around like it's a toy. Neither Intel/Nvidia is under any pressure to produce amazing products (by old standards), you can argue they are almost in a sense elongating current product life lines (blame it on nodes difficulties), and basically competing with themselves in their own markets.

I don't know if that's a completely fair assessment. The 290X was an insanely good GPU, it's failings had nothing to do with hardware and virtually everything to do with execution. Had it 1) not been much later to the market than the 780, 2) initially released with custom variants ready in the wings, and 3) Driver development been even a little farther along at release, I think it would have been a smashing success. I believe I would have opted for the 290X over the 780/780Ti had the above been true. By the time it all came together a good chunk of the target market had already invested in the competitor's solution. Timing is important, AMD has killed themselves on this point with every release after the 5870. Even with the 7970 which got to the market first (itself a great piece of hardware), by the time it was behaving the way it should nVidia had already countered.

 

[railven]

I don't know if that's a completely fair assessment. The 290X was an insanely good GPU, it's failings had nothing to do with hardware and virtually everything to do with execution. Had it 1) not been much later to the market than the 780, 2) initially released with custom variants ready in the wings, and 3) Driver development been even a little farther along at release, I think it would have been a smashing success. I believe I would have opted for the 290X over the 780/780Ti had the above been true. By the time it all came together a good chunk of the target market had already invested in the competitor's solution. Timing is important, AMD has killed themselves on this point with every release after the 5870. Even with the 7970 which got to the market first (itself a great piece of hardware), by the time it was behaving the way it should nVidia had already countered.

The 290X's failure was mostly due to bitmining and the price gouging vendors did. But I'm not even talking about market share here, or even total sales.

Tahiti got smacked around by Nvidia's middle chip. (Sure, now Tahiti is the top dog, but when people were actually buying these chips, seeing it having to get OC'ed (ghz editions) and use more power wasn't a very big seeling point).

AMD responded with Hawaii, and then NV just rolled out their big chip which was basically sitting in the sidelines racking in $1,000 Titan sales.

And before the dust settled in came Maxwell2's middle chip to just cement a victory.

Enter in Fiji and it's losing to the other half of Maxwell2 that was just sitting around in the background.

NV doesn't even have to bring out it's big guns to beat AMD. DX12 is going to change some of that, but by then Pascal might upset the DX12 momentum and I bet it will do it with the middle chip too.

EDIT: On the Intel side, they aren't even chasing performance gains anymore. It's all about power efficiency right now. All the talk about how Zen is gonna catch up to IVY or even Haswell, and yet I would be surprised if Intel has their own "Nvidia" waiting in the background.

 

[showb1z]

I think you need to look at some benchmarks, the only AMD chip that gets "smacked around" by its Nvidia counterpart right now is Fiji.
Most people use their GPU for several years, not just the month they get released in. Over 2 years of ownership AMD would've been the better choice in every match-up you described.
The performance of the hardware is not the issue.

 

[Mondozei]

3) Going back to point #1, since GTX680 and on, customers as a whole group have shown by voting with their wallets that they no longer care if the chip is mid-range or true flagship -- they will pay flagship prices for more performance regardless on where the chip lies in the NV architectural designation, even if it's just 15-25% more rather than the flagship historical 50-100% more.

I liked your post/comment overall but this section needs a clarification.

If you look at flagship to flagship, so 780 Ti to 980 Ti you absolutely see performance increases in the 50-60% range. Same for 780 to 980.

You have to compare to the same card in the last generation. So comparing the 980 should be done to the 780 and not to the 780 Ti.

So you're wrong on this, claiming that NV is offering less performance compared to previous flagships. That is only true if you make the incorrect comparison, which you obviously shouldn't do. NV also doesn't market the cards differently.
Everyone knows by now that the 780 was the GK104 and 780 Ti was GK110. So their naming scheme for the 900-series is identical and you as a consumer know as much.

Not sure why you try to pretend that isn't the case.

 

[Mondozei]

Nvidia successfully -and permanently- moved the price of high end up the ladder and AMD has followed suit. ~$500 is now the mid-die price from Nvidia and ~$400 is the new mid-die price from AMD.

No the mid tier is the 390/970. They are in the $330'ish range.

The GTX 980 is in many ways an odd bird, which is why many people who can afford it skip straight to 980 Ti instead and those who can't buy the 970. I got my EVGA 980 as a used card for 350, but I would never buy it new post-980 Ti.

 

[tviceman]

No the mid tier is the 390/970. They are in the $330'ish range.

The GTX 980 is in many ways an odd bird, which is why many people who can afford it skip straight to 980 Ti instead and those who can't buy the 970. I got my EVGA 980 as a used card for 350, but I would never buy it new post-980 Ti.

The naming conventions changed on Nvidia's side. The x80 used to be the absolute fastest single GPU card you could buy from them but now it's the second or third fastest of the last 2/3 product lineup releases, and those said two cards did not feature the flagship die. Even then the other x80 card featured a significantly cut down die and now even the current x80 TI card is cutfown.

I am not saying they're doing anything wrong, I am saying the x80 denomination no longer denotes the best of that family's lineup of products. It may be the best when it comes out, but it gets eclipsed by a x80 TI and a Titan now.

 

[n0x1ous]

Everyone knows by now that the 780 was the GK104 and 780 Ti was GK110

Epic fail. GTX 780 is cut down GK110. know what you are talking about before criticizing another's post.

 

[railven]

I think you need to look at some benchmarks, the only AMD chip that gets "smacked around" by its Nvidia counterpart right now is Fiji.
Most people use their GPU for several years, not just the month they get released in. Over 2 years of ownership AMD would've been the better choice in every match-up you described.
The performance of the hardware is not the issue.

Wow, it's like people don't read what other's post. Pretty sure I covered that.

Regardless of where those cards stand now, during their initial launch when most buyers looked at them, the numbers weren't as they are now.

Again, these claims that "people buy GPUs for 2-3 years" means nothing IF people bought their GPUs 2-3 years ago, before AMD got this huge gain in performance.

When Tahiti launched, GK104 for less outperformed it.

When Hawaii launched due to gouging, GK110 for less rivaled it, full GK110 still for less beat it, when the gouging subsided AMD's sales were already in the toilet. (You'll have some pound their chest that bitmining was the best thing that could happen to AMD since they sold cards. It created demand in Q1/Q2 of 2014, AMD responded by trying to ramp up production, bubble burst, used sales cannibalized new sales, AMD sat with a bunch of unused stock Q3/Q4, just look at their earning reports.)

And the trend repeats with Fiji losing to GM200. When these cards launch and make their initial (often lasting splash since reviewers often don't go back and rebench cards until something new is out) AMD is once again behind.

 

[ocre]

Epic fail. GTX 780 is cut down GK110. know what you are talking about before criticizing another's post.

. If you follow his point, it's clear he meant the Gtx 680 or the 770.

It was 3am when he posted, it's probably just a typo

 

[SteveGrabowski]

When Hawaii launched due to gouging, GK110 for less rivaled it, full GK110 still for less beat it, when the gouging subsided AMD's sales were already in the toilet. (You'll have some pound their chest that bitmining was the best thing that could happen to AMD since they sold cards. It created demand in Q1/Q2 of 2014, AMD responded by trying to ramp up production, bubble burst, used sales cannibalized new sales, AMD sat with a bunch of unused stock Q3/Q4, just look at their earning reports.)

I wonder how much mining cost them in RMAs? No GPU is designed to be run at 100% load 24/7. I imagine the mining was a huge reason why failure rates were so high in R9 200 series cards.

 

[tential]

Wow, it's like people don't read what other's post. Pretty sure I covered that.

Regardless of where those cards stand now, during their initial launch when most buyers looked at them, the numbers weren't as they are now.

Again, these claims that "people buy GPUs for 2-3 years" means nothing IF people bought their GPUs 2-3 years ago, before AMD got this huge gain in performance.

When Tahiti launched, GK104 for less outperformed it.

When Hawaii launched due to gouging, GK110 for less rivaled it, full GK110 still for less beat it, when the gouging subsided AMD's sales were already in the toilet. (You'll have some pound their chest that bitmining was the best thing that could happen to AMD since they sold cards. It created demand in Q1/Q2 of 2014, AMD responded by trying to ramp up production, bubble burst, used sales cannibalized new sales, AMD sat with a bunch of unused stock Q3/Q4, just look at their earning reports.)

And the trend repeats with Fiji losing to GM200. When these cards launch and make their initial (often lasting splash since reviewers often don't go back and rebench cards until something new is out) AMD is once again behind.

Really you just are bad at reading benchmarks. When I see fury x behind the 980ti I purchase it right away. Because 2 years later, it'll be faster than a 980ti and then I can say I made a better choice! Sure, for 2 years I had a slower chip, true, I can replace my card for a midrange card that's faster, buy who cares? I made the better longterm choice...
/sarcasm

I specifically didn't buy fury x when I saw the performance sucked and oc sucked. I don't buy chips and pray they get faster. Really, it pisses me off to no end that amd needs that long to get mature drivers out. It's actually a reason I don't want to buy their cards. Fix your drivers from the start. Don't make me wait 2 years for good drivers.
Oh, voltage control on fury x? Still Mia. Amd is a waiting game, I don't pay 600+ to wait. This is why I can't stand fiji. If ai is the same way, I don't even know what I'll do. Maybe I'll just sli nvidia, screw freesync and hope nvidia is fast enough to actually lock 60 fps, and get a "lightboost" TV from sony.

Ill know more when I see big pascal what I plan to do although I'll still go amd, I won't be happy with it.

That's my issue, I'm happy with my cpu purchase. Intel makes me happy. Amd makes me miserable, and so does nvidia with gsync and how they lock everything down.
It's picking the better of 2 bad choices....

 

[Timmah!]

I liked your post/comment overall but this section needs a clarification.

If you look at flagship to flagship, so 780 Ti to 980 Ti you absolutely see performance increases in the 50-60% range. Same for 780 to 980.

You have to compare to the same card in the last generation. So comparing the 980 should be done to the 780 and not to the 780 Ti.

So you're wrong on this, claiming that NV is offering less performance compared to previous flagships. That is only true if you make the incorrect comparison, which you obviously shouldn't do. NV also doesn't market the cards differently.
Everyone knows by now that the 780 was the GK104 and 780 Ti was GK110. So their naming scheme for the 900-series is identical and you as a consumer know as much.

Not sure why you try to pretend that isn't the case.

Nope, because 980 is sold for flagship price, therefore its directly comparable to previous flagship, which happens to be 780ti. If 980 was sold for 250-300 as once gtx *60 cards were (since its 204 as they were), then your point would stand.

 

[tential]

Nope, because 980 is sold for flagship price, therefore its directly comparable to previous flagship, which happens to be 780ti. If 980 was sold for 250-300 as once gtx *60 cards were (since its 204 as they were), then your point would stand.

"nope because it doesn't fit my point".

Clarified that for you.

 

[Azix]

Really you just are bad at reading benchmarks. When I see fury x behind the 980ti I purchase it right away. Because 2 years later, it'll be faster than a 980ti and then I can say I made a better choice! Sure, for 2 years I had a slower chip, true, I can replace my card for a midrange card that's faster, buy who cares? I made the better longterm choice...
/sarcasm

Actually, let's make a bet. Within the next 6 months to a year 980 ti owners will regret buying that card. Either nvidia will screw them over when pascal releases or Fury X will beat it in every released dx12 game. Betting Fury X beats it in released dx12 games. Actually, lets make it even better. The 390x will be putting out performance equal to or better than the 980ti in that time period on released dx12 games.

The wager is $100. I win you, give me $100. you win, you give me $100.

 

[jpiniero]

The thing about DX12 is that it's even more dependent on the game (and not the drivers) for optimizations. Of course... who is doing the optimizations? AMD/nVidia.

I've gone back and forth on how quickly nVidia will abandon Maxwell, mostly because I am expecting Maxwell 2 rebrands to fill in where Pascal isn't. And the rebrands could be quite high up in the chain for another year+.

 

[Sabrewings]

Actually, let's make a bet. Within the next 6 months to a year 980 ti owners will regret buying that card. Either nvidia will screw them over when pascal releases or Fury X will beat it in every released dx12 game. Betting Fury X beats it in released dx12 games. Actually, lets make it even better. The 390x will be putting out performance equal to or better than the 980ti in that time period on released dx12 games.

The wager is $100. I win you, give me $100. you win, you give me $100.

Does it have to be all 980 Ti owners? I won't regret it when Pascal releases because it gave me a year of great performance. Time to move on.

 

[happy medium]

The 390x will be putting out performance equal to or better than the 980ti in that time period on released dx12 games.

The wager is $100. I win you, give me $100. you win, you give me $100

6 months to a year? make it 500$, paypal, and i'll bet you

My bet is by the time direct x 12 matters, a gtx980ti will be old tech and not many will care.

 

[Azix]

6 months to a year? make it 500$, paypal, and i'll bet you

My bet is by the time direct x 12 matters, a gtx980ti will be old tech and not many will care.

first world problems.

 

[happy medium]

first world problems.

"put up or shut up" as my pop would say.:whiste:

 

[railven]

Does it have to be all 980 Ti owners? I won't regret it when Pascal releases because it gave me a year of great performance. Time to move on.

I'm on this boat. Only way I'd regret my 980 Ti purchase is if Corsair never releases the stupid bracket they promised, because I'm feeling pretty stupid right now for buying a Ref PCB Zotac when the EVGA SC+ are also ref PCB.

I've noticed most people dropping $500+ on a GPU aren't the kind that care when something is faster. Since they'll most likely be upgrading to it.

It's the penny pinchers that seem to care most what those $500+ people are doing.

 

[tential]

Does it have to be all 980 Ti owners? I won't regret it when Pascal releases because it gave me a year of great performance. Time to move on.

Pretty much the reason I regret buying a 980Ti day one. If I had bought it day 1, I would have been happy. But because I waited for Fury X. Then waited for Fury X Voltage unlock results (THe Voltage unlock isn't even out yet!), and was thoroughly disappointed, it's now so close to the nodeshrink I feel like it's a little late when I know I'll buy a nodeshrink high end card too. And I don't even need the performance, I'm not playing new games, all of the games I'm playing are old, but I can do above 4K DSR, and that'd be pretty cool.

But then I saw the Wasabi Mango monitor and I want that, but it will tear me apart to spend $650+ next gen to get a slower AMD card if that's what happens just like it pained me this generation.

 

[railven]

Pretty much the reason I regret buying a 980Ti day one. If I had bought it day 1, I would have been happy. But because I waited for Fury X. Then waited for Fury X Voltage unlock results (THe Voltage unlock isn't even out yet!), and was thoroughly disappointed, it's now so close to the nodeshrink I feel like it's a little late when I know I'll buy a nodeshrink high end card too. And I don't even need the performance, I'm not playing new games, all of the games I'm playing are old, but I can do above 4K DSR, and that'd be pretty cool.

But then I saw the Wasabi Mango monitor and I want that, but it will tear me apart to spend $650+ next gen to get a slower AMD card if that's what happens just like it pained me this generation.

Just be confident! If 390X > 980 Ti in DX12 games as is being wagered, then Fury X be like 980 Ti SLI!

Wait...where did I hear that claim before...

 

[tential]

Does it have to be all 980 Ti owners? I won't regret it when Pascal releases because it gave me a year of great performance. Time to move on.

Pretty much the reason I regret buying a 980Ti day one. If I had bought it day 1, I would have been happy. But because I waited for Fury X. Then waited for Fury X Voltage unlock results (THe Voltage unlock isn't even out yet!), and was thoroughly disappointed, it's now so close to the nodeshrink I feel like it's a little late when I know I'll buy a nodeshrink high end card too. And I don't even need the performance, I'm not playing new games, all of the games I'm playing are old, but I can do above 4K DSR, and that'd be pretty cool.

But then I saw the Wasabi Mango monitor and I want that, but it will tear me apart to spend $650+ next gen to get a slower AMD card if that's what happens just like it pained me this generation.

first world problems.

Exactly the point...
You're trying to penny pinch in a luxury market are you kidding me? I was EXCITED for Fury X, because it was a luxury product. Not just because it was a new GPU, but AMD was making a beautiful looking product that looked like time and care went into it. Too bad the performance wasn't there to back it, and they had to lie about that performance/OC too, which really irritated me enough into not getting it (Not that I could anyway it's sold out). Just like I won't buy a GTX 970, I won't buy a Fury X. The Pump issue further pissed me off, who launches a Luruxy high end product, and doesn't even test 1 of the cards? Clearly no QC, that was a joke.

I'm on this boat. Only way I'd regret my 980 Ti purchase is if Corsair never releases the stupid bracket they promised, because I'm feeling pretty stupid right now for buying a Ref PCB Zotac when the EVGA SC+ are also ref PCB.

I've noticed most people dropping $500+ on a GPU aren't the kind that care when something is faster. Since they'll most likely be upgrading to it.

It's the penny pinchers that seem to care most what those $500+ people are doing.

I think they're thinking that you'll hold onto the card for a long time or that we won't upgrade whenever we want to. I didn't buy even my 7950 to "hold onto for a long time." I just happened to do so since I wasn't gaming, and I STILL tried to upgrade to the R9 290. But it was bit coin miners that rampaged the price. Which now I'm noticing a common trend. When AMD launches a high end card these last 2 gens, it was unavailable at launch.... If someone hadn't posted that Wasabi Mango monitor, I'd have a MSI GTX 980Ti. But Freesync/Gsync is just necessary for 4K to me and only AMD provides it, so I'm stuck "Waiting". I pretty much associate at this point anything I want to do with AMD with "Waiting".

 

[Azix]

You guys think all people who buy these things have flexibility. There are those who saved up and can't make up the difference even if they sold the 980ti but that's just one. There are also those who would have difficulties importing/buying such cards or the prices are just crazy for them to begin with.

For those people, the AMD cards holding up is heaven. Those who throw cash around like an STD don't really have to care about anything. In fact you could well buy a fury x for the fun of it, throw that away then buy a 750ti to see how well it does at crysis 4k, then buy a 980ti to play worms Armageddon on your brand new CRT. meh.

Just be confident! If 390X > 980 Ti in DX12 games as is being wagered, then Fury X be like 980 Ti SLI!

Wait...where did I hear that claim before...

Fury X is a weird one. I don't know who expects SLI performance but if bottlenecks don't come into play it should be faster than overclocked 980tis.

I can't wait to see what nvidia decided to do with pascal. With AMD they can just fix the potential bottlenecks and pack more performance and call it a day. Maybe add 12.1 if anybody cares. But nvidia has some changes to make.

Now that dx12 is coming AMDs TFLOPs/mm2 "advantage" comes more into play. Take into consideration the fable legends benchmark. That was even less favorable to AMD than ashes, yet a 390 was faster than a stock 980 (slower than OC-ed 980) but you see the direction things are going there. That had even less usage of async. Compare the TFLOPs of the 2. 390 is 5.12 TFLOPS, 980 is 4.616 TFLOPS. The more compute is used the more this matters. Pascal has to become more competitive in terms of raw performance now that games are going to be more dependent on it.

Also their first HBM product, hardware or software scheduler, the double precision performance they promised...

 

[railven]

You guys think all people who buy these things have flexibility. There are those who saved up and can't make up the difference even if they sold the 980ti but that's just one. There are also those who would have difficulties importing/buying such cards or the prices are just crazy for them to begin with.

Odd, I never said that. I said "most" where as you implied all. Big difference between what you and I said.

I'm on this boat. Only way I'd regret my 980 Ti purchase is if Corsair never releases the stupid bracket they promised, because I'm feeling pretty stupid right now for buying a Ref PCB Zotac when the EVGA SC+ are also ref PCB.

I've noticed most people dropping $500+ on a GPU aren't the kind that care when something is faster. Since they'll most likely be upgrading to it.

It's the penny pinchers that seem to care most what those $500+ people are doing.

Actually, let's make a bet. Within the next 6 months to a year 980 ti owners will regret buying that card. Either nvidia will screw them over when pascal releases or Fury X will beat it in every released dx12 game. Betting Fury X beats it in released dx12 games. Actually, lets make it even better. The 390x will be putting out performance equal to or better than the 980ti in that time period on released dx12 games.

The wager is $100. I win you, give me $100. you win, you give me $100.

For those people, the AMD cards holding up is heaven. Those who throw cash around like an STD don't really have to care about anything. In fact you could well buy a fury x for the fun of it, throw that away then buy a 750ti to see how well it does at crysis 4k, then buy a 980ti to play worms Armageddon on your brand new CRT. meh.

It is heaven for them. AMD, on the other hand, they aren't partaking in the celestial celebrations.

As for the red, woof, salty? Pro-AMD guys sure do like to insinuate a lot.

Fury X is a weird one. I don't know who expects SLI performance but if bottlenecks don't come into play it should be faster than overclocked 980tis.

Twas a joke. But a long time ago some one here did predict that once Mantle hit with 290X it would beat SLI 780. DX12 being the spawn of Mantle, my joke pretty much followed the same logic. Haha.

I can't wait to see what nvidia decided to do with pascal. With AMD they can just fix the potential bottlenecks and pack more performance and call it a day. Maybe add 12.1 if anybody cares. But nvidia has some changes to make.

Now that dx12 is coming AMDs TFLOPs/mm2 "advantage" comes more into play. Take into consideration the fable legends benchmark. That was even less favorable to AMD than ashes, yet a 390 was faster than a stock 980 (slower than OC-ed 980) but you see the direction things are going there. That had even less usage of async. Compare the TFLOPs of the 2. 390 is 5.12 TFLOPS, 980 is 4.616 TFLOPS. The more compute is used the more this matters. Pascal has to become more competitive in terms of raw performance now that games are going to be more dependent on it.

Also their first HBM product, hardware or software scheduler, the double precision performance they promised...

Yes, AMD is finally seeing a little reward for their long investments. Question is, can they hold onto it? NV has this habit of either producing rabbits from a hat or fistfuls of money.